- Email: [email protected]
Hierarchical accountability in government
Hierarchical Accountability in Government Razvan Vlaicu, Alexander Whalley PII: DOI: Reference:
S0047-2727(15)00215-7 doi: 10.1016/j.jpubeco.2015.12.011 PUBEC 3639
To appear in:
Journal of Public Economics
Received date: Revised date: Accepted date:
30 May 2014 17 December 2015 22 December 2015
Please cite this article as: Vlaicu, Razvan, Whalley, Alexander, Hierarchical Accountability in Government, Journal of Public Economics (2016), doi: 10.1016/j.jpubeco.2015.12.011
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Razvan Vlaicuy
RI P
T
Hierarchical Accountability in Government Alexander Whalleyz
NU
SC
December 2015
Abstract
MA
This paper studies a setting where a relatively uninformed voter holds a policymaker accountable through an informed intermediary. In equilibrium the voter uses the intermediary to insulate the policymaker from pandering incentives when the voters policy
ED
expertise is low or the policymakers congruence is high. The voter can thus enjoy the bene ts of bureaucratic expertise without forfeiting electoral responsiveness. We examine the models predictions using U.S. city-level data, and nd that hierarchically-
PT
accountable managers reduce popular city employment, and adjust it more exibly, than electorally-accountable mayors. The estimated incentive e¤ects are smaller in
CE
cities with high voter expertise and larger during election years, and are robust to instrumentation by precipitation shocks that inuenced early 20th century manager government adoptions for reasons obsolete today.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
JEL Classi cation: D72, D73, H70. Keywords: pandering, accountability, incentives, city manager.
We would like to thank Jim Davies, Allan Drazen, Fernando Ferreira, Price Fishback, Lisa George, Gabriele Gratton, Alex Hirsch, Shawn Kantor, Ethan Kaplan, Ken Shotts, Gergely Ujhelyi, Richard Van Weelden, Emanuel Vespa, Joel Waldfogel, John Wallis, and seminar participants at Florida State University, IIES, New York University, Stanford University, Stockholm University, University of Chicago, University of Houston/Rice University, University of Illinois at Urbana, University of Pittsburgh, Canadian Public Economics Conference, Econometric Society NAWM, and Princeton Political Economy Conference. Erin Moody, Krystal Tapia, Michael Wall, and Marko Zivanovic provided excellent research assistance. y Northwestern University, Kellogg School of Management, 2001 Sheridan Road, MEDS 5th Floor, Evanston, IL 60208-2001, USA. Email: [email protected]. z University of California - Merced, School of Social Sciences, Humanities & Arts, 5200 North Lake Road, Merced, CA 95343, USA; and NBER, 1050 Massachusetts Ave., Cambridge, MA 02138, USA. E-mail: [email protected].
1
ACCEPTED MANUSCRIPT 1
Introduction
The accountability of government o¢cials to the general public is a fundamental principle of democratic governance. Yet, this principles most direct manifestation electoral
T
accountability while a useful safeguard may nevertheless introduce distortions in the poli-
RI P
cymaking process. Having to periodically face a public whose policy expertise is limited gives electorally-accountable o¢cials incentives to disregard their private expertise and adopt pop-
SC
ular policies not necessarily in the public interest by engaging in pandering (Canes-Wrone, Herron, and Shotts 2001) or electoral manipulation (Rogo¤ 1990). Delegating policymaking instead to unaccountable technocrats may allow expertise to be used without fear of
NU
electoral retribution but runs the risk that technocrats may pursue private goals that can disconnect policymaking from the public interest (Maskin and Tirole 2004). An important
MA
question then is how to resolve this tension between expertise and responsiveness, that is, how to allow public o¢cials expertise to be expressed in policymaking while also preserving accountability to the general public.
ED
In this paper we focus on a setting where the policymaker is accountable to the voter through an intermediary. Policymakers such as prime ministers, city managers, and school
PT
district superintendents cannot be directly removed by voters. However, they remain accountable to the voter in the sense that they can be replaced at will by a popularly elected
CE
intermediary, e.g., legislature, city council, school board. Following contract theory terminology we refer to this type of principal-intermediary-agent relationship as hierarchical accountability.1
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
At rst sight, holding a policymaker accountable through an intermediary seems to disconnect policymaker choices from voter preferences. If the voter is fully informed he can, nevertheless, exploit the intermediarys reelection motivation to control policymaker moral hazard just as e¤ectively as the voter could directly (Persson, Roland, and Tabellini 1997).2 In an incomplete information environment, however, the voter also faces an adverse selection problem which weakens electoral accountability by producing pandering incentives. Even if the voters best interest is to limit policymaker pandering, a commitment problem renders him unable to do so. Ex ante an uninformed voter may prefer to insulate a better-informed policymaker. Ex post, however, he would rather replace an unpopular policymaker because 1
About half of U.S. cities are run by city managers. "If the manager is not responsive to the governing body, it has the authority to terminate the manager at any time." (ICMA 2007, p. 2). Other forms of indirect accountability grant the policymaker a xed term, i.e., top regulators, central bank governors. 2 One caveat is possible policymaker-intermediary (executive-legislature) collusion. In that case, the executive would have more discretion to engage in rent-seeking.
2
ACCEPTED MANUSCRIPT unpopular policies signal the policymaker may have di¤erent preferences.3 When accountability to an imperfectly-informed voter distorts the policymakers incentives in this way, delegating policymaker accountability to an informed intermediary seems
T
justi able. The case for delegation is less clear-cut, however, if the intermediarys prefer-
RI P
ences may also di¤er from the voters. In that case the voter needs to provide incentives for the intermediary to act in the voters interest. Under what conditions can an informed intermediary help the voter insulate the policymaker from pandering incentives?
SC
To understand behavior in this setting we rst develop a hierarchical agency model (voterintermediary-policymaker model) where the voter is uncertain about the optimal policy but
NU
leans toward a "popular" policy. The intermediary is a political expert, i.e., informed about the policymakers type, and the policymaker is a policy expert, i.e., informed about the optimal policy. The models key insight is that the voter can credibly use the intermediary
MA
to insulate the policymaker from popular pressure when pandering is relatively detrimental to the voter. That happens when the voters policy expertise is low or the policymakers
ED
congruence, i.e., alignment with voter preferences, is high. In these cases the voter is ex ante better o¤ retaining an unpopular policymaker and the intermediary allows him to commit to do so ex post. This is because an unpopular policymaker is still more likely congruent,
PT
so retaining an unpopular policymaker signals intermediary congruence. Hierarchical accountability thus o¤ers the voter the exibility to promote policymaking expertise without
CE
forfeiting electoral responsiveness. Voters can credibly and optimally switch between erring on the side of expertise and erring on the side of responsiveness when their informational and political environment changes.4
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
In our empirical application we study policymaking by U.S. city managers, hierarchicallyaccountable o¢cials with the same major policy responsibilities as electorally-accountable mayors, i.e., writing the budget and hiring personnel. While manager government provides a natural measure of hierarchical accountability, how to measure pandering behavior is less clear. We propose that pandering behavior can be detected empirically in policy issues that satisfy three conditions: (i) is a primary (not secondary) policy issue over which the policymaker has jurisdictional control, (ii) the optimal policy is state-contingent, and (iii) the public has a clear stance on this issue. We argue that police o¢cer employment satis es 3
Besley and Smart (2007) and Smart and Sturm (2013) notice the commitment issue in the context of scal rules and term limits, respectively. 4 In contract theory and corporate nance, hierarchical agency models study optimal incentives through wage contracts (e.g., Strausz 1997, Park 2000). Our main theory result echoes the nding in this literature that under asymmetric information an intermediary allows the principal to commit to a broader range of incentive structures for the agent.
3
ACCEPTED MANUSCRIPT these requirements as (i) crime has consistently ranked among the top two local policy issues in Gallup surveys of local attitudes since 1959 (see Gallup 2000) and crime prevention among the top two city services with high "resident sensitivity to quality" (Levin and Tadelis 2010);
T
city executives have juristictional control over public safety, a substantial local budget item,
RI P
(ii) the probability of crime uctuates with economic and social conditions, changing the optimal policy response, and (iii) due to constant voter concern over public safety, demand for police o¢cers by relatively uninformed voters can be expected to persist even when the
SC
probability of crime is low; in contrast, civilian police employment, e.g., administrators, dispatchers, as well as other city employee categories, should not elicit such a clear popular
NU
preference. Across a number of speci cations we nd that on average managers employ 814% fewer police o¢cers per capita than mayors but a comparable number of police civilians and non-police employees per capita.
MA
A central challenge to estimating institutional e¤ects is that institutions may be endogenous to policymaking, for instance through unobserved voter preferences (Aghion, Alesina,
ED
and Trebbi 2004). To address potential endogeneity in accountability form we propose an instrument for manager government. The instrument is based on the observation that before the 1936 Flood Control Act transferred ood prevention from local governments to the fed-
PT
eral Army Corps of Engineers cities often responded to ood-related infrastructure crises by adopting manager government because it facilitated the ascension of engineers into the top
CE
executive o¢ce. We document that pre-1936 precipitation shocks inuenced early switches to manager government for reasons obsolete today and nd that the city employment patterns noted above also appear in this IV setting.5
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
We further explore the theory models incentive mechanisms and nd that policy volatility is higher in manager governments because managers can more often, i.e., when insulated, act on their expertise to adjust the policy to the stochastic state. We also propose a measure of voter expertise and nd that the manager-mayor o¢cer employment di¤erential decreases in cities a¤ected by the crack epidemic, where voters should be more aware of the state of crime. It is also more pronounced in election years, when the theory model predicts incentives should be sharper. We note that these patterns cannot be fully explained by alternative mechanisms, such as pure patronage motivations or policymaker type selection. Our paper relates to the literature on the career concerns of expert policymakers. Maskin and Tirole (2004) show how relying on unaccountable technocrats, e.g., unelected bureaucrats, tenured judges, eliminates pandering incentives but makes removing noncongruent 5 To our knowledge this is the rst instrument for manager government in the literature, if we exclude Baqirs (2002) use of lags of city institutions as instruments for current city institutions.
4
ACCEPTED MANUSCRIPT policymakers harder; thus there is a tradeo¤ between expertise and responsiveness in policymaking. Similarly, media providers improve the monitoring of public o¢cials, but face incentives to be yes men themselves (Ashworth and Shotts 2010); allowing decisionmaking
T
behind closed doors (Fox 2007), or imposing term limits (Smart and Sturm 2013), both
RI P
reduce the electoral pressure on congruent policymakers to be popular, but also allow noncongruent ones to make suboptimal policies; candidate competition reduces pandering, but encourages anti-pandering (Kartik, Squintani, and Tinn 2013). Our theory shows, by con-
SC
trast, that hierarchical accountability a¤ords the voter the exibility to optimally choose between insulating congruent policymakers, on the one hand, and providing pandering in-
NU
centives to noncongruent policymakers, on the other. The voter can thus enjoy the bene ts of bureaucratic expertise without forfeiting electoral responsiveness. Our contribution to the empirical literature is two-fold. First, we provide an empirical
MA
measure of pandering behavior grounded in the logic of the theoretical pandering literature and use it to quantify how pandering incentives respond to informational and electoral
ED
factors. In the previous literature pandering has been measured by looking at how U.S. presidents budgetary proposals respond to public opinion surveys of citizen spending preferences (Canes-Wrone and Shotts 2004), how judges with reelection concerns or facing stronger com-
PT
petition issue more punitive sentences (Huber and Gordon 2004, Gordon and Huber 2007), or how elected judges sentencing behavior follows the ideological preferences of their con-
CE
stitutents (Lim 2013). Second, we contribute to the literature on U.S. city managers which has largely focused on di¤erences in public spending from city mayors. Coate and Knight (2011) survey this literature, note that results have been mixed, and provide new evidence
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
that managers outspend mayors.6 A smaller literature has looked at other policy outcomes: Managers are more likely to privatize city services (Levin and Tadelis 2010), more exibly adjust tax revenues (Vlaicu and Whalley 2011), and reduce full-time city employment (Enikolopov 2014). In contrast, our paper documents di¤erences in policymaking incentives for popular and neutral policies, and the mechanisms behind them. Also, while the previous literature has treated city government form as exogenous, we o¤er an instrumental variable strategy that addresses the di¢cult issue of government form endogeneity. The rest of the paper is organized as follows. Section 2 presents the model. Section 3 provides historical background and introduces the data. Section 4 presents the empirical strategy and results. Section 5 concludes. 6
Coate and Knights (2011) citizen-candidate model attributes di¤erences in spending not to incentives, but to voters electing di¤erent councilmember types. Other papers (Baqir 2002, MacDonald 2008) found no spending di¤erential between managers and mayors.
5
ACCEPTED MANUSCRIPT 2
Incentives Model
To gain insight into how hierarchical accountability operates we extend a standard principalagent model of electoral accountability by introducing an asymmetrically informed intermedi-
T
ary into the voter-policymaker relationship. The goal is to understand how being accountable
RI P
through an informed intermediary a¤ects the policymakers incentives. In what follows, hierarchical accountability is de ned as a setting where the policymaker
SC
can be removed at any time by an intermediary that in turn can be replaced by the voter periodically in an election. Electoral accountability, by contrast, refers to a setting where
NU
the policymaker can be directly replaced by the voter periodically in an election. Setup. At time t a policy choice xt needs to be made from the set of policy alternatives
MA
f0; 1g: One can think of xt = 1 as a high level of the policy, and xt = 0 as a low level. The policys optimality is contingent on the state of the world st 2 f0; 1g prevailing in that period, namely the optimal policy is the one that matches the state. The state is i.i.d. across
ED
periods. Let P fst = 1g denote the probability of the high state. There are three kinds of players: voter, intermediary, and policymaker. The voter gets
PT
a unit of payo¤ when the policy matches the state, zero otherwise: v(xt ; st ) = I fxt = st g. Here I is an indicator function. Notice that, as long as voter information is limited to the state prior ; the voter strictly prefers the high policy xt = 1 if and only if the high state is
CE
more likely, > 21 . In this case we say that the high policy is the popular policy, borrowing Maskin and Tiroles (2004) terminology. In this informational context, voter expertise can be measured by how far is from a half.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
The voter delegates policymaking to a policymaker (P ) and delegates policymaker accountability to an intermediary (I). These two agents are both policy-motivated and o¢cemotivated. Their policy motivation depends on their type, biased or congruent: Pt ; It 2 fB; Cg. For each type, the preferred policy gives him a payo¤ of one, and he receives zero otherwise. A congruent type is fully aligned with the voter, wanting policy xt = 1 in state st = 1 and policy xt = 0 in state xt = 0: A biased type prefers policy xt = 0 with probability in both states. One may think of the biased type wanting to match an "alternative" state that is uncorrelated with the true state st : This payo¤ structure is similar to the one adopted by Fox (2007), who assumes that the biased type always prefers policy xt = 0 in both states. An agent obtains his policy payo¤s as long as the agent is in o¢ce; out of o¢ce an agents payo¤ is normalized to zero.7 7
A previous version of the paper assumed the policymakers type could be either congruent, always
6
ACCEPTED MANUSCRIPT This setting captures the intuition that an unpopular policy signals an unappealing policymaker type. For instance, in a world where the voter is concerned about crime, hiring too few police o¢cers signals that the policymaker is soft on crime, as opposed to tough on
T
crime; or, in a world where the voter is concerned about ination, a low interest rate signals
RI P
that the policymaker is an ination dove, as opposed to an ination hawk. To see this, note that a popular voter preference toward the high policy, i.e., > 21 ; coincides with the biased policymaker leaning toward the low policy, and vice-versa. Since the biased type prefers
SC
the low policy more than half the time, and the congruent type less than half the time, an unpopular policy signals policymaker bias toward the unpopular low policy.
NU
The timing of the in nite-horizon game is as follows. (I) Every period, the policymaker observes the state st 2 f0; 1g and chooses between the unpopular/popular policy: xt 2 f0; 1g: (II) Every period, the intermediary observes the policy and policymaker type (xt ; Pt ) and
MA
decides whether to replace/retain the policymaker: yt 2 f0; 1g: (III) Every other period the voter, having observed two periods of choices [(xt 1 ; yt 1 ); (xt ; yt )] ; but not types Pt ; It
ED
or the current state st ; decides to replace/reelect the intermediary: zt 2 f0; 1g: An agent
exiting at t is succeeded by an agent (challenger) whose type is a new draw from the type distribution. An exiting agent cannot run for o¢ce again.8
PT
For expositional simplicity we assume that reelection incentives a¤ect only election-period behavior (cf. Rogo¤ 1990, Shi and Svensson 2006); in the Theory Robustness section below
CE
we discuss an extension that relaxes this assumption. Let ( jt )t=1;2;::: be two independent sequences of i.i.d. preference shocks, for j = P; I; where jt 2 fB; Cg and P jt = C = j
< P < 1:9 An agents type is either last periods or the current periods preference shock: jt 2 jt 1 ; jt and P jt = jt 1 = :
is the congruent types frequency. Assume
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
1 2
Assume 0 < < 1: These assumptions imply: First, that an agents type is correlated over time between two adjacent periods, but uncorrelated between two non-adjacent periods. In particular, an agents type at t + 1 (non-election period) is correlated with his type at t (election period), but uncorrelated with his type at t 1 (non-election period).10 Second, that although policymaker and intermediary types are initially independent, correlation between preferring the state-matching policy, or dissonant, always preferring the policy that mismatches the state (see also Maskin and Tirole 2004). The analysis is similar and available upon request. 8 A key feature of hierarchical accountability is that the policymaker can be replaced between elections. A city manager, for example, serves at the city councils pleasure. His "job tenure is only secure until the next council meeting" (Stillman 1977, p. 664). The models timing captures this feature by assuming that the intermediary can replace the policymaker in the middle of an electoral term. 9 Using a structural approach, Aruoba, Drazen, and Vlaicu (2015) estimate that over half of incumbent U.S. state governors are congruent with the publics preferences. 10 The type persistence probability is Pfjt = jt 1 g = Pfjt = jt 1 ; jt 1 = jt 1 g = (1 ) ; for any t:
7
ACCEPTED MANUSCRIPT their types can nevertheless develop during the game through the intermediarys equilibrium retention decisions. The policymaker cares about policies he himself chooses. The intermediary cares about Formally, lifetime utility for a
T
the choices of the policymaker he is keeping in o¢ce.
RI P
policymaker at the beginning of an electoral term, i.e., at t 1; is u(xt 1 ; st 1 jPt 1 ) + Q P1 i t+1 u(xi ; si jPi ) i =t y 1 ; for an intermediary it is [u(xt 1 ; st 1 jIt 1 )+ u(xt ; st jIt )]+ i=t Qi P1 2i 2t+2 I I ; s j ; s j [u(x ) + u(x )] 2i t+1 2i t+2 2i t+1 2i t+2 2i t+1 2i t+2 =t z2 t : Here is a i=t
SC
time discount factor, with 0 < < 1:
As is standard, our policymaker is a policy expert who knows whether the popular policy
NU
is optimal (the state st ). The intermediary, on the other hand, is a political expert, in the sense of Gailmard and Jenkins (2009), who knows the policymakers preferences (his type Pt ). This second feature seems reasonable in settings where the intermediary has frequent contact
MA
with the policymaker. That is probably the case for a city council since the policymaker they choose, the city manager, has to attend all city council meetings and sometimes committee
ED
meetings. The voter, by contrast, knows neither the popular policys optimality (moral hazard) nor the policymakers preferences (adverse selection). Thus, this model features two forms of the classic delegation tradeo¤: by delegating policymaking to a policy expert the
PT
voter may bene t from more informed policy choices but may su¤er from the policymakers bias; by delegating policymaker accountability to a political expert the voter may bene t
CE
from more informed retention decisions but may su¤er from the intermediarys bias. We denote voter posterior beliefs about agent types by jt (Xt ; Yt ) P jt+1 = CjXt ; Yt ;
where j = P; I and (Xt ; Yt ) = (x ; y )t =1 is the public history at time t; i.e., policy choices
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
and retention decisions up to time t. We refer to voter posterior beliefs about an agents type as that agents "reputation." Equilibrium. The game described above is a principal-intermediary-agent model with incomplete information. We solve for its Perfect Bayesian Equilibrium, namely: (i) policymaker, intermediary, and voter strategies (xt ; yt ; zt ) that are sequentially rational; and (ii) voter beliefs Pt ; It that are consistent with agent strategies. We add the requirement
that voter strategies be robust to a small fraction of non-strategic agents, implying that in a pooling equilibrium o¤-equilibrium path beliefs reect sincere, i.e., preference-based, behavior; see Maskin and Tirole (2004) for details on this re nement. We restrict attention to pure strategies. Throughout this section we assume that the discount factor is large
enough that both agent types would rather override their current preferred policy for the sake of remaining in o¢ce; the equilibrium will thus characterize agents incentives, i.e., 8
ACCEPTED MANUSCRIPT extrinsic motivation. In the Discussion section below we also consider the case where agents have no career concerns and instead follow their preferences, i.e., intrinsic motivations; that equilibrium captures behavior driven by selection.
T
Since the voter can only replace the policymaker through the intermediary, his problem that following a two-period electoral term [t
RI P
is to reelect that intermediary which seems more in tune with his preferences. We show 1; t], after the voter has observed choices
SC
[(xt 1 ; yt 1 ); (xt ; yt )], the incumbent intermediarys equilibrium public reputation It (Xt ; Yt ) increases whenever ^ It (xt ; yt ) P It+1 = Cj (xt ; yt ) ; It+1 = It = It I : In words, the
voter predicts future congruence based on election-period behavior (xt ; yt ) conditional on
NU
the intermediarys current type persisting into the next period. Since the voters reelection decision at t depends only on election-period variables (xt ; yt ), in non-election periods the intermediary is free to retain only policymakers of his own type; this makes the intermediarys
MA
type in future non-election-periods t + 1; t + 3; and so on, matter to the voter. The following proposition states the implications of this voter calculation for policymaker
ED
incentives. For contrast we also include the case of no intermediary, i.e., electoral accountability, which is the mainstay of the current literature. Proposition 1 For popular policy issues > 12 : (i) Hierarchical accountability either
PT
insulates the policymaker pre-election, if P > ; or creates policymaker pandering incen-
tives pre-election, otherwise; (ii) Electoral accountability always creates policymaker pan-
CE
dering incentives pre-election; (iii) Agents follow their preferences post-election under both accountability forms, e.g., an intermediary retains only policymakers of his own type. Proof. See Online Appendix A.1.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Under electoral accountability the voter uses the policy signal contained in the policy choice xt to learn about the policymakers type. Since > 21 ; the high policy xt = 1 is the popular policy and the biased type leans toward xt = 0. Thus a popular policy is a signal of policymaker congruence, and an unpopular policy is a signal of policymaker bias. The voters best response then is to reelect only after the popular policy. This creates a pandering incentive for both policymaker types, i.e., needing to choose the popular policy regardless of its optimality. Hierarchical accountability, on the other hand, provides the voter an additional signal: the retention signal contained in the intermediarys retention decision yt : Because the intermediary prefers to retain only policymakers of his own type, retention is a signal of intermediary congruence if the policymaker is more likely congruent. Thus, because for a popular policymaker that is always the case, as his reputation Pt (Xt ; Yt ) improves from the prior P > 21 ;
9
ACCEPTED MANUSCRIPT retaining a popular policymaker signals intermediary congruence, and replacing a popular policymaker signals intermediary bias. For an unpopular policymaker the voters reaction is more subtle because an unpopular
T
policymaker, while less appealing than the challenger, may or may not be more likely con-
RI P
gruent, and that is what the intermediarys reputation hinges on. Whether an unpopular policymaker is believed more likely congruent turns out to depend on the relative magni tudes of P and , namely whether P (1 ) > 1 P . This says that the frequency of
SC
congruent policymakers wanting the unpopular policy exceeds the frequency of biased policymakers wanting the unpopular policy. This happens if the state is su¢cienly likely to be
NU
low, or the policymakers prior reputation is su¢ciently good. In such cases, retaining an unpopular policymaker, while never a sequentially rational action for the voter, since the voter expects him to do worse than a challenger, nevertheless signals intermediary congruence;
MA
conversely, replacing an unpopular policymaker signals intermediary bias. Consequently, the voter will reelect an intermediary that retains an unpopular policymaker.
ED
The key implication for incentives is that, when the retention signal dominates the policy signal, i.e., P larger than , the voter gives the intermediary incentives to retain the policymaker regardless of the policymakers policy choices, thus endogenously insulating the
PT
policymaker. Conversely, if the policy signal dominates the retention signal, then retaining an unpopular policymaker signals a biased intermediary. In this case the voter gives
CE
the intermediary incentives to retain only popular policymakers, which in turn gives the policymaker the incentive to pander. Policymaker insulation yields ex ante election-period voter welfare equal to P 1 + 1 P [ (1 ) + (1 ) ] whereas under pandering voter welfare is P + 1 P :
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
The pandering equilibrium thus worsens the high types incentives, but may improve the low types incentives. The bene t of insulation is then P (1 ), while the potential cost is 1 P [2 (1 ) ] ; which is no larger than 1 P . It follows that when the voter plays an insulation equilibrium, that is, when P (1 ) > 1 P , or P > ;
voter welfare always improves. In that sense, the hierarchical structure o¤ers the voter the exibility to switch between insulation and pandering to optimally induce, through the intermediary, the types of policymaker incentives that best serve the voters interest.11 One may argue that another way in which hierarchically-accountable policymakers di¤er 11
The two types of intermediary strategies, leading to either insulation or pandering, are consistent with case studies of U.S. city councils that distinguish between council strategies that insulate the manager, termed "blind faith," versus council strategies that transmit popular preferences, termed "political." See Stillman (1977).
10
ACCEPTED MANUSCRIPT from electorally-accountable policymakers is that they have weaker motivations to deliver political patronage to various constituencies. While patronage could potentially a¤ect popular policies, such as police protection or low taxation, it also typically inuences neutral
T
policies such as in-kind transfers or city jobs that may not elicit a clear popular preference. indi¤erent between the policies xt = 1 and xt = 0. Proposition 2 For neutral policy issues =
1 2
RI P
Neutral policy issues can be modeled by setting equal to a half, making the voter ex ante policymaking outcomes do not vary
SC
with the accountability form in election periods or non-election periods. Proof. See Online Appendix A.1.
NU
Intuitively, a balanced state prior silences the policy signal. Both policymaker types, congruent and biased, would choose either policy half the time. The voter can now credibly insulate an electorally-accountable policymaker from pandering incentives since the voter
MA
cannot link either policy to a particular type. He also nds it welfare-enhancing to do so, as P (1 ) > 1 P [2 (1 ) ] ; making voter welfare under insulation dominate
ED
voter welfare under pandering. For hierarchically-accountable policymakers the equilibrium is also insulating because the retention signal is always stronger than the policy signal, since
P > = 12 . The implication is that pandering incentives are absent on neutral issues for
PT
both accountability forms.
The key comparative statics implications for policymaker incentives, based on the analysis
CE
above, can be stated as follows.
Proposition 3 Hierarchical accountability weakens the policymakers pre-election incentives to pander to the voter, compared to electoral accountability, and results in more equilib-
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
rium matching of the policy to the stochastic state, when: the relevant policy issue elicits a popular preference, and the voters policy expertise on this issue is low or the policymakers congruence P is high. This result implies incentive-driven di¤erences in both the mean and the variance of policy choices between the two accountability forms. Hierarchically-accountable policymakers on average reduce the prevalence of popular policies, and adjust policy more exibly in response to changes in the state, relative to electorally-accountable policymakers. Intuitively the presence of the intermediary allows the voter to credibly insulate the policymaker when pandering is relatively detrimental to the voter. Policymaker insulation is possible when voter expertise is low or policymaker congruence P is high. In these cases the voter is ex ante better o¤ retaining both popular and unpopular policymakers, although ex post the voter would replace an unpopular policymaker. The intermediary provides a commitment de-
11
ACCEPTED MANUSCRIPT vice for the voter to implement his ex ante preferences. This is because when P > >
1 2
an
unpopular policymaker is still more likely congruent, so retaining an unpopular policymaker signals intermediary congruence. Thus, it is sequentially rational for the voter to reelect
T
an intermediary that retained an unpopular policymaker, even though it is not sequentially
RI P
rational for the voter to retain an unpopular policymaker himself. This insulating equilibrium results in less popular and more volatile policy choices than the pandering equilibrium
SC
prevalent under electoral accountability.
Theory Robustness. The theoretical results above are robust to a setup where the voter factors into his reelection decision both election-period and non-election-period policy (cf.
NU
Martinez 2009); see Online Appendix A.2 for a solution to the model under this relaxed assumption. We nd that election-period pandering is driven by the following policymaker
MA
intertemporal tradeo¤: choosing a popular policy against preferences at the beginning of the term results in a sure loss, whereas postponing it for the end of the term results in a potential loss, i.e., because the election periods state may be such that the policymaker
ED
actually prefers the popular policy. The same logic applies to the intermediary. In Online Appendix A.3 we also develop a version of the model that allows for variation
PT
in pandering intensity. The main insight of that model is that the pandering probability is equal to the discounted reelection probability premium obtained for choosing the popular
CE
policy. This premium shrinks monotonically as the policy signal becomes weaker. In the limit, as the state prior approaches 12 ; the policy signal gradually fades out, and the policymakers incentive to pander becomes arbitrarily close to zero. In this sense, the strict
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
separation we made above between popular vs. neutral issues can be seen as a more contin uous characteristic of policy issues as varies in the interval 21 ; 1 . Discussion. Our theoretical results show how the incentives of hierarchically-accountable policymakers may depend on their policy, informational, and political environment; observationally, they predict a negative election-period policy di¤erential in popular policies between hierarchical and electoral accountability, when di¤erential, when
1 2
1 2
< < P ; but no election-period policy
< P < ; or on neutral policies, i.e., close to a half. The propositions
also imply that in the rst case policy outcomes are more volatile under hierarchical accountability because this institutional setting gives the policymaker more freedom to adjust the policy to the stochastic state. The model can also be solved under the assumption that the accountability forms primary role is not to provide incentives but to select certain types of policymakers; see Online 12
ACCEPTED MANUSCRIPT Appendix A.4 for the details of this variant of the model. The key insight of the selection model is that policymaker selection through an informed intermediary is more precise, since the intermediary is better informed, and more frequent, since it can also occur between
T
elections, but may back re if intermediary selection by the voter is weak.
RI P
This tradeo¤ notwithstanding, the timing of policy choices does yield an unambiguous prediction. (a) Under the incentives model, a popular policy di¤erential should arise in election years because policymakers change their behavior pre-election to increase their chances
SC
to stay in o¢ce; see Proposition 1. (b) Under the pure selection model, a popular policy di¤erential could be observed in both non-election and election years; electoral selection
NU
produces its e¤ects post-election only, while hierarchical selection produces its e¤ects both post-election and pre-election. Since selection increases the prevalence of congruent policymakers, and the popular policy is more likely optimal, better selection should be associated
MA
with more popular policies. Thus, the election-year popular policy di¤erential is negative under an incentives model, but positive under a selection model. The non-election-year ; i.e., depends on whether intermediary con-
ED
popular policy di¤erential has the sign of I
gruence dominates voter expertise. The di¤erent predictions of the incentives and selection
PT
models allows us to distinguish between the two models empirically.
Empirical Application
CE
3
Our theoretical results for policymaker incentives, summarized in Proposition 3 above, imply di¤erential policy patterns according to the accountability form. Our empirical application
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
examines popular (police o¢cer employment) and neutral (police civilian employment, nonpolice employment) policy issues in U.S. cities where the executive is either hierarchicallyaccountable (manager) or electorally-accountable (mayor). Assuming >
1 2
seems reason-
able for policy issues such as police o¢cer employment. This makes the high policy xt = 1 of high o¢cer employment the popular policy option. Stucky (2005) reviews the literature on how the publics crime probability prior, also known as "fear of crime" in the criminology literature, a¤ects popular preference for police.12 12
Gallups annual Crime Survey asks the question "Is there more crime in your area than there was a year ago, or less?" Since the survey started in 1973 the percentage of respondents who say "More" exceeded those that say "Less" except for three years 1998, 2000, 2001 (see Gallup 2010). In this context, assuming the state is i.i.d. across periods means that past policy choices do not a¤ect the voters current crime probability prior. This seems to be a good approximation if police have only a short-term e¤ect on committed crime (actual crime), while the probability of crime (latent crime) is driven by more fundamental forces like economic inequality, economic growth, and race relations.
13
ACCEPTED MANUSCRIPT City managers operate under hierarchical accountability because they can be removed at any time by the city council, whose composition can in turn be changed by voters in local elections. City mayors operate under electoral accountability because they can be replaced
T
by voters periodically in local elections. The testable hypotheses can be stated as follows.
RI P
(H1) Managers are less likely than mayors to adopt a popular policy. Neutral policy issues do not exhibit this mean policy di¤erential.
(H2) Managers are more likely than mayors to make changes to popular policies. Neutral
SC
policy issues do not exhibit this policy volatility di¤erential.
(H3) The negative mean di¤erential on popular issues should decrease in voter expertise.
NU
(H4) The negative mean di¤erential on popular issues should be more pronounced in election years.
MA
In what follows we examine the evidence for these predictions. Historical Background. Early U.S. city executives were appointed by state governors or city councils. After major cities like Boston and St. Louis started to popularly elect their
ED
mayors in 1822, electoral accountability for city executives became nearly universal. By the end of the 19th century city politics had became dominated by "machine" politicians
PT
drawing their support from workers and immigrants, and often using illegal means to stay in o¢ce. This crisis in city government accountability sparked an urban reform movement that
CE
coincided with broader social reforms taking place during the Progressive Era (roughly 18901930). Initially reformers targeted cities electoral systems: from district-based partisan to at-large non-partisan elections. The goal was to dilute the power of minorities and parties
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
on which the machines thrived. The Progressives, who wanted not only cleaner government but also more e¢cient government, were later inspired by the "e¢ciency movement"s success with the new corporate form in business enterprises, i.e., CEO accountable to shareholders through a board of directors, and sought to apply this model to city governance. Manager government, rst experimented with in the small city of Staunton, Virginia in 1908, attained broad recognition when the National Municipal League picked it as their recommended government form in the 1915 edition of the Model City Charter. During the 1910s and 1920s most major cities, including New York City, were debating switching to a manager govenment. Despite strong opposition from incumbent mayors and party bosses manager government advanced steadily.13 Knoke (1982) attributes successful switches to manager to strong 13 A number of 87 cities adopted manager government between 1913-1918, another 153 between 1918-1923, and 84 more between 1923-1928 (Judd and Swanstrom 2010).
14
ACCEPTED MANUSCRIPT business interests, weak unions, high population mobility, small immigrant population, and small city size. As these factors are likely to persist and independently a¤ect policy today, empirically identifying the policy e¤ects of manager government seems challenging.
T
Manager government, however, also lled a growing need for technical expertise at the
RI P
top of city government, a need felt more acutely in times of crisis. To mitigate the e¤ects of natural disasters the new technologically-intensive public infrastructure, e.g., levees and bridges, had to be e¤ectively managed. The city of Dayton, Ohio provides an illustration.
SC
In March 1913 after days of heavy rainfall the Great Miami River overowed the citys levees causing a ood that destroyed over 20,000 homes. In the immediate aftermath local leaders
NU
sought to rebuild the ood control system with a large public works campaign. Expediency dictated the adoption of manager government so that an engineer could be appointed to lead the reconstruction e¤ort.
MA
Subsequent crises such as the Great Mississippi Flood of 1927 and the Northeast Flood of 1936 resulted in substantial losses across multiple local jurisdictions and helped swing the
ED
balance toward federal takeover of ood control from local authorities. In 1936 Congress passed the Flood Control Act (FCA) that moved responsibility for ood prevention and management to the Army Corps of Engineers. The hundreds of miles of levees and 375 major
PT
reservoirs constructed by the Army Corps of Engineers after 1936 signi cantly weakened the link between heavy precipitation and the incidence of oods.
CE
That adopting a manager government had been a local response to ood hazards before 1936 suggests an IV identi cation strategy: using precipitation shocks during the local ood control era (1900-1936) to isolate exogenous variation in manager government.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Data. Our sample consists of all U.S. cities with year-1900 Census population over 17,500 residents. This sample selection criterion is not a¤ected by government form choice, since manager reforms do not occur until the small city of Staunton, Virginia (pop. 7,289 in 1900) starts experimenting with it in 1908. After dropping Washington DC, since it has a federally-appointed city government until 1973, a number of 248 present-day cities satisfy this criterion. The sample period for our panel is 1960-2000.14 Our government form data comes from ICMAs Municipal Year Book. We point out two features of government form variation in our sample. First, the maps in Figures 1 and 2 display little geographic clustering in manager government. Second, Figure 3 shows that most manager charter adoptions in our sample occur before 1960. In fact, the majority of 14
Two intervening annexations and one merger slightly alter the sample: Pittsburgh, PA annexed Allegheny, PA in 1907; Omaha, NE annexed South Omaha, NE in 1915; and West Hoboken, NJ merged with Union Hill, NJ to form Union City, NJ in 1925.
15
ACCEPTED MANUSCRIPT adoptions occurred before the federal takeover of ood control in 1936, and only ten occurred after 1960. We measure local policymaking behavior using city employment during 1960-2000. This
T
policy area has the attractive feature that it allows us to distinguish popular from neu-
RI P
tral policy issues by disaggregating police employment into o¢cer and civilian employees, according to the distinction made in the FBIs Uniform Crime Reports.15 One may wonder how much inuence city managers have over the hiring of city employees.
SC
One way to examine this question is by looking at city managers "de jure" personnel powers as established in the charters of manager cities. Another approach is to gauge city managers
NU
"de facto" powers over city employment by estimating manager xed e¤ects. In terms of "de jure" powers, a standard handbook of "model" city charters prescribes under the section "Powers and Duties of the City Manager" that "The city manager shall
MA
appoint and, when necessary for the good of the service, suspend or remove all city employees and appointive administrative o¢cers provided for, by or under this charter, except as
ED
otherwise provided by law, this charter or personnel rules adopted pursuant to this charter. The city manager may authorize any administrative o¢cer subject to the managers direction and supervision to exercise these powers with respect to subordinates in that o¢cers
PT
department, o¢ce or agency." (Kemp 2003, p. 33) (our emphases added). In terms of "de facto" powers, Online Appendix Table A1 shows estimates of manager xed e¤ects for police
CE
and non-police city employees in the manager cities in our sample, using the methodology of Bertrand and Schoar (2003). If city managers have di¤erent employment policy stances and are e¤ective at implementing them, then we should nd that manager xed e¤ects capture
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
statistically signi cant variation in employment. Indeed, for all three employment categories manager xed e¤ects are highly signi cant, producing increases in the adjusted R2 for police o¢cers as high as 7:8 percentage points. We construct our instrument using 1900-2000 weather reports from the U.S. Historical Climatology Networks Daily Temperature, Precipitation, and Snow Data. This dataset contains daily readings for temperature extremes, rainfall, and snowfall collected from weather stations throughout the U.S. We match a city to the closest weather station based on geographic coordinates.16 Our main precipitation measure is the yearly sum of rainfall and
15
Census of Governments city employment data has the advantage that it distinguishes between full-time and part-time city employees, but distinguishes between police o¢cers and civilians only starting in 1977. 16 The U.S. has 126 weather stations reporting in 1900. The median distance to the closest weather station for our sample cities is 47 miles. As the opening of new stations could be related to changes in local weather or local economy we keep every city matched to the same station over time.
16
ACCEPTED MANUSCRIPT density-adjusted snowfall at the station level.17 We de ne an extreme precipitation event as a year when precipitation exceeds the 99th percentile of the national 20th century yearly precipitation distribution. Our cross-sectional measure of precipitation shocks for a given
T
city is the frequency of extreme precipitation events in a given period, referred to below (1900-1936) are referred to as LFC precipitation shocks.
RI P
as precipitation shocks. For instance, precipitation shocks in the local ood control period In addition to these key variables, we analyzed an extensive set of geographic, demo-
SC
graphic, economic, and crime variables. The Data Appendix (see Online Appendix A.6) provides the complete list of variables, with details about their measurement and sources.
NU
Table 1 reports descriptive statistics for our major variables by government form. Panel A shows that manager cities employ on average about 21% fewer o¢cers per capita than mayor cities, and virtually the same number of civilians per capita. Interestingly, Panel B
MA
suggests that before the advent of manager government there were no major di¤erences in either type of police employment. Other statistically signi cant di¤erences are: manager
ED
cities have on average higher reported crime, higher fraction of cleared crime (Panel A), higher incidence of Progressive institutions and Republican mayors (most manager cities maintain an honorary mayor position), lower non-police employment per capita (Panel C),
PT
and higher fraction college graduates (Panel D). Figure 4 provides a rst look at how manager government a¤ected police employment
CE
historically. Manager governments employ fewer police o¢cers from 1960 onward, but not in 1900. In Panel B we divide cities based on whether they experienced LFC precipitation shocks and nd that cities hit by these shocks have lower o¢cer employment after 1960, with
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
the di¤erence attenuating over time. Managers and Popular Policy. We rst estimate a simple model of the form: log(Employmenti;t ) = 1 M anageri;t + 02 xi;t +
t
+ i;t
(1)
where Employmenti;t is city employees per capita for city i in year t, M anageri;t is a dummy variable indicating manager form of city government, xi;t is a vector of controls,
t s
are year
xed e¤ects, and i;t is the error term. The coe¢cient 1 measures the conditional percentage di¤erence in mean employment between manager and mayor cities. According to hypothesis (H1) we expect 1 < 0 for police o¢cer employment and 1 = 0 for neutral employment categories. 17 We adjust snowfall for water density by dividing it by ten, as suggested by the U.S. Department of Agriculture. See http://www.ak.nrcs.usda.gov/snow/data.
17
ACCEPTED MANUSCRIPT This model allows us to account for measurable city geographic, demographic, economic, and political characteristics, and to control for national trends a¤ecting local police employment. As government form changes infrequently during 1960-2000 and the observations are
T
unlikely to be independent within a city we cluster the standard errors at the city level.
RI P
Table 2 presents OLS estimates of the o¢cer employment di¤erential. Accounting for year and division xed e¤ects (the U.S. has nine Census divisions)18 column (1) shows that manager governments employ 11:9% fewer o¢cers per capita. Adding demographic
SC
controls to account for voter preferences in column (2) shows that the point estimate remains negative and highly statistically signi cant.19 Manager government may be related to other
NU
institutional and political factors that could a¤ect policy. The descriptive statistics in Table 1 reect the historical fact that manager reform often came on the heels of other urban reforms: non-partisan ballots, at-large elections, and civil service rules. The literature has
MA
sometimes packaged these other reforms together with manager reform, distinguishing only between "traditional" and "reformed" cities (Stucky 2005). Manager cities may also di¤er
ED
in media penetration and police unionization, which may a¤ect the voter expertise () and policymaker congruence ( P ). Adding these institutional and political controls in column (3) does little to alter the estimate. Finally, in column (4) manager remains negatively
PT
correlated with o¢cer employment when controlling for city spending. In our theory model hierarchical accountability performs an informational role in the
CE
sense that it more often allows the policymaker to act on his private expertise. It is then useful to compare the magnitudes in Table 2 with impacts of information on government policy estimated in prior work. For instance, Strombergs (2004) estimated 0.201 elasticity
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
of federal unemployment relief spending with respect to radio penetration (see his Table II column IV) would imply a 20.1% upper bound, in absolute value, on the o¢cer di¤erential, corresponding to a change from zero radio penetration to full penetration. All the estimates of the o¢cer di¤erential in Table 2 are within this range. Endogeneity. Even if the OLS speci cations above control for all relevant confounds, estimates may still be biased by reverse causality. Despite the infrequency of actual changes in accountability form (e.g., 0.8% of 1,420 U.S. cities had changed their government form between 1980-90 according to Table 1 in Baqir 2002) city charters are endogenous by virtue 18
The U.S. Census divisions are: New England, North Atlantic, South Atlantic, East North Central, East South Central, West North Central, West South Central, Mountain, and Paci c. 19 Voter or special interest preferences can a¤ect institutional choice. Morelli and Van Weelden (2013) argue that pandering can depend on how divisive an issue is for the electorate and Aghion, Alesina, and Trebbi (2004) discuss how the distribution of voter preferences may a¤ect the choice of institutions.
18
ACCEPTED MANUSCRIPT of being subject to revision by popular referendum. Measurement error is another potential source of bias in OLS estimates leading to attenuation. Because manager cities typically maintain an honorary mayor, city clerks in these cities sometimes mistakenly report a mayor
T
form of government on ICMA survey forms (see Coate and Knight 2011).
RI P
To address concerns with bias in the OLS estimates we develop an IV approach. Our strategy is based on public infrastructure crises triggering early 20th century switches to manager government to facilitate the ascension of engineers into the top executive o¢ce. Floods
SC
caused by extreme precipitation before the 1936 Flood Control Act were one such relevant crisis; see Historical Background above. We thus instrument for manager government using
NU
the frequency of extreme precipitation events in the local ood control era (1900-1936), in short, LFC precipitation shocks. Formally, the rst stage model is:
MA
M anageri;t = 1 LF C_P recipitation_Shocksi + 02 wi;t +
t
+ i;t
(2)
where LF C_P recipitation_Shocksi is the frequency of precipitation shocks in the local
ED
ood control era for city i, wi;t is a vector of controls,
t s
are year xed e¤ects, and i;t is
the error term. The controls include Century_P recipitation_Shocksi , the frequency of city
PT
precipitation shocks during the 20th century, and M edian_P recipitationi , median annual city precipitation during the 20th century. These variables help make the exclusion restriction
CE
credible as previous research found that climate a¤ects economic growth (Dell, Jones, and Olken 2012), crime (Jacob, Lefgren, and Moretti 2007), conict (Miguel, Satyanath, and Sergenti 2004), and the origins of trust (Durante 2010).20
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Because we control for century precipitation shocks, the identi cation comes from the timing of precipitation shocks (pre vs. post 1936) and not simply their occurence. Our identi cation assumptions are that, conditional on typical local precipitation patterns: (i) cities hit by precipitation shocks in the LFC era have the same average unobserved characteristics as spared cities, and (ii) LFC precipitation shocks a¤ect police employment decades later only through their e¤ect on government form. In the Online Appendix we provide supportive evidence for these conditions. In support of instrument validity and the exclusion restriction we nd that: (i) LFC precipitation shocks are not correlated with early city characteristics (Online Appendix Table A2 Panel A); (ii) LFC precipitation shocks are not related to other present-day city institutions (Online Appendix Table A2 Panel B), and (iii) in contrast to 20
In a related IV strategy using country-level data Bruckner and Ciccone (2011) exploit the fact that in non-democratic societies the cost of popular opposition to an authoritarian regime is lower during times of economic distress, making negative rainfall shocks correlated with democracy.
19
ACCEPTED MANUSCRIPT LFC precipitation shocks, federal ood control (FFC) precipitation shocks (1936-1960) are not related to police o¢cer employment today (Online Appendix Table A3). Trends in city managers educational backgrounds provide additional support for our
T
identi cation strategy. Progressive Era city managers were disproportionately engineers.
RI P
Later in the century, however, the manager profession bacame dominated by individuals trained in public administration (Stillman 1977). More than 95 percent of city managers were engineers in 1918, 77 percent in 1934, and only 18 percent in 1977. In other words, LFC
SC
precipitation shocks are related to manager government for reasons largely obsolete today. Table 3 presents IV estimates of the o¢cer employment di¤erential controlling for typical
NU
local precipitation patterns, together with OLS counterparts. Columns (1),(2) control for year xed e¤ects and typical precipitation patterns. Columns (3),(4) add water proximity controls, e.g., swamps, rivers, to account for the fact that residents of cities close to water
MA
may implement private solutions to ood risk. Columns (5),(6) add division xed e¤ects. The rst stage shows a strong relationship between LFC precipitation shocks and present-
ED
day manager government: the F -statistic always exceeds the critical value of 10 below which nite-sample weak-instrument bias could be a concern (Bound, Jaeger, and Baker 1995). In the second stage we nd signi cantly lower o¢cer employment in manager cities. The larger
PT
IV point estimates relative to the baseline OLS results suggest that measurement error in government form might be present in our sample (Coate and Knight 2011); another factor
CE
could be unobserved voter preferences, e.g., more conservative cities are both more likely to adopt manager government and have a tougher stance on crime by demanding more police. Overall the IV results uphold the substantive conclusions derived from the OLS estimates.21
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
If the costs of changing government form in response to precipitation shocks are heterogeneous, the identi ed e¤ects will be local to the subset of responsive cities and potentially di¤erent from the population-wide treatment e¤ect. For example, Acemoglu, Robinson, and Torvik (2013) argue that voters dismantle exogenously imposed checks and balances when special interests are strong. One way to explore this issue is to examine the sensitivity of the IV estimates to changes in instrument construction. If the e¤ects of government form are heterogeneous and highly speci c to cities induced to change government form by our proposed instrument, alternative instrument de nitions would likely change the magnitudes of the estimated e¤ects. Online Appendix Tables A4 and A5 show that the IV results are by and large robust to altering the instrument de nition along several dimensions, strengthening their external validity. 21 Baqir (2002) and Whalley (2013) nd little evidence that related forms of institutional endogeneity contaminate OLS results in the context of U.S. local governments.
20
ACCEPTED MANUSCRIPT Using a di¤erent sample, a di¤erent period, and exploiting switches to manager government Coate and Knight (2011) nd that managers spend more than mayors. We nd that manager governments employ fewer police o¢cers. To reconcile these seemingly incongruous
T
results we examine total city government spending and police department spending data in
RI P
Online Appendix Table A6. Columns (1),(2) replicate Coate and Knights (2011) nding in our sample; accounting for endogeneity in column (3) weakens the manager-mayor total spending di¤erence although the point estimate remains positive and sizable. Columns
SC
(4),(5) imply little di¤erence in police spending per capita while column (6) shows smaller police spending by managers, with the estimate not statistically signi cant.
NU
Online Appendix Table A7 reports sensitivity checks for our preferred IV speci cation and two related OLS speci cations. We examine robustness to sample alterations, alternative inference procedures, and alternative dependent variable scaling; the estimates seem quite
MA
robust and inferences from both OLS and IV continue to hold. As one might expect, the manager e¤ects are somewhat larger in bigger cities, model (2), and prior to the federal
ED
intervention in local policing through the COPS program, model (4). Our hypothesized causal mechanism from LFC precipitation shocks to government form runs through ood-related infrastructure crises, yet direct measures of ood damage or ood
PT
intensity at the beginning of the twentieth century are sparse due to local management of ood control before 1936. We nevertheless provide two ways to make this connection. The
CE
rst uses a cross-sectional soil-based measure of ood incidence developed by U.S. Department of Agriculture scientists. The second uses a time-varying measure of ooding based on riverow readings as monitored by the U.S. Geological Service. We develop alternative
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
measures of precipitation shocks, based on di¤erent critical volume thresholds and di¤erent time aggregation, in order to check which ones are more tightly linked with these ooding proxies. The results are in Online Appendix Tables A8 and A9. Overall, we nd signi cant correlations between extreme precipitation and ooding. While there are tradeo¤s between the di¤erent precipitation measures, the one based on the top 1% volume threshold and annual aggregation seems to perform best with our yearly dataset. Other Employment Categories. Table 4 reports OLS and IV estimates of managermayor di¤erences in city employment other than police o¢cers. The rst three columns look at civilians hired in the police department, e.g., dispatchers, administrators, forensic sta¤. This can be considered a neutral policy issue since voters likely have no clear ex ante preference over high or low police civilian employment. The OLS estimates are positive and closer to zero than the negative estimates in the police o¢cer regressions, and not 21
ACCEPTED MANUSCRIPT statistically signi cant. The IV estimate is also positive, and somewhat larger, possibly reecting a substitution response between o¢cer and civilian employment, although still not signi cantly di¤erent from zero. While the lower number of o¢cers in manager cities may
T
be consistent with managers weaker patronage motivations (Enikolopov 2014) the lack of a
RI P
similar pattern for civilians seems to indicate that other types of incentives also matter. The last three columns consider non-police employees, e.g., re ghters, sanitation workers, assessors. We note that for 65.7% of our sample cities public education is provided
SC
through independent school districts; these cities have little direct control over teacher employment which falls under the jurisdiction of the school board and the school districts
NU
superintendent. The OLS estimates for non-police employees are similar in magnitude to the ones for o¢cers in Table 2, although less statistically signi cant; the IV estimate is substantially smaller. These negative point estimates may reect weaker pandering incentives
MA
for some of these employment categories, or may be consistent with a patronage explanation; we explore this issue further below in Table 5 and Online Appendix Table A10.
ED
Policy Volatility. In the insulation equilibrium of our model the manager is free to adjust the policy to match the stochastic state. The mayor, in contrast, is politically constrained by
PT
public pressure to pick the popular policy regardless of the state. This implies that popular policies should be more volatile in manager cities (hypothesis H2). Table 5 reports OLS and
CE
IV estimates of manager-mayor di¤erences in policy volatility.22 In columns (1)-(3) we nd a positive and statistically signi cant di¤erence in volatility for the popular policy, namely police o¢cer employment per capita. Columns (4)-(6) show that
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
the managers higher propensity to adjust o¢cers is not due to a higher crime rate volatility in manager cities. As we assume that the probability of crime (latent crime) is exogenous to government form, the di¤erences in committed crime (actual crime) volatility estimated here suggest that police employment a¤ects committed crime di¤erentially by government form. In the bottom half of the table we nd no signi cant di¤erences in volatility for police civilians per capita or non-police employees per capita. The estimates have for the most part a negative sign, opposite to the pattern for o¢cer volatility, and are closer to zero. The contrasting coe¢cient patterns for popular vs. neutral policy issues provide support for the types of incentives highlighted by our model. 22
volatility in a time-series variable yi;t as the log of relative deviations from mean, We measure PT 1 PT 1 log yi;t T t=1 yi;t = T t=1 yi;t . As extreme outliers may represent measurement error we trim the sample by removing the top and bottom 1% extreme values. For ease of comparison we use the 1975-2000 period when crime data is available.
22
ACCEPTED MANUSCRIPT Informational Mechanism. In our theory model a popular policy di¤erential strictly speaking only exists for low-voter-expertise high-policymaker-congruence cities where voters can credibly commit to insulate the policymaker. In Table 6 we examine (H3) which states
T
that the policy di¤erential for popular issues should decrease in the voters policy expertise.
RI P
We use the "crack epidemic" that started to a¤ect U.S. cities in the early 1980s as a shock to voter expertise measured by the voters crime probability prior (). The emergence of crack cocaine and associated media attention increased the match between the popular
SC
policy of high police and the optimal policy, as it heightened public awareness of crime in urban areas.23 Importantly for identi cation the geographic spread of crack was driven
NU
by proximity to a few entry points and other topographic conditions, rather than by local enforcement activities (Evans, Garthwaite, and Moore 2012). Based on Proposition 1 we expect the increase in the voters crime probability prior due to the crack epidemic to produce
MA
a manager pandering equilibrium, leading to a reduction in the negative manager-mayor o¢cer di¤erential toward zero. In Table 6 we add to equation (1) an interaction of M anageri;t
ED
with this proxy for the informational mechanism. The results in columns (1)-(3) show that the crack epidemic shrinks the negative o¢cer di¤erential, consistent with a pandering
PT
equilibrium in manager cities with high voter expertise. Electoral E¤ects. Our incentives model predicts that pandering is strongest pre-election
CE
because the voter rationally relies more heavily on recent information about policy choices. We measure election proximity by coding the year before the election date as an election year.24 To estimate electoral e¤ects we add to equation (1) the election indicator
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Electioni;t and the interaction M anageri;t Electioni;t , as well as city xed e¤ects. In this model M anager measures the manager-mayor policy di¤erential in non-election years, while M anager + M anagerElection measures the manager-mayor policy di¤erential in election years. Hypothesis (H4) states that the election year employment di¤erential is more negative than the non-election year di¤erential for popular issues, meaning M anagerElection < 0 for o¢cers. Table 7 column (1) shows that the interaction coe¢cient is negative, consistent with (H4), yet somewhat statistically weak. To increase precision we notice from the robustness 23
See Blenko (1993). Economists and other social scientists have provided evidence that the crack epidemic increased crime. For example, Grogger and Willis (2000) conclude that "Our results indicate that, in the absence of crack cocaine, the crime rate in 1991 would have remained below its previous peak in the early 1980s." (p. 528). 24 Election dates are set in the city charter; special elections outside the regular election cycle, say to replace a mayor resigning before the end of the term, are infrequent and we ignore them. For manager government we use mayoral-council elections. We cross-checked elections data against multiple sources to resolve inconsistencies we found among sources.
23
ACCEPTED MANUSCRIPT analysis, Online Appendix Table A7 Model (2), that mean policy di¤erentials are greater in larger cities.25 Columns (2)-(4) therefore restrict the sample to the subset of 174 cities with population over 50,000 in 1960. The negative interaction coe¢cient from column (1)
T
reappears in columns (2)-(3) and is now also statistically signi cant at conventional levels.
RI P
The employment cycle could be confounded by a spending cycle although in Online Appendix Table A10 we nd neither a di¤erential spending cycle nor a di¤erential employment cycle in other employment categories. The further addition of government spending in colThe election-year di¤erential is
0:008
0:013; or
SC
umn (4) leaves the estimate of M anagerElection and its standard error virtually unchanged. 0:8 percentage points and statistically
NU
negative, while the non-election-year di¤erential is
2:1 percentage points, and statistically
zero. Note also that the election-year di¤erential is not driven by fewer election-year o¢cers in manager cities, but by more election-year o¢cers in mayor cities, namely 0:9% more.
MA
These results are consistent with the predictions of our incentives model as opposed to a selection model by preference type which would predict a positive election-year di¤erential
Conclusion
PT
4
ED
(see Section 2 and Online Appendix A.4).
This paper sheds light on the incentive e¤ects of an accountability form that is pervasive at
CE
various levels of governments. Our principal-intermediary-agent model gives informational asymmetries a central role and shows that hierarchical accountability creates either insulation or pandering incentives depending on informational and political factors. It implies
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
that in a setting with government-voter expertise asymmetries this structure rather than hindering democratic accountability may in fact increase the exibility of voter control over policymaking incentives, by reducing pandering when it is detrimental to voter welfare. The theory guides our empirical analysis of U.S. city managers in a salient area of local policy, namely public safety. The estimates for the main institutional e¤ect and its mechanisms are consistent with the theoretically characterized incentives. For simplicity we assumed that informational asymmetries are exogenous to the institutional environment, yet motivations for acquiring information by the voter or the policymaker (Stephenson 2011) may depend on the accountability form: on one hand, the voter might be less interested in acquiring information if he cannot directly remove the policymaker; on the 25
Levitt (1997) also argues that electoral cycles should be more pronounced in larger cities. He exploits the correlation between election years and police hiring in a panel of 59 large U.S. cities to identify the e¤ect of police on crime.
24
ACCEPTED MANUSCRIPT other hand, a hierarchically-accountable policymaker, who enjoys more policy discretion, may acquire more policy information. Thus, the informational asymmetry may be more severe under hierarchical accountability. Empirically, a more stringent test of the model
T
would be to verify that voter and intermediary strategies, in addition to policy outcomes,
RI P
adjust to the informational and political environment as shown here to provide optimal incentives for a hierarchically-accountable policymaker. This requires more detailed data on council elections and manager retention (see Enikolopov 2014 for an e¤ort in this direction).
SC
Finally, one could apply our framework to study the incentives of indirectly-accountable policymakers with xed terms. We conjecture that in that setting voters have less inuence
CE
PT
ED
MA
NU
over policymaker incentives since the retention signal would be weaker.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
25
ACCEPTED MANUSCRIPT FIGURE 1: Manager Cities, 1960
MA
NU
SC
RI P
T
Notes: Authors’ calculations for the sample of the 248 largest self-governing cities as of 1900. Manager government locations are plotted based on 1960 form of government. The map reflects state jurisdictional boundaries.
CE
PT
ED
FIGURE 2: Mayor Cities, 1960
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Notes: Authors’ calculations for the sample of the 248 largest self-governing cities as of 1900. Mayor government locations are plotted based on 1960 form of government. The map reflects state jurisdictional boundaries.
26
ACCEPTED MANUSCRIPT FIGURE 3: Timing of Manager Charter Adoptions
1900
NU
0
1
SC
Frequency 2 3
RI P
T
4
5
1920
1940
1960
1980
2000
Year
MA
Notes: Authors’ tabulations using adoption data from the Municipal Year Book, newspapers, and city charters for the sample of the 248 largest selfgoverning cities as of 1900. Year of manager government adoption is based on year of charter approval by voters.
PT
ED
FIGURE 4: Police Employment, by Indicated Cohort
CE
Panel B: Officers Per Capita; by LFC Precipitation Shocks .4
1
Panel A: Officers Per Capita; by Government Form
.2 Mean Residuals -.2 0 -.6
1900
1920
1940
1960
1980
2000
1920
1940
1960
1980
2000
Notes: Authors’ calculations for the sample of the 248 largest self-governing cities as of 1900. Panel B plots the mean residual from a regression of log(Police Officers Per Capita) on Century Precipitation Shocks and Median Precipitation.
27
1900
Year
Year
LFC Precipitation Shocks No LFC Precipitation Shocks
-.4
Mean log(Officers Per Capita) .2 .4 .6 .8
AC
Manager 1960 Mayor 1960
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
ACCEPTED MANUSCRIPT TABLE 1: Descriptive Statistics, By Government Form Manager Cities mean std. dev. (1) (2)
Mayor Cities mean std. dev. (3) (4)
mean diff. p-value (5)
0.53 0.24 227 0.43 32.07 0.11
2.25 0.39 198 0.31 81.00 0.25
1.08 0.36 213 0.44 39.27 0.15
0.000 0.890 0.576 0.266 0.003 0.005
1.10 0.06 0.05
0.46 0.06 0.04
0.935 0.074 0.022
0.50 0.43 0.46 0.45 1.88 2,681 12.47
0.000 0.000 0.886 0.004 0.610 0.548 0.040
RI P
1.86 0.38 204 0.37 92.96 0.28
SC
Police Officers per Capita Police Civilians per Capita Police Dept. Spending per Capita Fraction Officers Unionized (1968) Crime Rate (1975-2000) Fraction Crimes Cleared (1975-2000)
T
Panel A: Police Department, 1960-2000
Panel B: Police Department, 1900
Police Officers per Capita Police Civilians per Capita Arrests per Capita
0.34 0.11 0.05
MA
NU
1.10 0.08 0.07
Panel C: City Government, 1960-2000
PT
ED
Non-Partisan Elections (1960) Fraction At-Large Seats (1960) Early Civil Service (1960) Incumbent Mayor Republican Local Newspaper Sales (1990-2000) City Gov. Spending per Cap. Non-Police Empl. per Cap. (1972-2000)
0.85 0.83 0.69 0.47 0.42 2,115 15.62
0.35 0.34 0.46 0.50 1.06 2,584 11.69
0.53 0.55 0.70 0.27 0.51 2,208 18.59
Panel D: City Demographics, 1960-2000
CE
Population Population Density Fraction Non-White Fraction College Graduate Household Income Observations
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
146,198 184,417 5,280 19,744 0.19 0.15 0.15 0.08 28,653 7,477 3,622
252,521 712,626 6,556 8,837 0.20 0.20 0.13 0.08 28,241 7,498 6,546
0.074 0.732 0.900 0.020 0.487 10,168
Notes: Authors’ calculations with 1960-2000 city data, unless otherwise indicated. The unit of observation is a city-year for the sample of 248 largest, as of 1900, self-governing cities in the United States. Each entry in columns (1)-(4) presents the mean/std. dev. for the indicated variable and column (5) has pvalues for the hypothesis that the difference between the means in columns (1) and (3) is zero, using standard errors clustered at the city level. The number of observations reflects the maximum sample size across all the reported variables. All monetary values are expressed in 2000$.
28
ACCEPTED MANUSCRIPT TABLE 2: Government Form and Popular Policy Model:
SC
NU
Year Fixed Effects Division Fixed Effects Demographics Institutional Fiscal Period Cities Observations
RI P
T
Manager
OLS OLS OLS OLS (1) (2) (3) (4) Dependent Variable = log(Police Officers per Capita) í0.119*** í0.083*** í0.069*** í0.076*** (0.033) (0.027) (0.030) (0.029) Yes Yes Yes Yes Yes Yes Yes Yes No Yes Yes Yes No No Yes Yes No No No Yes 1960-2000 1960-2000 1960-2000 1960-2000 248 248 248 248 9,974 9,974 9,974 9,974
CE
PT
ED
MA
Notes: Authors’ calculations with 1960-2000 city data as described in the Data Appendix. The unit of observation is a city-year for the sample of the largest 248 self-governing cities in the United States as of 1900. The table reports OLS estimates of ȕ1 from equation (1) in the text. Standard errors clustered at the city level in parentheses. Demographic controls: Population Density; Fraction Non-White; Fraction College Graduate; Median Household Income. Institutional controls: Incumbent Mayor Republican; NonPartisan Elections 1960; Fraction At-Large Seats 1960; Early Civil Service; Police Unionization 1968; Local Newspaper Sales 1990-2000. Fiscal controls: log(City Government Spending per Capita). Indicator variables for missing values added. * indicates significance at the 10 percent level, ** significance at the 5 percent level, and *** significance at the 1 percent level.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
29
ACCEPTED MANUSCRIPT TABLE 3: Government Form and Popular Policy: Endogeneity
[p-value]
OLS IV OLS IV OLS (2) (3) (4) (5) (6) Dependent Variable = log(Police Officers per Capita) í0.284** í0.143*** í0.270** í0.098*** í0.394* í0.086*** (0.140) (0.035) (0.127) (0.031) (0.207) (0.032) First Stage: Dependent Variable = Manager 11.675*** – 13.676*** – 9.875*** – (2.851) (2.566) (2.676) 16.76 – 28.40 – 13.62 – [0.0001] [0.0003] [0.0000]
Year Fixed Effects Weather Water Division Fixed Effects Period Cities Observations
Yes Yes No No 1960-2000 248 9,974
Model:
Yes Yes No No 1960-2000 248 9,974
Yes Yes Yes No 1960-2000 248 9,974
NU
F-stat
MA
LFC Precip. Shocks
SC
RI P
T
Manager
IV (1)
Yes Yes Yes No 1960-2000 248 9,974
Yes Yes Yes Yes 1960-2000 248 9,974
Yes Yes Yes Yes 1960-2000 248 9,974
CE
PT
ED
Notes: Authors’ calculations with 1960-2000 city data as described in the Data Appendix. The unit of observation is a city-year for the sample of largest 248 self-governing cities in the United States as of 1900. The first panel reports estimates of ȕ1 in equation (1) in the text. The second panel reports estimates of D1 in equation (2) in the text. Standard errors clustered at the city level in parentheses. Fstatistic corresponds to the excluded instrument LFC Precipitation Shocks with p-values in brackets. Weather controls: Century Precipitation Shocks; Median Precipitation. Water controls: number of Lakes; number of Swamps; one indicator for location on Coast; three indicators for Small Rivers, Medium Rivers, Large Rivers. * indicates significance at the 10 percent level, ** significance at the 5 percent level, and *** significance at the 1 percent level.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
30
ACCEPTED MANUSCRIPT TABLE 4: Government Form and Other Employment
NU
Year Fixed Effects Weather Water Division Fixed Effects Period Cities Observations
T
Manager
OLS OLS IV (4) (5) (6) log(Non-Police Empl. Per Capita) í0.083 í0.104* í0.036 (0.068) (0.056) (0.361) Yes Yes Yes Yes Yes Yes Yes Yes Yes No Yes No 1972-2000 1972-2000 1972-2000 246 246 246 6,702 6,702 6,702
RI P
Dep. Var. =
OLS OLS IV (1) (2) (3) log(Police Civilians per Capita) 0.082 0.015 0.162 (0.070) (0.070) (0.253) Yes Yes Yes Yes Yes Yes Yes Yes Yes No Yes No 1960-2000 1960-2000 1960-2000 248 248 248 9,850 9,850 9,850
SC
Model:
CE
PT
ED
MA
Notes: Authors’ calculations with city data as described in the Data Appendix. The unit of observation is a city-year for the sample of the largest 248 self-governing cities in the United States as of 1900. Columns (1)-(6) report estimates of ȕ1 from equation (1) in the text. Standard errors clustered at the city level reported in parentheses. Weather controls: Century Precipitation Shocks; Median Precipitation. Water controls: number of Lakes; number of Swamps; one indicator for location on Coast; three indicators for Small Rivers, Medium Rivers, Large Rivers. * indicates significance at the 10 percent level, ** significance at the 5 percent level, and *** significance at the 1 percent level.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
31
ACCEPTED MANUSCRIPT TABLE 5: Government Form and Volatility
Model: Dep. Var. =
Manager Cities Observations Year Fixed Effects Weather Water Division Fixed Effects Period
OLS OLS IV (7) (8) (9) log(Police Civil. Variation Coef.) –0.112 –0.088 –0.325 (0.076) (0.083) (0.259) 247 247 247 6,310 6,310 6,310 Yes Yes Yes Yes Yes Yes Yes Yes Yes No Yes No 1975-2000 1975-2000 1975-2000
T
RI P
Cities Observations
OLS OLS IV (4) (5) (6) log(Crime Rate Variation Coef.) í0.147*** í0.149*** í0.455** (0.056) (0.056) (0.205) 248 248 248 6,024 6,024 6,024 OLS OLS IV (10) (11) (12) log(Non-Police Variation Coef.) í0.090 í0.071 0.002 (0.097) (0.090) (0.270) 246 246 246 5,850 5,850 5,850 Yes Yes Yes Yes Yes Yes Yes Yes Yes No Yes No 1975-2000 1975-2000 1975-2000
SC
Manager
NU
Dep. Var. =
OLS OLS IV (1) (2) (3) log(Police Officer Variation Coef.) 0.210*** 0.148** 0.507* (0.070) (0.067) (0.288) 246 246 246 6,336 6,336 6,336
MA
Model:
CE
PT
ED
Notes: Authors’ calculations with 1975-2000 city data as described in the Data Appendix. The unit of observation is a city-year. The sample is of the largest 248 self-governing cities in the United States as of 1900. Columns (1)-(6) report estimates of ȕ1 from equation (1) in the text. Standard errors clustered at the city level reported in parentheses. The dependent variable is calculated as log(|yit – average[yit]|/average[yit]), for the yit observations between the bottom 1 percentile and the top 1 percentile of its sampling distribution. Weather controls: Century Precipitation Shocks; Median Precipitation. Water controls: number of Lakes; number of Swamps; one indicator for location on Coast; three indicators for Small Rivers, Medium Rivers, Large Rivers. * indicates significance at the 10 percent level, ** significance at the 5 percent level, and *** significance at the 1 percent level.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
32
ACCEPTED MANUSCRIPT TABLE 6: Government Form and Popular Policy: Mechanisms OLS (1)
Model: Dep. Variable =
OLS (3)
RI P
T
Manager
OLS (2)
log(Police Officers per Capita) –0.096*** –0.087*** –0.095*** (0.026) (0.031) (0.031) 0.041* 0.047** 0.050** (0.023) (0.023) (0.023) –0.030 –0.036 –0.034 (0.028) (0.035) (0.035) Yes Yes Yes Yes Yes Yes Yes Yes Yes No Yes Yes No No Yes 1960-2000 1960-2000 1960-2000 248 248 248 9,974 9,974 9,974
Manager × Crack Epidemic
SC
Crack Epidemic
MA
NU
Year Fixed Effects Division Fixed Effects Demographics Institutional Fiscal Period Cities Observations
CE
PT
ED
Notes: Authors’ calculations with 1960-2000 city data as described in the Data Appendix. The unit of observation is a city-year for the sample of largest 248 self-governing cities in the United States as of 1900. Columns (1)-(3) report OLS estimates of equation (1) in the text augmented with the indicated variables and interaction terms. Standard errors clustered at the city level reported in parentheses. Demographic controls: Population Density; Fraction Non-White; Fraction College Graduate; Median Household Income. Institutional controls: Incumbent Mayor Republican; Non-Partisan Elections 1960; Fraction At-Large Seats 1960; Early Civil Service; Police Unionization 1968; Local Newspaper Sales 1990-2000. Fiscal controls: log(City Government Spending per Capita). Indicator variables for missing values added. * indicates significance at the 10 percent level, ** significance at the 5 percent level, and *** significance at the 1 percent level.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
33
ACCEPTED MANUSCRIPT TABLE 7: Government Form and Popular Policy: Electoral Effects Model: Dep. Variable =
OLS (2)
OLS (3)
OLS (4)
log(Police Officers per Capita) –0.078 –0.064 –0.045 –0.008 (0.056) (0.084) (0.048) (0.057) –0.009* –0.015** –0.013*** –0.013*** (0.005) (0.007) (0.005) (0.005) 0.006 0.010** 0.009** 0.009** (0.004) (0.004) (0.004) (0.004) – – – –0.005 (0.007) – – – 0.005 (0.004) Yes Yes Yes Yes Yes Yes Yes Yes No No Yes Yes 1960-2000 1960-2000 1960-2000 1960-2000 248 174 174 174 9,974 7,033 7,033 7,033 All Above 50k Above 50k Above 50k
Manager × Election Election Manager × log(City Gov. Spend. per Cap.)
NU
log(City Gov. Spend. per Cap.)
SC
RI P
T
Manager
OLS (1)
MA
Year Fixed Effects City Fixed Effects Demographics Period Cities Observations 1960 City Population
CE
PT
ED
Notes: Authors’ calculations with 1960-2000 city data as described in the Data Appendix. The unit of observation is a city-year. In column (1) the sample is the largest 248 self-governing cities in the United States as of 1900. In the remaining columns the sample is the 174 cities that also have at least 50,000 residents in 1960. The table reports OLS estimates of equation (1) in the text augmented with the indicated variables and interaction terms. Standard errors clustered at the city level reported in parentheses. Demographic controls: Population Density; Fraction Non-White; Fraction College Graduate; Median Household Income. Indicator variables for missing values added. * indicates significance at the 10 percent level, ** significance at the 5 percent level, and *** significance at the 1 percent level.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
34
ACCEPTED MANUSCRIPT References Acemoglu, D., J. Robinson, and R. Torvik (2013) "Why Do Voters Dismantle Checks and Balances?" Review of Economic Studies 80(3): 845-875.
T
Acemoglu, D., G. Egorov, and K. Sonin (2012) "A Political Theory of Populism," Quar-
RI P
terly Journal of Economics 128(2): 771-805.
Aghion, P., A. Alesina, and F. Trebbi (2004) "Endogenous Political Institutions," Quar-
SC
terly Journal of Economics 119(2): 565-611.
Aruoba, B., A. Drazen, and R. Vlaicu (2015) "A Structural Model of Electoral Accountability," NBER working paper 21151.
NU
Ashworth, S. and K.W. Shotts (2010) "Does Informative Media Commentary Reduce Politicians Incentives to Pander?" Journal of Public Economics 94(11-12): 838-847.
MA
Baqir, R. (2002) "Districting and Government Overspending," Journal of Political Economy 110(6): 1318-1354.
Bertrand, M. and A. Schoar (2003) "Managing with Style: The E¤ect of Managers on
ED
Firm Policies," Quarterly Journal of Economics 118 (4): 1169-1208. Besley, T. and M. Smart (2007) "Fiscal Restraints and Voter Welfare," Journal of Public
PT
Economics 91(3-4): 755-773.
Blenko, S. (1993) Crack and the Evolution of Anti-Drug Policy, Westport, CT: Green-
CE
wood Press.
Bound, J., D.A. Jaeger, and R.M. Baker (1995) "Problems with Instrumental Variables Estimation when the Correlation Between the Instruments and the Endogenous Explanatory
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Variable is Weak," Journal of the American Statistical Association 90(430): 443-450. Bruckner, M. and A. Ciccone (2011) "Rain and the Democratic Window of Opportunity," Econometrica 79(3): 923-947. Canes-Wrone, B., M. Herron, and K. Shotts (2001) "Leadership and Pandering: A Theory of Executive Policymaking," American Journal of Political Science 45(3): 532-550. Canes-Wrone, B. and K. Shotts (2004) "The Conditional Nature of Presidential Responsiveness to Public Opinion," American Journal of Political Science 48(4): 690-706. Coate, S. and B. Knight (2011) "Government Form and Public Spending: Theory and Evidence from U.S. Municipalities," AEJ: Economic Policy 3(3): 82-112. Dell, M., B. Jones, and B. Olken (2008) "Climate Shocks and Economic Growth: Evidence from the Last Half Century," AEJ: Macroeconomics 4(3): 66-95. Durante, R. (2010) "Risk, Cooperation, and the Economic Origins of Social Trust: An Empirical Investigation," working paper, Sciences Po. 35
ACCEPTED MANUSCRIPT Enikolopov, R. (2014) "Politicians, Bureaucrats, and Targeted Redistribution," Journal of Public Economics 120: 74-83. Evans, W., C. Garthwaite, and T.J. Moore (2012) "The White/Black Educational Gap,
T
Stalled Progress, and the Long Term Consequences of the Emergence of Crack Cocaine
RI P
Markets," NBER working paper 18437.
Fox, J. (2007) "Government Transparency and Policymaking," Public Choice 131(1): 23-44.
SC
Gailmard, S. and J. Jenkins (2009) "Agency Problems, the 17th Amendment, and Representation in the Senate," American Journal of Political Science 53(2): 324-342.
NU
Gallup, Inc. (2000) "Crime Tops List of Americans Local Concerns," by Lydia Saad. http://www.gallup.com/poll/2800/crime-tops-list-americans-local-concerns.aspx. Gallup, Inc. (2010) "Americans Still Perceive Crime as on the Rise," by Je¤rey M. Jones.
MA
http://www.gallup.com/poll/144827/americans-perceive-crime-rise.aspx. Gordon, S.C. and G.A. Huber (2007) "The E¤ect of Electoral Competitiveness on In-
ED
cumbent Behavior," Quarterly Journal of Political Science 2(2): 107-138. Grogger, J. and M. Willis (2000) "The Emergence of Crack Cocaine and the Rise in Urban Crime Rates," The Review of Economics and Statistics 82(4): 519-529.
PT
Huber, G.A. and S.C. Gordon (2004) "Accountability and Coercion: Is Justice Blind when It Runs for O¢ce?" American Journal of Political Science 48(2): 247263.
CE
ICMA (2007) Council-Manager Form of Government, Washington, DC: ICMA Press. Jacob, B., L. Lefgren, and E. Moretti (2007) "The Dynamics of Criminal Behavior," Journal of Human Resources 42(3): 489-527.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Kartik, N., F. Squintani, and K. Tinn (2013) "Information Revelation and Pandering in Elections," working paper, Columbia University. Kemp, R.L. (ed.) (2003) Model Government Charters: A City, County, Regional, State, and Federal Handbook, McFarland Press. Knoke, D. (1982) "The Spread of Municipal Reform: Temporal, Spatial, and Social Dynamics," American Journal of Sociology, 87(6): 1314-1339. Levin, J. and S. Tadelis (2010) "Contracting for Government Services: Theory and Evidence from U.S. Cities," Journal of Industrial Economics 58(3): 507-541. Levitt, S. (1997) "Using Electoral Cycles in Police Hiring to Estimate the E¤ect of Police on Crime," American Economic Review 87(3): 270-290. Lim, C. (2013) "Preferences and Incentives of Appointed and Elected Public O¢cials: Evidence from State Trial Court Judges," American Economic Review 103(4): 1360-1397.
36
ACCEPTED MANUSCRIPT MacDonald, L. (2008) "The Impact of Government Structure on Local Public Expenditures," Public Choice 136(3-4): 457-473. Martinez, L. (2009) "A Theory of Political Cycles," Journal of Economic Theory 144(3):
T
11661186.
RI P
Maskin, E. and J. Tirole (2004) "The Politician and the Judge: Accountability in Government," American Economic Review 94(4): 1034-1054.
Miguel, E., S. Satyanath, and E. Sergenti, (2004) "Economic Shocks and Civil Conict:
SC
An Instrumental Variables Approach," Journal of Political Economy 112(4): 725-753. Morelli, M. and R. Van Weelden (2013) "Ideology and Information in Policymaking,"
NU
Journal of Theoretical Politics 25 (3): 412-439.
Park, C. (2000) "Monitoring and Structure of Debt Contracts," Journal of Finance 55(5): 2157-95.
MA
Persson, T., G. Roland and G. Tabellini (1997) "Separation of Powers and Political Accountability," Quarterly Journal of Economics 112: 1163-1202.
ED
Rogo¤, K. (1990) "Equilibrium Political Budget Cycles," American Economic Review 80(1): 21-36.
Shi, M. and Svensson, J. (2006) "Political Budget Cycles: Do They Di¤er Across Coun-
PT
tries and Why?" Journal of Public Economics 90(8-9): 1367-1389. Smart, M. and D. Sturm (2013) "Term Limits and Electoral Accountability," Journal of
CE
Public Economics 107: 93-102.
Stephenson, M. (2011) "Information Acquisition and Institutional Design," Harvard Law Review 124: 1422-1483.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Stillman, R.J. (1977) "The City Manager: Professional Helping Hand, or Political Hired Hand?" Public Administration Review 37(6): 659-670. Strausz, R. (1997) "Delegation of Monitoring in a Principal-Agent Relationship," Review of Economic Studies 64(3): 337-357. Stromberg, D. (2004) "Radios Impact of Public Spending," Quarterly Journal of Economics 119(1): 189-221. Stucky, T.D. (2005) "Local Politics and Police Strength," Justice Quarterly 22: 139-169. Vlaicu, R. and A. Whalley (2011) "Do Housing Bubbles Generate Fiscal Bubbles? Evidence from California Cities," Public Choice 149(1-2): 89-108. Whalley, A. (2013) "Elected Versus Appointed Policymakers: Evidence from City Treasurers," Journal of Law and Economics 56(1): 39-81.
37
0DQXVFULSW &OLFNKHUHWRYLHZOLQNHG5HIHUHQFHV
Online Appendix (Not for Publication) Section A.1 contains formal proofs of Propositions 1-2 in the text. Sections A.2-A.3 provide
T
theory robustness results. Section A.4 solves a selection model. Section A.5 contains a
RI P
discussion of supplementary empirical results. Section A.6 is the Data Appendix containing data definitions and sources, followed by supplementary Tables A1-A10.
SC
A.1 Incentives Model
NU
∞ Assumption 1 Given per-period i.i.d. preference shocks η jt t=1 such that η jt ∈ {B, C} with P η jt = C = γ j , agent j’s type at t + 1 is θjt+1 ∈ η jt , η jt+1 with P θjt+1 = η jt = λ. β[1+β(1−π)] 1−β 2
MA
Assumption 2 The discount factor β satisfies: Proposition 1 For popular policy issues π > P
1 2
> 1.
: (i) Hierarchical accountability either in-
sulates the policymaker pre-election, if γ > π, or creates policymaker pandering incentives
ED
pre-election, otherwise; (ii) Electoral accountability always creates policymaker pandering incentives pre-election; (iii) Agents follow their preferences post-election under both account-
PT
ability forms, e.g., an intermediary retains only policymakers of his own type. Proof of Proposition 1.
We first establish that voter beliefs about agent types are
CE
based on current period actions. To see this, note that at t + 1 incumbent j’s type may be based on either the current or a past preference shock: γ jt (Xt , Yt ) ≡ P θjt+1 = C| (Xt , Yt ) = P θjt+1 = C| (Xt , Yt ) , θjt+1 = η jt+1 (1 − λ)+P θjt+1 = C| (Xt , Yt ) , θjt+1 = θjt = η jt λ (1 − λ)+ P θjt+1 = C| (Xt , Yt ) , θjt+1 = η jt , θjt = η jt−1 λ2 . When θjt+1 = θjt = η jt , earlier observations
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
ACCEPTED MANUSCRIPT
(Xt−1 , Yt−1 ) are generated by preference types η jt−1 and earlier, which are independent of η jt . Thus, P θjt+1 = C| (Xt , Yt ) , θjt+1 = θjt = η jt = P θjt+1 = C| (xt , yt ) , θjt+1 = θjt = η jt . When θjt+1 6= θjt , the period t + 1 type is independent of period t type, since θjt+1 , θjt ∈ j η t+1 , η jt , η jt+1 , η jt−1 , η jt , η jt−1 . Thus, P θjt+1 = C| (Xt , Yt ) , θjt+1 6= θjt = γ j . Using these observations, it follows that γ jt (Xt , Yt ) = γˆ jt (xt , yt ) λ (1 − λ) + γ j [1 − λ (1 − λ)] . One can also write:
γ jt (Xt , Yt ) − γ j = λ (1 − λ) γˆ jt (xt , yt ) − γ j
(1)
Under electoral accountability, i.e., no intermediary and every other period it is the voter who decides whether to reelect the policymaker, the voter reelects the incumbent policymaker at the end of electoral term [t − 1, t] iff γˆ Pt (xt ) ≥ γ P . Thus, at t − 1 the incumbent follows his preferences. Suppose the equilibrium has separating election-period strategies: xt (st |C) 6= xt (st |B) for some state st . Then, either γˆ Pt (1) > γ P > γˆ Pt (0) i
ACCEPTED MANUSCRIPT or γˆ Pt (0) > γ P > γˆ Pt (1) , so the voter’s best response is to condition reelection on the election-period policy xt that improves the policymaker’s reputation. By Assumption IIa both types of policymaker prefer to override their preferences and comply at t, because the
RI P
T
resultant expected payoff is bounded below by the low type’s future expected payoff, namely P∞ 2τ −1 . Therefore, policymaker strategies in an election [1 + β (1 − π)] = β[1+β(1−π)] τ =1 β 1−β 2 period are pooling in both states: xt (st |C) = xt (st |B) for st = 0, 1. With pooling electionperiod strategies off-equilibrium path beliefs are pinned down by policymaker preferences,
SC
due to the assumption of a small fraction of non-strategic agents. Off-path, preference-based, policymaker equivalent reputations would be γˆ Pt (1) =
NU
γ P (1−π) . Thus pooling on xt = 0 cannot be an γ P (1−π)+(1−γ P )π P P γ t (0) = γ , which makes the probability of reelection
γP π γ P π+(1−γ P )(1−π)
> γ P > γˆ Pt (0) =
equilibrium because γ Pt (1) > γ P and
for xt = 1 no smaller than for xt = 0,
so each type would sometimes choose xt = 1 in election periods. The unique equilibrium then
MA
has election-period pooling on xt = 1, supported by equilibrium voter beliefs γ Pt (1) = γ P and γ Pt (0) < γ P .
Under hierarchical accountability the voter reelects the incumbent intermediary at the
ED
end of electoral term [t − 1, t] iff γˆ It (xt , yt ) ≥ γ I . Thus, at t − 1 the incumbent intermediary follows his preferences, namely retains only policymakers of his own type. Then,
PT
the policymaker also follows his preferences at t − 1. Suppose the equilibrium has separating election-period strategies: yt (xt |C) 6= yt (xt |B) for some policy xt . Then, either
CE
γˆ It (xt , 1) > γ I > γˆ It (xt , 0) or γˆ It (xt , 0) > γ I > γˆ It (xt , 1) , so given xt the voter’s best response is to condition reelection on the retention decision yt that improves the intermediary’s reputation. Overriding the preferred retention decision at t in order to get reelected is preferred
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
by both intermediary types because that yields positive future payoffs whereas following preferences yields zero. Thus, yt (xt |C) = yt (xt |B) for xt = 0, 1, and election-period strategies cannot be separating. With pooling election-period strategies off-equilibrium path beliefs are pinned down by agent preferences, due to the assumption of a small fraction of nonstrategic agents. Off-path, preference-based, intermediary equivalent reputations would be γ I (1−γ P )(1−π) γI γP π I γˆ It (1, 1) = γ I γ P π+(1−γ ˆ It (1, 0) = γ I (1−γ P )(1−π)+(1−γ I )γ P π (< γ I ); γˆ It (0, 1) = I )(1−γ P )(1−π) (> γ ); γ γ I (1−γ P )π γ I γ P (1−π) I ; γ ˆ (0, 0) = . Therefore, if xt = 1 off-path beliefs I P I P I P t γ γ (1−π)+(1−γ )(1−γ )π γ (1−γ )π+(1−γ I )γ P (1−π) can only support intermediary pooling on retaining the policymaker. If xt = 0 the form of the equilibrium strategies depends on whether γ P (1−π) ≶ 1 − γ P π, or γ P ≶ π. When the left-hand side is larger, then γˆ It (0, 1) > γ I > γˆ It (0, 0) and so the resulting off-path beliefs can
only support intermediary pooling on retaining the policymaker. Otherwise, the resulting off-path beliefs can only support intermediary pooling on replacing the policymaker. The
ii
ACCEPTED MANUSCRIPT unique equilibrium can thus take two forms: (i) in election period t the intermediary retains regardless of xt ; or (ii) in election period t the intermediary retains iff xt = 1. In the first type of equilibrium the policymaker follows his preferences in an election period; in the second
T
type of equilibrium, by the argument in the previous paragraph and Assumption IIa, both
Proposition 2 For neutral policy issues π =
1 2
RI P
policymaker types choose the popular policy in an election period.
policymaking outcomes do not vary with
SC
the accountability form in election periods or non-election periods. Proof of Proposition 2. The claim follows by plugging π =
1 2
into the voter’s equilibrium
beliefs in the Proof of Proposition 1. Under electoral accountability election-period policy xt
NU
does not signal policymaker type θPt since γˆ Pt (1) = γˆ Pt (0) = γ P . Assuming that when indifferent between the incumbent and the challenger the voter reelects the incumbent, the voter
MA
reelects the incumbent policymaker regardless of policy; alternatively one could assume the voter plays the equilibrium with the highest voter welfare, which is the insulating equilibrium since γ P + 1 − γ P 2π (1 − π) > π. This insulates the policymaker in an election period, so
ED
xt (st |C) = st , and xt (st |B) = 1 with probability (1 − π) and xt (st |B) = 0 with probability
π. Under hierarchical accountability the retention signal is informative, ruling out separatγP >
1 2
PT
ing election-period intermediary strategies. Since only the case γ P > π is now possible, as = π, based on Proposition 1 only the isulating equilibrium can be supported, thus
CE
the unique pooling equilibrium features policymaker election-period insulation.
A.2 Theory Robustness I: Incentive Dynamics
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
In this section we explore the robustness of the incentives model results in Appendix A.1 by relaxing the constraint implicit in Assumption 1 that reelection incentives affect only electionperiod behavior (as in Rogoff 1990, Shi and Svensson 2006). Here we use the following assumption that allows reelection incentives to affect both election and non-election period behavior (as in Martinez 2009). ∞ Assumption 1(I) Given per-electoral-term i.i.d. preference shocks η j2τ −1 τ =1 such that η j2τ −1 ∈ {B, C} with P η j2τ −1 = C = γ j , agent j’s type at [t + 1, t + 2] is θjt+1 = θjt+2 ∈ j η t−1 , η jt+1 with P θjt+2 , θjt+1 = η jt−1 = λ. Assumption 2(I) The discount factor β satisfies:
β(1−π) 1−β 1 2
> 1.
: (i) Hierarchical accountability ei ther insulates the policymaker pre-election, if γ (1 − π)2 > 1 − γ P π 2 , or creates policy-
Proposition 1(I) For popular policy issues
π> P
iii
ACCEPTED MANUSCRIPT maker pandering incentives pre-election with probability γ I (1 − π) + 1 − γ I π, otherwise;
an intermediary retains only a policymaker of his own type.
T
(ii) Electoral accountability creates policymaker pandering incentives pre-election with proba bility γ P (1−π)+ 1 − γ P π; (iii) Agents follow their preferences post-election; in particular,
RI P
Proof of Proposition 1(I). For electoral accountability the voter reelects the incumbent
SC
policymaker at the end of electoral term [t − 1, t] iff γˆ Pt (xt−1 , xt ) ≥ γ P . To see this, note γ jt (Xt , Yt ) ≡ P θjt+1 = C| (Xt , Yt ) = P θjt+1 = C| (Xt , Yt ) , θjt+1 = η jt+1 (1 − λ) + P{θjt+1 = C| (Xt , Yt ) , θjt+1 , θjt = η jt−1 }λ (1 − λ) + P θjt+1 = C| (Xt , Yt ) , θjt+1 , θjt = η jt−1 , η jt−3 λ2 . If
MA
NU
θjt+2 , θjt+1 = θjt , θjt−1 = η jt−1 , earlier observations (Xt−2 , Yt−2 ) are generated by preference types η jt−3 and earlier, which are independent of η jt−1 . Thus, P θjt+1 = C| (Xt , Yt ) , θjt+1 , θjt = η jt−1 = P θjt+1 = C| (xt−1 , yt−1 , xt , yt ) , θjt+1 = θjt = η jt−1 . When θjt+1 = 6 θjt , the period t+1 type is in dependent of period t type, since θjt+1 , θjt ∈ η jt+1 , η jt−1 , η jt+1 , η jt−3 , η jt−1 , η jt−3 . Thus, P θjt+1 = C| (Xt , Yt ) , θjt+1 6= θjt = γ j . Using these observations, it follows that γ jt (Xt , Yt ) = γˆ jt (xt−1 , yt−1 , xt , yt ) λ (1 − λ) + γ j [1 − λ (1 − λ)] . One can also write:
ED
γ jt (Xt , Yt ) − γ j = λ (1 − λ) γˆ jt (xt−1 , yt−1 , xt , yt ) − γ j
(2)
PT
Suppose in equilibrium policymaker types play separating strategies. Then, because π > 21 , there is one pair of policies (xt−1 , xt ) that is chosen more often by the biased type. The
CE
voter’s best response is to not reelect when observing that policy pair. In a state pair (st−1 , st ) where the biased type prefers that policy pair, forgoing it at t yields at least P∞ τ β(1−π) τ =1 β (1 − π) = 1−β , whereas forgoing reelection yields 1 at t. Thus, by Assumption
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
2(I), policymaker strategies have to be pooling. In a pooling equilibrium off-equilibrium path voter beliefs are pinned down by policymaker preferences, due to the assumption of a small fraction of non-strategic agents. Then, γˆ Pt (1, 1) =
γˆ Pt (0, 1) = γ P ; γˆ Pt (0, 0) =
γ P (1−π)2 γ P (1−π)2 +(1−γ P )π 2
γ P π2 (> γ P π 2 +(1−γ P )(1−π)2
γ P ); γˆ Pt (1, 0) =
(< γ P ). Assuming that when indifferent be-
tween the incumbent and the challenger the voter reelects the incumbent, the voter reelects when at least one of the last two observed policies is popular. A reelection-seeking policymaker will then choose the popular policy at least once. When st−1 is such that the incumbent prefers the popular policy, he chooses xt−1 = 1 and then xt according to his period t preferences. When st−1 is such that the incumbent prefers the unpopular policy, choosing xt−1 = 1 and then xt according to period t preferences yields β during the term whereas choosing xt−1 according to period t − 1 preferences and then xt = 1 yields 1 + βπ for a congruent type and 1 + β (1 − π) for a biased type. Thus, for both types it is optimal to comply later in the term. Pandering in election period t occurs when the popular iv
ACCEPTED MANUSCRIPT policy went against the policymaker’s period t − 1 preferences, that is with probability γ P (1 − π) + 1 − γ P π. By Assumption 2(I) at t both policymaker types prefer to over-
RI P
T
ride their election-period preferences and comply with the voter’s reelection rule, because P 2τ −1 it yields ∞ {π (1 + β) + (1 − π) (1 + βπ)} = β[1+βπ(2−π)] for a congruent type and τ =1 β 1−β 2 P∞ 2τ −1 β [1+β (1−π 2 )] for a biased type. {(1 − π) (1 + β) + π [1 + β (1 − π)]} = τ =1 β 1−β 2 Under hierarchical accountability, the voter reelects the incumbent intermediary at the
end of electoral term [t − 1, t] iff γˆ It [(xt−1 , yt−1 ), (xt , yt )] ≥ γ I . See equation (2). >
1 , 2
Be-
given (xt−1 , xt ), there is one pair of retention decisions (yt−1 , yt ) that is
SC
cause γ
P
chosen more often by the biased type, which makes the voter not reelect when observ-
NU
ing that pair of intermediary reactions. But then the biased type would not choose that course of action in the first place because pursuing reelection yields a positive expected
MA
payoff. Thus, intermediary strategies have to be pooling. In a pooling equilibrium offequilibrium path voter beliefs are pinned down by intermediary preferences, due to the assumption of a small fraction of non-strategic agents. Then, after two popular policies γ I γ P π2
γ I γ P π 2 +(1−γ I )(1−γ P )(1−π)2
ED
γˆ It [(1, 1), (1, 1)] = 2
γ I (1−γ P ) (1−π)2
(> γ I ); γˆ It [(1, 0), (1, 1)] = γ I ; γˆ It [(1, 0), (1, 0)] =
(< γ I ); one popular and one unpopular policy γ I (1−γ P )2 (1−π)2 +(1−γ I )(γ P )2 π 2 γ I γ P π(1−π) I ˆ It [(0, 0), (1, 0)] = γˆ It [(1, 1), (0, 1)] = γ I γ P π(1−π)+(1−γ I )(1−γ P )(1−π)π (> γ ); γ
γˆ It [(0, 1), (1, 1)] =
CE
PT
γˆ It [(1, 0), (0, 0)] = 2 γ I (1−γ P )γ P π 2 γ I (1−γ P ) (1−π)π I I (> (< γ ); γ ˆ [(0, 0), (1, 1)] = 2 2 t γ I (1−γ P )γ P π 2 +(1−γ I )γ P (1−γ P )(1−π)2 γ I (1−γ P ) (1−π)π+(1−γ I )(γ P ) π(1−π) 2 I P P γ (1−γ )γ (1−π) γ I ); γˆ It [(1, 0), (0, 1)] = γ I (1−γ P )γ P (1−π)2 +(1−γ I )γ P (1−γ P )π2 (< γ I ); after two unpopular policies γ I γ P (1−π)2 γˆ It [(0, 1), (0, 1)] = γ I γ P (1−π) (> γ I iff γ P (1 − π)2 > 1 − γ P π 2 iff γ P > 2 +(1−γ I )(1−γ P )π 2
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
p); γˆ It [(0, 0), (0, 1)] = γ I ; γˆ It [(0, 0), (0, 0)] =
2
γ I (1−γ P ) π 2
γ I (1−γ P )2 π 2 +(1−γ I )(γ P )2 (1−π)2
(> γ I iff γ P < p),
and we set γˆ It [(xt−1 , 1), (xt , 0)] = γˆ It [(xt−1 , 0), (xt , 1)] to preserve symmetry, even though [(xt−1 , 1), (xt , 0)] are not consistent with the preferences of either type of intermediary. Assuming that when indifferent between the incumbent and the challenger the voter reelects the incumbent, the voter reelects (i) after two popular policies, iff the intermediary retains at least once; (ii) after one popular and one unpopular policy, iff the intermediary retains after the popular policy, and (iii) after two unpopular policies, iff the intermediary retains at least once, when γ P > p, regardless of the intermediary’s reaction, when γ P (1 − π)2 > 1 − γ P π 2 , or iff the intermediary replaces at least once, when γ P (1 − π)2 < 1 − γ P π 2 . It remains to
show that the intermediary prefers to comply with the voter’s flexible reelection rule at the
end of the term rather than at the beginning of the term. When γ P (1 − π)2 > 1 − γ P π 2 the voter’s strategy is to reelect the intermediary if there
is at least one retention and, in the case of only one popular policy, the retention has to follow v
ACCEPTED MANUSCRIPT the popular policy. If the intermediary faces a policymaker of the same type, retaining in both periods is both consistent with intermediary preferences and guarantees reelection. If however the intermediary faces a policymaker of the opposite type, he could either retain at
T
the beginning of the term or at the end. Retaining at t − 1 and replacing at t produces a zero
RI P
payoff at t, because the policymaker will follow his preferences at t; replacing at t − 1 and retaining at t produces a positive expected payoff at t. Thus, the intermediary always retains at the end of the term. When γ P (1 − π)2 < 1 − γ P π 2 the voter’s strategy is to reelect the
SC
intermediary if there is at least one retention after popular or one replacement after unpopu-
lar, and in the case of only one popular policy, the retention has to follow the popular policy.
NU
If the intermediary faces a popular same type policymaker or an unpopular opposite type policymaker, acting on preferences in both periods results in reelection. In the opposite cases,
MA
the intermediary can either replace an unpopular same type policymaker at t−1 yielding γ P + β π + βγ P P θPt+1 = θIt+1 |θPt = θIt = C + β 1 − γ P P θPt+1 = θIt+1 +Vt+3 and 1 − γ P + β (1 − π) + βγ P P θPt+1 = θIt+1 + β 1 − γ P P θPt+1 = θIt+1 |θPt = θIt = B +Vt+3 for a con-
ED
gruent and biased type respectively, or retain at t − 1 and retain again at t iff he is pop ular yielding π + β P θPt+1 = θIt+1 |θPt = θIt = C + βπ + Vt+3 for a congruent type and (1 − π) + β P θPt+1 = θIt+1 |θPt = θIt = B + β (1 − π) + Vt+3 for a biased type; the inter-
PT
mediary can either retain a popular opposite type policymaker at t − 1 and replace at t,
yielding zero at t, or replace at t − 1 and retain at t iff the incoming policymaker is pop-
CE
ular, yielding a positive expected payoff at t. If the type persistence probability λ (1 − λ) is large enough, the strategy of complying with the voter’s reelection rule at the end of the term has a higher payoff for both intermediary types. The frequency of election-period pan dering then is γ I γ P (1 − π) + 1 − γ P (1 − π) + 1 − γ I γ P π + 1 − γ P π which equals γ I (1 − π) + 1 − γ I π.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Proposition 2(I) For neutral policy issues π =
1 2
policymaking outcomes do not vary with
the accountability form in an election or a non-election period. Proof of Proposition 2(I). The claim follows by plugging π =
1 2
into the voter’s equilib-
rium beliefs in the Proof of Proposition 1(I). Under electoral accountability policies (xt−1 , xt ) do not signal type θPt−1 . Assuming that when indifferent between the incumbent and the challenger the voter reelects the incumbent, the voter reelects the incumbent policymaker regardless of policies chosen in the electoral term [t − 1, t]; alternatively, one could assume that the voter plays the equilibrium with the highest voter welfare, which in this case is the insulating equilibrium, since (1 + β) γ P + 1 − γ P 2π (1 − π) > γ P [π (1 + β) + (1 − π) (1 + βπ)] + 1 − γ P π (1 − π + βπ). This insulates the policymaker throughout the electoral term. Unvi
ACCEPTED MANUSCRIPT der hierarchical accountability the retention signal is still informative, since γ P > 21 , ruling out separating equilibria. Since only the case γ P (1 − π)2 > 1 − γ P π 2 is now possible,
based on Proposition 1(I) the unique pooling equilibrium is policymaker insulation in both
RI P
T
halves of an electoral term.
A.3 Theory Robustness II: Variable Pandering Intensity
SC
Assumption 1(II) The process for types ( θPt , θIt ) is the same as in the Incentives Model in the main text and Online Appendix A.1. The policy payoffs of the congruent and biased types are as in the Incentives Model in the main text and Online Appendix A.1, with the
NU
modification that the preferred policy gives, instead of a payoff of 1, a payoff equal to a random fraction ϕ ˜ of an agent’s value of office Vθj , and that random fraction is uniformly
MA
distributed: ϕ ˜ ∼ U [0, 1] .
Assumption 2(II) The probability γ˜ j that the challenger is a congruent type is uniformly
ED
distributed: γ˜ j ∼ U [0, 1] .
Proposition 1(II) The claims of Propositions 1 and 2 continue to hold. Reelection prob-
PT
abilities satisfy 0 < rt (0) < rt (1) < 1. Moreover, when pre-election pandering occurs in equilibrium, the pandering probability pt is equal to the discounted reelection probability pre-
CE
mium, pt = β [rt (1) − rt (0)] , and is monotonically increasing in π, from pt = 0 when π = 12 , to pt < 1 when π → 1.
Proof of Proposition 1(II). Under electoral accountability, the voter reelects the in-
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
cumbent policymaker if and only if γ Pt (Xt ) ≥ γ˜ P , which becomes, using equation (1), λ (1 − λ) γˆ Pt (xt ) − γ P ≥ γ˜ P − γ P . Since γ˜ P ∼ U [0, 1] , reelection probabilities are rt (xt ) = γ P + λ (1 − λ) γˆ Pt (xt ) − γ P , for xt = 0, 1. We first show the existence of a pandering equilibrium on policy xt = 1 and then rule out other equilibria.
In an equilibrium with pandering on the popular policy xt = 1 the voter’s equilibrium beliefs are determined by γˆ Pt (1) = γ P (1−π) γ P (1−π)+(1−γ P )π
γ P [pt +(1−pt )π] γ P [pt +(1−pt )π]+(1−γ P )[pt +(1−pt )(1−π)]
> γ P > γˆ Pt (0) =
and equation (1). Thus, rt (1) > rt (0) . An incumbent policymaker faces a
tradeoff when the current preferred policy is not the most rewarded electorally, i.e., it is not policy xt = 1. In these cases he will override his current policy preference and pick the popular policy xt = 1 if and only if rt (1) βVθP ≥ ϕ ˜ VθP + rt (0) βVθP . Since ϕ ˜ ∼ U [0, 1] the incumbent policymaker picks policy xt = 1 with probability pt = β [rt (1) − rt (0)] . Using the derived expression for rt (xt ) , this can be rewritten as pt = βλ (1 − λ) γˆ Pt (1) − γˆ Pt (0) . vii
ACCEPTED MANUSCRIPT This last equation, together with the expressions for γˆ Pt (1) , γˆ Pt (0) above, give an implicit function in pt and π. Using the implicit function theorem: ∂ P ∂ P γˆ t (1) − ∂π γˆ t (0) βλ (1 − λ) ∂π ∂pt = ∂π 1 − βλ (1 − λ) ∂p∂ t γˆ Pt (1) where it can be easily verified that
∂ P γˆ ∂π t
∂pt ∂π
then γˆ Pt (1) =
> 0. Note also that when π =
1 2
RI P
T
(3)
∂ P γˆ (0) and ∂p∂ t γˆ Pt (1) < ∂π t γˆ Pt (0) = γ P and so pt = 0.
(1) > 0 >
0. Therefore,
SC
We now show that equilibria where rt (1) ≤ rt (0) can be ruled out. In the case rt (1) = rt (0) the incumbent policymaker acts on his current policy preference, implying that γˆ Pt (1) = (0) =
γ P (1−π) . γ P (1−π)+(1−γ P )π
But since γˆ Pt (1) > γ P > γˆ Pt (0) then, rt (1) > γ P >
NU
γP π , γˆ Pt γ P π+(1−γ P )(1−π)
rt (0) , a contradiction. In the case rt (1) < rt (0) , both incumbent types sometimes pander γP π , γˆ Pt γ P π+(1−γ P )(1−π)
(0) =
MA
on xt = 0. This implies γˆ Pt (1) =
γ P [pt +(1−pt )(1−π)] . γ P [pt +(1−pt )(1−π)]+(1−γ P )[pt +(1−pt )π]
Then, since γˆ Pt (1) > γ P > γˆ Pt (0) again rt (1) > γ P > rt (0) , a contradiction. Thus the unique equilibrium is the one that features pandering on the popular policy xt = 1.
ED
Under hierarchical accountability the equilibrium can be either insulation or pandering, depending on whether γ P > π. The proof closely follows the structure of the Proof for
PT
Proposition 1. In a pandering equilibrium, the intermediary plays the role of the voter in electoral accountability, so the policymaker’s pandering probability has the same properties
CE
as the ones derived above.
A.4 Selection Model
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Assumption 1(III) The process for types and the payoff structure are as in the Incentives Model in the main text and Online Appendix A.1. Assumption 2(III) The discount factor β satisfies: β = 0. Proposition 1(III) For popular policy issues π >
1 2
: (i) Hierarchical accountability per-
forms policymaker selection through intermediary hselection. The increase i in intermediary γI γP I I I congruence post-election is γ t+1 − γ = λ (1 − λ) γ I γ P +(1−γ I )(1−γ P ) − γ , when γ P > π, h i γI π I and γ It+1 − γ I = λ (1 − λ) γ I π+(1−γ − γ , when γ P < π. The increase in policymaker I )(1−π) h i γP γI P P P congruence is γ t+1 − γ = λ (1 − λ) γ P γ I +(1−γ P )(1−γ I ) − γ post-election; pre-election it is γ P γ It+1 P P P γ t+2 −γ = λ (1 − λ) γ P γ I +(1−γ P ) 1−γ I − γ ; (ii) Electoral accountability performs pol( t+1 ) t+1 icymaker selection post-election only. The increase in policymaker congruence post-election h i is γ Pt+1 − γ P = λ (1 − λ)
γP π γ P π+(1−γ P )(1−π)
− γP . viii
ACCEPTED MANUSCRIPT Proof of Proposition 1(III). By Assumption 2(III) both policymaker types as well as intermediary types follow their preferences rather than comply with a reelection or retention rule. As a consequence, their reputations rise and fall according to equation (1).
T
Under electoral accountability the voter reelects only policymakers that chose the pop-
RI P
ular policy in the most recent election period, by Proposition 1. Then, γ Pt+1 − γ P = i h γP π P . Under hierarchical accountabilλ (1 − λ) γˆ Pt (1) − γ P = λ (1 − λ) γ P π+(1−γ P )(1−π) − γ
ity the voter reelects only intermediaries that have retained the policymaker in the election
SC
period, if γ P > π, and reelects only intermediaries that have retained a popular policymaker and replaced an unpopular policymaker in the election period, if γ P < π. Intermediary selec-
MA
NU
tion happens after every election period. Then, for γ P > π a reelected reputai h intermediary’s I γI γP I I I tion improved by γ t+1 − γ = λ (1 − λ) γˆ t (xt , 1) − γ = λ (1 − λ) γ I γ P +(1−γ I )(1−γ P ) − γ I , γ I [γ P π+(1−γ P )π ] P I I I and for γ < π by γ t+1 − γ = λ (1 − λ) γ I [γ P π+(1−γ P )π]+γ I [γ P (1−π)+(1−γ P )(1−π)] − γ = h i γI π I λ (1 − λ) γ I π+(1−γ . Policymaker selection occurs every period as a result of the I )(1−π) − γ
ED
intermediary’s retention decision. Post-election the reputation goes up by h policymaker’s i P I γ γ P γ Pt+1 − γ P = λ (1 − λ) γˆ Pt (xt , 1) − γ P = λ (1 − λ) γ P γ I +(1−γ . The expression P )(1−γ I ) − γ P P P P for pre-election selection γ t+2 − γ = λ (1 − λ) γˆ t (xt+1 , 1) − γ follows by replacing γ I
PT
with γ It+1 . Thus γ Pt+1 > γ Pt+2 = γ P under electoral selection while γ Pt+1 < γ Pt+2 under hierarchical selection, since γ It+1 > γ I . Because π >
1 2
a congruent policymaker is more likely
CE
to choose the popular policy, creating a positive election-period popular policy differential between hierarchical and electoral accountability. Proposition 2(III) For neutral policy issues π >
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
1 2
: (i) Hierarchical accountability per-
forms policymaker selection through intermediaryhselection. The increase i in intermediary γI γP I I I congruence post-election is γ t+1 − γ = λ (1 − λ) γ I γ P +(1−γ I )(1−γ P ) − γ . The increase in h i γP γI P policymaker congruence is γ Pt+1 − γ P = λ (1 − λ) γ P γ I +(1−γ − γ post-election; preP )(1−γ I ) γP γI election it is γ Pt+2 − γ P = λ (1 − λ) γ P γ I +(1−γt+1 − γ P ; (ii) Electoral accountability P ) 1−γ I ( ) t+1 t+1 performs no policymaker selection pre-election or post-election, γ Pt+2 − γ P = γ Pt+1 − γ P = 0. Proof of Proposition 2(III). For electoral accountability the result follows by plugging in π=
1 2
into the electoral selection formula of Proposition 1(III). For hierarchical accountabilityi h
only the case γ P > π is now possible. Then γ It+1 − γ I = λ (1 − λ)
γI γP γ I γ P +(1−γ I )(1−γ P )
and the formulas for policymaker selection are those derived for Proposition 1(III).
ix
− γI
ACCEPTED MANUSCRIPT A.5 Supplementary Empirical Results Robustness. Online Appendix Table A7 reports sensitivity checks for our preferred IV specification and two related OLS specifications. Panel A checks the sensitivity of our results
T
to minor sample alterations: excluding extremely large/small cities, excluding the pre-1972
RI P
period, which was characterized by racial tensions, and the post-1994 period, when the federal government intervened more forcefully in local police employment through COPS
SC
grants, and dropping dependent variable outliers and cities far away from weather stations. Overall, the OLS and the IV results maintain their prior patterns. Online Appendix Table A7 Panel B varies the inference procedure. We examine three
NU
sets of alternative standard errors: clustering on both city and year, clustering on weather station (as this is the unit of observation for the weather variables), and accounting for
MA
spatial correlation (Conley 1999).1 Inferences from both the OLS and the IV estimates remain unchanged.
Thus far we have scaled the number of police employees by population, however a more
ED
relevant measure may be the crime level since police should be proportional to the extent of crime. A high officer-crime ratio may also better capture deviations from policy optimality,
PT
by providing a measure of "excess police." Online Appendix Table A7 Panel C presents estimates with employment scaled by two different crime measures: reported crime and
CE
cleared crime. In both OLS and IV specifications manager cities have significantly fewer officers per crime than mayor cities. Extreme Precipitation and Flooding. As the posited key determinant of city govern-
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
ment form is flood-related infrastructure crises, the ideal instrument would be based on the extent of flood damage in the local flood control era, 1900-1936. This data, however, is not available, not in least part because flood control in this period was handled by local authoritites. The next best approach is to produce an indirect measure of flooding, based on the much more readily available precipitation data. The existing literature on the link between extreme precipitation and flooding, however, has not reached a consensus on this question (Bedient, Huber, and Vieux 2008). To nevertheless validate our precipitation-based instrument we adopt two approaches. First, in Online Appendix Table A8 we use a cross-sectional measure of flooding, based 1
To implement Conley standard errors we allow for a spatial dependence of up to 0.5 degrees latitude and longitude, which corresponds to about 65 miles, quite close to the 58 miles mean distance of a city from the nearest weather station. We do not include year effects in this model as the unbalanced nature of the panel impedes the estimation of the Conley procedure. This slight change accounts for the difference between the point estimates in row (9) and previous ones.
x
ACCEPTED MANUSCRIPT on soil analysis by the USDA, to study correlations with several versions of our precipitationbased instrument. We find that the history of flooding in a location (revealed by the presence of flood-specific sediment deposited in the soil) is clearly related to the history of annual
T
precipitation shocks, variously defined, in the full cross section of cities. Overall, the top-1%
RI P
precipitation cutoff seems to perform best.
Second, in Online Appendix Table A9 we use a time-varying measure of flooding, based on USGS riverflow readings in locations throughout the U.S. This variable is not available
SC
for our full sample. However, we use it to study precipitation-flooding correlations, first by imputing missing observations as no flood, and second, by restricting the sample to
NU
reported observations. The imputation is based on the assumption that reported readings generally reflect higher-than-usual riverflow values. The correlations are positive for all methods of time aggregation, annual, quarterly, monthly, and 7-day max, and are more
MA
precisely estimated when using data at higher frequencies, e.g., quarterly, monthly. In the IV analysis in Online Appendix Table A5, the IV estimates based on quarterly or monthly
ED
LFC precipitation shocks are less precise, however (first stage F-statistics 5.03 and 6.38, compared to 28.40).
Putting these considerations in the balance, we choose the instrument based on annual
PT
aggregation for our preferred IV specification because it has (i) considerable validity in
CE
predicting flooding and (ii) a stronger first stage than the alternatives.
A.6 Data Appendix i) Police Protection
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Police Officers per Capita: Number of sworn police officers, per 1,000 residents. Sources: Uniform Crime Reports (1960-2000). Police Civilians per Capita: Number of civilian (non-sworn) police employees, per 1,000 residents. Sources: Uniform Crime Reports (1960-2000). Police Department Spending per Capita: Total police department expenditure, per resident, in 2000$. Sources: Census of Governments, City Government Finances (19602000). Fraction Officers Unionized 1968: Fraction of the police department unionized in 1968. Source: Municipal Year Book (1969). Crime Rate: Sum of violent and property crime rates. Violent crime rate is total number of murder, rape, and robbery crimes reported, per 1,000 residents. Property crime rate is
xi
ACCEPTED MANUSCRIPT total number of motor vehicle theft, larceny, burglary, and assault crimes reported, per 1,000 residents. Sources: Uniform Crime Reports (1975-2000). Fraction Crimes Cleared: Ratio of total offences cleared through arrest to total crimes
RI P
T
reported. Sources: Uniform Crime Reports (1975-2000). ii) City Characteristics in 1900 Sources: U.S. Census Bureau (1905, 1906).
SC
Police Officers per Capita: Number of sworn police officers, per 1,000 residents. Police Civilians per Capita: Number of civilian (non-sworn) employees, per 1,000 resi-
NU
dents.
Arrests per Capita: Number of police department arrests, per 1,000 residents.
MA
Sewers per Square Mile: Miles of sewers per square mile. Roads per Square Mile: Miles of paved roads per square mile.
ED
Population: Number of city residents. iii) City Government
PT
Manager: Dummy variable equal to 1 if the city has a manager form of government, and 0 otherwise. Sources: Municipal Year Book (1960-2000).
CE
Manager Name: First and last name of the city manager in manager cities. Sources: World Almanac (1960-2000), ICMA, city clerk offices. Election: Indicator variable equal to 1 in an election year, and 0 otherwise. If the election
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
takes place before July 31 the previous year is coded as an election year. Election years are coded based on mayor elections and in case no mayor office exists in a manager city, based council elections. Sources: Municipal Year Book (various years), World Almanac and Book of Facts (various years), www.ourcampaigns.com, city charters, newspaper articles. Non-Partisan Elections: Indicator variable equal to 1 if the city charter in effect in 1960 mandates non-partisan elections, and 0 otherwise. Source: Municipal Year Book (1960). Fraction At-Large Seats: Fraction of city council seats elected at-large in 1960. Source: Municipal Year Book (1960). Early Civil Service: Indicator variable equal to 1 if the city has a non-political civil service before 1937, and 0 otherwise. Source: Civil Service Assembly (1938). Incumbent Mayor Republican: Takes a value of 1 if the incumbent mayor’s party is Republican, and 0 otherwise. Most manager cities maintain a separate and honorary office of the mayor. Sources: Ferreira and Gyourko (2009). xii
ACCEPTED MANUSCRIPT Local Newspaper Sales: Average daily sales of local newspapers, per capita, 1990-2000. Source: George and Waldfogel (2006). City Government Spending Per Capita: Total city government expenditure, in thou-
T
sands of 2000$, per capita. Sources: U.S. Census Bureau, Census of Governments, City
RI P
Government Finances (1960-2000).
Non-Police Employees Per Capita: Total city government employees minus total police protection employees, per 1,000 residents. Sources: U.S. Census Bureau, Census of
SC
Governments (1972-2000).
NU
iv) Demographics
Population: Number of city residents, in thousands. Based on 1960, 1970, 1980, 1990, and
MA
2000 Census of Population numbers, linearly interpolated in intercensal years. Sources: U.S. Census Bureau, Census of Population (various years). Population Density: Number of city residents, in thousands, divided by square miles
ED
of city area. Population based on 1960, 1970, 1980, 1990, and 2000 Census of Population numbers, linearly interpolated in intercensal years. City area based on 1900 and 2000 Census
PT
numbers, linearly interpolated between 1900 and 2000. Sources: Authors’ calculations. Fraction Non-White: Fraction of city population that are non-white. Based on 1960, 1970, 1980, 1990, and 2000 Census of Population numbers, linearly interpolated in intercensal
CE
years. Sources: U.S. Census Bureau, Census of Population (various years). Fraction College Graduate: Fraction of city population that are college graduates. Based on 1960, 1970, 1980, 1990, and 2000 Census of Population numbers, linearly interpolated in
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
intercensal years. Sources: U.S. Census Bureau, Census of Population (various years). Household Income: Median household income of city residents, in 2000$. Based on 1960, 1970, 1980, 1990, and 2000 Census of Population numbers, linearly interpolated in intercensal years. Sources: U.S. Census Bureau, Census of Population (various years). v) Weather Sources: The U.S. Historical Climatology Network’s Daily Temperature, Precipitation, and Snow Data contains daily readings for precipitation, snowfall, and temperature extremes collected from weather stations throughout the U.S. We construct yearly variables based on this dataset. LFC Precipitation Shocks: Fraction of years in the Local Flood Control era (1900-1936) with annual precipitation in the top 1 percent of the national precipitation distribution. xiii
ACCEPTED MANUSCRIPT FFC Precipitation Shocks: Fraction of years in the Federal Flood Control era (1937-1960) with annual precipitation in the top 1 percent of the national precipitation distribution. Century Precipitation Shocks: Fraction of twentieth century (1900-2000) years with
T
annual precipitation in the top 1 percent of the national precipitation distribution.
RI P
Median Precipitation: Median annual precipitation from 1900-2000. Quarterly Precipitation Shocks: Fraction of years in the relevant era (LFC or Century) with quarterly precipitation in the top 1 percent of the national precipitation distribution.
SC
Monthly Precipitation Shocks: Fraction of years in the relevant era (LFC or Century) with monthly precipitation in the top 1 percent of the national precipitation distribution.
NU
7-Day Max Precipitation Shocks: Fraction of years in the relevant era (LFC or Century) with the annual maximum cumulative 7-day precipitation in the top 1 percent of the national
MA
precipitation distribution. vi) Water
ED
Sources: Fishback, Horrace, and Kantor (2005, 2006). The data are reported at the county level. We match it to our sample cities.
PT
Lakes: Number of lakes in the county.
Swamps: Number of swamps in the county. Coast: Dummy variable that indicates if county is located on the coast. to 20 counties.
CE
Small Rivers: Dummy variable that indicates if county has a river that goes through 11 Medium Rivers: Dummy variable that indicates if county has a river that goes through
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
21 to 50 counties.
Large Rivers: Dummy variable that indicates if county has a river that goes through more than 50 counties. vii) Flooding Soil-Based Flooding: Standardized average flood class for the county, based on contemporary surveys by U.S. Department of Agriculture soil scientists. Soil surveys measure the frequency of flood occurrence in locations across the United States. The surveys collect a soil profile for each location. When a flood occurs at a location a thin stratum of gravel, sand, silt, or clay is deposited by the floodwater. These strata in the soil profile produce estimates of flood frequency at the location. Flood classes: (1) Frequent floods (> 50 times per 100 years), (2) Occasional floods (5 to 50 times per 100 years), (3) Rare floods (1-5 times per xiv
ACCEPTED MANUSCRIPT 100 years) or (4) No measurable flood activity. We standardize the county-level averages to have mean zero and standard deviation one. Sources: Fishback, Horrace, and Kantor (2005, 2006).
T
River-Based Flooding: Peak streamflow, in cubic feet per second (cfs), captures the
RI P
maximum instantaneous discharge of a stream or river at a given location. Usually occurs at or near the time of maximum stage, i.e. when water levels are high enough to cause a flood. Based on monitoring station nearest to city downtown or the river. When missing
SC
data is imputed as no flood occurence, the variable is turned into a dummy that takes the value one in each year with a positive reading. Sources: U.S. Geological Service National
NU
Water Information System.
MA
viii) Mechanisms
Crack Epidemic: Dummy variable equal to 1 if crack cocaine arrived in the state in a prior year, based on cocaine-related deaths, and 0 otherwise. Sources:
CE
PT
ED
Moore (2012).
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
xv
Evans, Garthwaite, and
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 Å
Å Å
Å
Å Å
Å Å
Å
Å
Å
Å Å
MA
Å Å
Å
ED
Å Å
Å
Å
Å Å
Å
Å /,3Å /,3Å /,3Å /,3Å Å Å Å Å ÅLOG0OLICEÅ#IVILIANSÅ0ERÅ#AP Å Å ÅÅ Å Å Å Å nÅÅ nÅÅ nÅ Å nÅ Å Å Å Å ÅÅ nÅ Å nÅ Å Å Å Å Å Å Å ÅÅ 9ESÅ 9ESÅ 9ESÅ 9ES 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ 9ESÅ .OÅ 9ESÅ
PT
Å
CE
AC
/,3Å /,3Å /,3Å /,3Å Å Å Å Å ÅLOG0OLICEÅ/FFICERSÅPERÅ#AP Å Å ÅÅ Å Å Å Å nÅÅ nÅÅ nÅ Å nÅ Å Å Å Å ÅÅ nÅ Å nÅ Å Å Å Å Å Å Å ÅÅ 9ESÅ 9ESÅ 9ESÅ 9ES 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ 9ESÅ .OÅ 9ESÅ Å Å
Å
Å Å
Å
Å Å
Å
Å Å
Å
/,3Å /,3Å /,3Å /,3Å Å Å Å Å Å Å Å LOG.ON 0OLICEÅ%MPLÅ0ERÅ#APÅ Å Å Å Å Å nÅÅ nÅÅ nÅ Å nÅ Å Å Å Å ÅÅ nÅ Å nÅ Å Å Å Å Å Å Å ÅÅ 9ESÅ 9ESÅ 9ESÅ 9ES 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ 9ESÅ .OÅ 9ESÅ
XVIÅ
RI P
SC
NU
T
.OTESÅ!UTHORSjÅCALCULATIONSÅWITHÅ ÅCITYÅDATAÅASÅDESCRIBEDÅTHEÅ$ATAÅ!PPENDIXÅ4HEÅUNITÅOFÅOBSERVATIONÅISÅAÅCITY YEARÅ4HEÅFULLÅSAMPLEÅCONSISTSÅOFÅ THEÅÅCITIESÅOUTÅOFÅTOTALÅ ÅTHATÅUSEÅMANAGERÅGOVERNMENTÅATÅSOMEÅPOINTÅDURINGÅTHEÅSAMPLEÅPERIODÅ)NÅEACHÅOFÅTHEÅTHREEÅPANELSÅTHEÅFIRSTÅTWOÅCOLUMNSÅUSEÅ THEÅFULLÅAVAILABLEÅSAMPLEÅWHILEÅTHEÅLASTÅTWOÅCOLUMNSÅRESTRICTÅTHEÅSAMPLEÅTOÅTHOSEÅCITIESÅTHATÅHAVEÅATÅLEASTÅONEÅMANAGERÅSERVINGÅINÅATÅLEASTÅONEÅOTHERÅCITYÅINÅ THEÅSAMPLEÅ4HEÅ&%SÅAREÅBASEDÅONÅAÅCITYÅMANAGERjSÅFIRSTÅANDÅLASTÅNAMEÅ4HEREÅAREÅÅMANAGERSÅINÅTHISÅSAMPLE ÅOFÅWHICHÅÅSERVEDÅINÅATÅLEASTÅTWOÅCITIESÅ 4HEÅREPORTEDÅNUMBERÅ OFÅCONSTRAINTSÅFALLSÅ BELOWÅTHISÅMAXIMUMÅDUEÅTOÅ MISSINGÅDATAÅ ONÅTHEÅDEPENDENTÅVARIABLESÅ4HEÅ TABLEÅREPORTSÅ&Å TESTÅSTATISTICS ÅANDÅ CORRESPONDINGÅP VALUES ÅFORÅTHEÅJOINTÅSTATISTICALÅSIGNIFICANCEÅOFÅADDINGÅ-ANAGERÅ&IXEDÅ%FFECTSÅTOÅAÅREGRESSIONÅMODELÅTHATÅINCLUDESÅYEARÅANDÅCITYÅFIXEDÅEFFECTSÅ 3TANDARDÅERRORSÅAREÅCLUSTEREDÅATÅTHEÅCITYÅLEVELÅ
0ERIODÅ -ANAGERÅ#ITIESÅ /BSERVATIONSÅ
Å $EPÅ6ARÅ Å Å& STATÅ-ANAGERÅ&%ÅÅ P VALUEÅ-ANAGERÅ&%Å Å#ONSTRAINTSÅ !DJUSTEDÅ2Å Å9EARÅ&IXEDÅ%FFECTSÅ #ITYÅ&IXEDÅ%FFECTSÅ -ANAGERÅ&IXEDÅ%FFECTSÅ
-ODELÅ
4!",%Å!Å-ANAGERÅ&IXEDÅ%FFECTSÅ
ACCEPTED MANUSCRIPT
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
/,3Å Å Å
ED
RI P
SC
NU
MA
Å Å Å Å Å Å Å Å 0ANELÅ"ÅÅ#HARACTERISTICS Å0OST ,&#Å Å Å Å Å %ARLYÅ#IVILÅ -ANAGERÅ .ON 0ARTISANÅ &RACTIONÅ3EATSÅÅ !T ,ARGEÅ 3ERVICEÅ %LECTIONSÅ Å Å Å Å
Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ #ROSS 3ECTIONÅ #ROSS 3ECTIONÅ #ROSS 3ECTIONÅ #ROSS 3ECTIONÅ
CE
PT
Å
ÅÅ Å 9ESÅ 9ESÅ #ROSS 3ECTIONÅ
ÅÅ
Å Å ÅÅ
Å ಥÅ Å Å Å ÅÅ
LOG0OPUL Å
/,3Å Å Å
XVIIÅ
Å
T
.OTESÅ!UTHORSjÅCALCULATIONSÅWITHÅCITYÅDATAÅASÅDESCRIBEDÅINÅTHEÅ$ATAÅ!PPENDIXÅ4HEÅUNITÅOFÅOBSERVATIONÅISÅAÅCITYÅFORÅTHEÅSAMPLEÅOFÅLARGESTÅÅSELF GOVERNINGÅCITIESÅINÅTHEÅ5NITEDÅ3TATESÅASÅOFÅÅ4HEÅTABLEÅREPORTSÅ/,3ÅESTIMATESÅOFÅ α1 Å FROMÅEQUATIONÅ ÅINÅTHEÅTEXT ÅWITHÅ -ANAGERÅREPLACEDÅBYÅ THEÅ INDICATEDÅ VARIABLEÅ 3TANDARDÅ ERRORSÅ CLUSTEREDÅ ATÅ THEÅ CITYÅ LEVELÅ INÅ PARENTHESESÅ 7EATHERÅ CONTROLSÅ #ENTURYÅ 0RECIPITATIONÅ 3HOCKSÅ -EDIANÅ 0RECIPITATIONÅ7ATERÅCONTROLSÅNUMBERÅOFÅ,AKESÅNUMBERÅOFÅ3WAMPSÅONEÅINDICATORÅFORÅLOCATIONÅONÅ#OASTÅTHREEÅINDICATORSÅFORÅ3MALLÅ2IVERS Å-EDIUMÅ 2IVERS Å ,ARGEÅ2IVERSÅ ÅINDICATESÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL Å
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL ÅANDÅ
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅ LEVELÅ
7ATERÅ 0ERIODÅ
ÅÅ Å 9ESÅ 9ESÅ #ROSS 3ECTIONÅ
ÅÅ
,&#Å0RECIPITATIONÅ3HOCKSÅ Å
Å/BSERVATIONSÅ Å7EATHERÅ
Å Å ÅÅ
Å Å Å ÅÅ
ÅÅ $EPÅ6ARIABLEÅ ÅÅ
ÅÅ Å
Å/BSERVATIONSÅ
/,3Å Å Å
/,3Å /,3Å /,3Å Å Å Å Å Å Å 0ANELÅ!ÅÅ#HARACTERISTICS Å0RE ,&#Å LOG0OLICEÅ#IVÅ LOG!RRESTSÅPERÅ Å LOG3EWERSÅPERÅ LOG2OADSÅPERÅÅ 3QÅ-ILE Å PERÅ#AP Å #APITA Å 3QÅ-ILEÅ Å Å Å Å Å ಥÅ Å ಥ Å ಥÅ Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å /,3Å /,3Å /,3Å /,3Å
AC
Å ÅÅ $EPÅ6ARIABLEÅ ÅÅ LOG0OLICEÅ/FFÅ PERÅ#AP Å Å Å ,&#Å0RECIPITATIONÅ3HOCKSÅ ಥÅ
-ODELÅ
4!",%Å!Å)NSTRUMENTÅ#OVARIATEÅ"ALANCEÅANDÅ%XCLUSIONÅ2ESTRICTIONÅ
ACCEPTED MANUSCRIPT
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
Å
PT
CE
AC ED
MA
Å /,3Å /,3Å Å Å Å -ANAGERÅ Å
Å Å
Å Å Å ಥÅ ಥÅ Å Å Å 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ .OÅ Å Å Å Å Å Å /,3Å /,3Å Å Å Å Å LOG0OLICEÅ/FFICERSÅPERÅ#APITA Å Å ಥ
Å ಥ
Å Å Å ಥÅ Å Å Å Å 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ .OÅ Å Å Å Å Å Å
XVIIIÅ
RI P
SC
NU
T
.OTESÅ!UTHORSjÅCALCULATIONSÅWITHÅ ÅCITYÅDATAÅASÅDESCRIBEDÅINÅTHEÅ$ATAÅ!PPENDIXÅ4HEÅUNITÅOFÅOBSERVATIONÅISÅ AÅ CITY YEARÅ FORÅ THEÅ SAMPLEÅ OFÅ LARGESTÅ Å SELF GOVERNINGÅ CITIESÅ INÅ THEÅ 5NITEDÅ 3TATESÅ ASÅ OFÅ Å 4HEÅ TABLEÅ REPORTSÅ ESTIMATESÅ OFÅ α1 Å INÅ EQUATIONÅ Å INÅ THEÅ TEXTÅ INÅ COLUMNSÅ Å ANDÅ )6Å REDUCEDÅ FORMSÅ INÅ COLUMNSÅ Å 3TANDARDÅ ERRORSÅ CLUSTEREDÅ ATÅ THEÅ CITYÅ LEVELÅ INÅ PARENTHESESÅ 7EATHERÅ CONTROLSÅ #ENTURYÅ 0RECIPITATIONÅ 3HOCKSÅ -EDIANÅ 0RECIPITATIONÅ 7ATERÅ CONTROLSÅ NUMBERÅ OFÅ ,AKESÅ NUMBERÅ OFÅ 3WAMPSÅ ONEÅ INDICATORÅ FORÅ LOCATIONÅ ONÅ #OASTÅ THREEÅ INDICATORSÅ FORÅ 3MALLÅ 2IVERS Å -EDIUMÅ 2IVERS Å ,ARGEÅ 2IVERSÅ Å INDICATESÅ SIGNIFICANCEÅ ATÅ THEÅ Å PERCENTÅ LEVEL Å
Å SIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL ÅANDÅ
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVELÅ
7EATHERÅ 7ATERÅ $IVISIONÅ&IXEDÅ%FFECTSÅ 0ERIODÅ #ITIESÅ /BSERVATIONSÅ
Å9EARÅ&IXEDÅ%FFECTSÅ
&&#Å0RECIPITATIONÅ3HOCKSÅ
Å Å $EPÅ6ARIABLEÅ Å Å ,&#Å0RECIPITATIONÅ3HOCKSÅ
-ODELÅ
4!",%Å!Å)6Å2EDUCEDÅ&ORMÅANDÅ&ALSIFICATIONÅ
ACCEPTED MANUSCRIPT
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
Å
ED
RI P
SC
NU
MA
XIXÅ
Å Å
T
.OTESÅ!UTHORSjÅCALCULATIONSÅWITHÅ ÅCITYÅDATAÅASÅDESCRIBEDÅINÅTHEÅ$ATAÅ!PPENDIXÅ4HEÅUNITÅOFÅOBSERVATIONÅISÅAÅ CITY YEARÅ FORÅ THEÅ SAMPLEÅ OFÅ LARGESTÅ Å SELF GOVERNINGÅ CITIESÅ INÅ THEÅ 5NITEDÅ 3TATESÅ ASÅ OFÅ Å 4HEÅ TABLEÅ REPORTSÅ )6Å ESTIMATESÅOFÅȕÅ FROMÅEQUATIONÅ ÅINÅTHEÅTEXTÅINÅTHEÅFIRSTÅPANELÅANDÅFIRST STAGEÅ/,3ÅESTIMATESÅOFÅ α1 ÅINÅEQUATIONÅ ÅINÅ THEÅ TEXTÅ INÅ THEÅ SECONDÅ PANELÅ 3TANDARDÅ ERRORSÅ CLUSTEREDÅ ATÅ THEÅ CITYÅ LEVELÅ INÅ PARENTHESESÅ 7EATHERÅ CONTROLSÅ #ENTURYÅ 0RECIPITATIONÅ3HOCKSÅ -EDIANÅ 0RECIPITATIONÅ 7ATERÅ CONTROLSÅ NUMBERÅ OFÅ ,AKESÅ NUMBERÅ OFÅ 3WAMPSÅ ONEÅ INDICATORÅ FORÅ LOCATIONÅONÅ#OASTÅTHREEÅINDICATORSÅFORÅ 3MALLÅ2IVERS Å -EDIUMÅ2IVERS Å ,ARGEÅ2IVERSÅ ÅINDICATESÅSIGNIFICANCEÅATÅTHEÅÅ PERCENTÅLEVEL Å
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL ÅANDÅ
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVELÅ
PT
CE
Å )6Å )6Å )6Å )6Å Å Å Å Å ,&#Å%RAÅ$EFINITIONÅ 0RECIPITATIONÅ3HOCKÅ$EFINITIONÅ Å Å 4OPÅÅ$UMMYÅ 4OPÅÅ$UMMYÅ Å Å Å Å Å 3ECONDÅ3TAGEÅ$EPENDENTÅ6ARIABLEÅ ÅLOG 0OLICEÅ/FFICERSÅPERÅ#APITA Å ಥ Å ಥ
Å ಥ Å ಥ Å Å Å Å Å Å Å Å Å &IRSTÅ3TAGEÅ$EPENDENTÅ6ARIABLEÅ Å-ANAGERÅ Å
Å
Å
Å
Å Å Å Å Å Å Å Å Å ;=Å ;=Å ;=Å ;=Å Å Å Å Å Å 2EDUCEDÅ&ORMÅ$EPENDENTÅ6ARIABLEÅ ÅLOG 0OLICEÅ/FFICERSÅPERÅ#APITA Å Å Å Å ಥ Å ಥ
Å ಥ Å ಥ Å Å Å Å Å Å Å Å Å 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ .OÅ .OÅ .OÅ Å Å Å Å Å Å Å Å Å Å Å Å
AC
Å Å ÅÅ Å
Å ,&#Å0RECIPITATIONÅ3HOCKSÅ Å Å & STATISTICÅ ;P VALUE=Å Å Å ,&#Å0RECIPITATIONÅ3HOCKSÅ Å Å Å9EARÅ&IXEDÅ%FFECTSÅ 7EATHERÅ 7ATERÅ $IVISIONÅ&IXEDÅ%FFECTSÅ 0ERIODÅ #ITIESÅ /BSERVATIONSÅ
Å
Å
-ANAGERÅ
Å
-ODELÅ
4!",%Å!Å!LTERNATIVEÅ)NSTRUMENTÅ$EFINITIONSÅ)Å
ACCEPTED MANUSCRIPT
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
ED
Å
RI P
SC
NU
MA
XXÅ
T
.OTESÅ!UTHORSjÅCALCULATIONSÅWITHÅ ÅCITYÅDATAÅASÅDESCRIBEDÅINÅTHEÅ$ATAÅ!PPENDIXÅ4HEÅUNITÅOFÅOBSERVATIONÅISÅAÅ CITY YEARÅ FORÅ THEÅ SAMPLEÅ OFÅ LARGESTÅ Å SELF GOVERNINGÅ CITIESÅ INÅ THEÅ 5NITEDÅ 3TATESÅ ASÅ OFÅ Å 4HEÅ TABLEÅ REPORTSÅ )6Å ESTIMATESÅOFÅȕÅ FROMÅEQUATIONÅ ÅINÅTHEÅTEXTÅINÅTHEÅFIRSTÅPANELÅANDÅFIRST STAGEÅ/,3ÅESTIMATESÅOFÅ α1 ÅINÅEQUATIONÅ ÅINÅ THEÅ TEXTÅ INÅ THEÅ SECONDÅ PANELÅ 3TANDARDÅ ERRORSÅ CLUSTEREDÅ ATÅ THEÅ CITYÅ LEVELÅ INÅ PARENTHESESÅ 7EATHERÅ CONTROLSÅ #ENTURYÅ 0RECIPITATIONÅ3HOCKSÅ -EDIANÅ 0RECIPITATIONÅ 7ATERÅ CONTROLSÅ NUMBERÅ OFÅ ,AKESÅ NUMBERÅ OFÅ 3WAMPSÅ ONEÅ INDICATORÅ FORÅ LOCATIONÅONÅ#OASTÅTHREEÅINDICATORSÅFORÅ 3MALLÅ2IVERS Å -EDIUMÅ2IVERS Å ,ARGEÅ2IVERSÅ ÅINDICATESÅSIGNIFICANCEÅATÅTHEÅÅ PERCENTÅLEVEL Å
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL ÅANDÅ
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVELÅ
Å Å9EARÅ&IXEDÅ%FFECTSÅ 7EATHERÅ 7ATERÅ $IVISIONÅ&IXEDÅ%FFECTSÅ 0ERIODÅ #ITIESÅ /BSERVATIONSÅ
Å,&#Å0RECIPITATIONÅ3HOCKSÅ
PT
CE
)6Å Å
)6Å )6Å )6Å Å Å Å 0RECIPITATIONÅ!GGREGATIONÅ !NNUALÅ 1UARTERLYÅ -ONTHLYÅ $AYÅ-AXÅ Å Å Å Å Å 3ECONDÅ3TAGEÅ$EPENDENTÅ6ARIABLEÅ ÅLOG0OLICEÅ/FFICERSÅPERÅ#APITA Å ಥ
Å ಥÅ ಥ Å ಥÅ Å Å Å Å Å Å Å -ANAGERÅ Å &IRSTÅ3TAGEÅ$EPENDENTÅ6ARIABLEÅ Å Å
Å
Å
Å Å Å Å Å Å Å Å Å Å ;=Å ;=Å ;=Å ;=Å Å Å Å Å Å 2EDUCEDÅ&ORMÅ$EPENDENTÅ6ARIABLEÅ ÅLOG 0OLICEÅ/FFICERSÅPERÅ#APITA Å Å Å Å ಥ
Å ಥ
Å ಥ
Å ಥÅ Å Å Å Å Å Å Å Å 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ .OÅ .OÅ .OÅ Å Å Å Å Å Å Å Å Å Å Å Å
AC
Å Å ÅÅ Å
Å ,&#Å0RECIPITATIONÅ3HOCKSÅ Å Å & STATISTICÅ ;P VALUE=Å ÅÅ
Å
Å
-ANAGERÅ
Å
-ODELÅ
4!",%Å!Å!LTERNATIVEÅ)NSTRUMENTÅ$EFINITIONSÅ))Å
ACCEPTED MANUSCRIPT
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
Å
PT
CE
AC ED
Å /,3Å /,3Å )6Å Å Å Å ÅLOG#ITYÅ3PENDINGÅPERÅ#APITA Å ÅÅ Å
Å
Å Å Å Å Å Å Å Å 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ 9ESÅ .OÅ Å Å Å Å Å Å Å Å Å /,3Å /,3Å )6Å Å Å Å Å Å Å Å LOG0OLICEÅ$EPTÅ3PENDINGÅPERÅ#APITA Å Å Å ಥÅ Å Å Å Å Å Å 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ 9ESÅ .OÅ Å Å Å Å Å Å Å Å Å
XXIÅ
RI P
SC
NU
MA
T
.OTESÅ!UTHORSjÅCALCULATIONSÅWITHÅ ÅCITYÅDATAÅASÅDESCRIBEDÅINÅTHEÅ$ATAÅ!PPENDIXÅ4HEÅUNITÅOFÅOBSERVATIONÅISÅAÅCITY YEARÅFORÅTHEÅ SAMPLEÅOFÅLARGESTÅÅSELF GOVERNINGÅCITIESÅINÅTHEÅ5NITEDÅ3TATESÅASÅOFÅÅ4HEÅTABLEÅREPORTSÅ/,3ÅANDÅ)6ÅESTIMATESÅOFÅȕÅ FROMÅEQUATIONÅ ÅINÅTHEÅMAINÅTEXTÅWITHÅLOG%MPLOYMENT ÅREPLACEDÅBYÅTHEÅINDICATEDÅDEPENDENTÅVARIABLESÅ3TANDARDÅERRORSÅCLUSTEREDÅATÅTHEÅCITYÅLEVELÅINÅ PARENTHESESÅ 7EATHERÅ CONTROLSÅ #ENTURYÅ 0RECIPITATIONÅ 3HOCKSÅ -EDIANÅ 0RECIPITATIONÅ 7ATERÅ CONTROLSÅ NUMBERÅ OFÅ ,AKESÅ NUMBERÅ OFÅ 3WAMPSÅONEÅINDICATORÅFORÅLOCATIONÅONÅ#OASTÅTHREEÅINDICATORSÅFORÅ3MALLÅ2IVERS Å-EDIUMÅ2IVERS Å,ARGEÅ2IVERSÅ ÅINDICATESÅSIGNIFICANCEÅATÅ THEÅÅPERCENTÅLEVEL Å
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL ÅANDÅ
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVELÅ Å Å Å
7EATHERÅ 7ATERÅ $IVISIONÅ&IXEDÅ%FFECTSÅ 0ERIODÅ #ITIESÅ /BSERVATIONSÅ
Å9EARÅ&IXEDÅ%FFECTSÅ
-ANAGERÅ
Å $EPÅ6ARIABLEÅ ÅÅ
-ODELÅ
4!",%Å!Å'OVERNMENTÅ&ORMÅANDÅ0UBLICÅ3PENDINGÅ
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT 4!",%Å!Å'OVERNMENTÅ&ORMÅANDÅ0OPULARÅ0OLICYÅ2OBUSTNESSÅ Å -ODELÅ
CE
PT
ED
MA
NU
SC
RI P
T
/,3Å /,3Å )6Å Å Å Å ÅLOG0OLICEÅ/FFICERSÅPERÅ#APITA Å $EPÅ6ARÅ ÅÅ ÅÅ Å ಥ
Å Å "ASELINEÅFROMÅ4ABLEÅ ÅCOLSÅ ÅANDÅ Å ಥ
Å ಥ
Å Å Å Å Å Å Å Å 0ANELÅ!Å!LTERNATIVEÅ3AMPLESÅ Å Å$ROPÅ#ITIESÅWITHÅÅ0OPULATIONÅOVERÅMILÅ Å Å Å ಥ
Å ಥ
Å ಥ
Å Å Å Å Å$ROPÅ#ITIESÅWITHÅÅ0OPULATIONÅUNDERÅKÅ ಥ
Å ಥ
Å ಥ Å Å Å Å Å$ROPÅPRE Å0ERIODÅ ಥ
Å ಥ
Å ಥ Å Å Å Å Å Å$ROPÅPOST #/03Å0ROGRAMÅ0ERIODÅ ಥ
Å ಥ
Å ಥ
Å Å Å Å Å$ROPÅ/FFICERÅPERÅ#APITAÅ/UTLIERÅ/BSERVATIONSÅ ಥ
Å ಥ
Å ಥ
Å Å Å Å Å$ROPÅ&ARÅFROMÅ7EATHERÅ3TATIONÅ#ITIESÅ ಥ
Å ಥ
Å ಥÅ Å Å Å Å Å Å Å 0ANELÅ"Å!LTERNATIVEÅ3TANDARDÅ%RRORÅ#ONSTRUCTIONÅANDÅ)NFERENCESÅ Å Å#LUSTERÅONÅBOTHÅ#ITYÅANDÅ9EARÅ Å Å Å ಥ
Å ಥ
Å ಥ
Å Å Å Å Å#LUSTERÅONÅ7EATHERÅ3TATIONÅ ಥ
Å ಥ
Å ಥ
Å Å Å Å Å#ONLEYÅ3TDÅ%RRORSÅ.OÅ9EARÅ&IXEDÅ%FFECTS Å ಥ
Å ಥ
Å ಥ
Å Å Å Å Å Å Å Å 0ANELÅ#Å!LTERNATIVEÅ0OLICEÅ/FFICERÅ3CALINGÅ Å Å0ERÅ2EPORTEDÅ#RIMEÅ Å Å Å ಥ
Å ಥ
Å ಥ
Å Å Å Å Å0ERÅ#LEAREDÅ#RIMEÅ ಥ
Å ಥ
Å ಥ
Å Å Å Å Å Å9EARÅ&IXEDÅ%FFECTSÅ 9ESÅ 9ESÅ 9ESÅ 7EATHERÅ 9ESÅ 9ESÅ 9ESÅ 7ATERÅ 9ESÅ 9ESÅ 9ESÅ $IVISIONÅ&IXEDÅ%FFECTSÅ .OÅ 9ESÅ .OÅ
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
.OTESÅ !UTHORSjÅ CALCULATIONSÅ WITHÅ Å CITYÅ DATAÅ ASÅ DESCRIBEDÅ INÅ THEÅ $ATAÅ !PPENDIXÅ 4HEÅ UNITÅ OFÅ OBSERVATIONÅ ISÅ AÅ CITY YEARÅ FORÅ THEÅ SAMPLEÅ OFÅ LARGESTÅ Å SELF GOVERNINGÅ CITIESÅ INÅ THEÅ 5NITEDÅ 3TATESÅ ASÅ OFÅ Å4HEÅTABLEÅREPORTSÅESTIMATESÅOFÅȕÅINÅEQUATIONÅ ÅINÅTHEÅMAINÅTEXTÅ3TANDARDÅERRORSÅCLUSTEREDÅATÅTHEÅ CITYÅ LEVELÅ REPORTEDÅ INÅ PARENTHESESÅ EXCEPTÅ INÅ 0ANELÅ " Å 7EATHERÅ CONTROLSÅ #ENTURYÅ 0RECIPITATIONÅ 3HOCKSÅ -EDIANÅ0RECIPITATIONÅ7ATERÅCONTROLSÅNUMBERÅOFÅ ,AKESÅNUMBERÅOFÅ 3WAMPSÅONEÅINDICATORÅFORÅLOCATIONÅONÅ #OASTÅ THREEÅ INDICATORSÅ FORÅ 3MALLÅ 2IVERS Å -EDIUMÅ 2IVERS Å ,ARGEÅ 2IVERSÅ 3EEÅ MAINÅ TEXTÅ FORÅ DETAILEDÅ DESCRIPTIONSÅOFÅTHEÅSPECIFICATIONSÅINÅPANELSÅ! Å" ÅANDÅ#Å ÅINDICATESÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL Å
Å SIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL ÅANDÅ
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVELÅ
Å Å
XXIIÅ
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
ÅÅ
Å Å
Å
Å
PT
CE
AC
MA
Å
ED
Å
SC
Å
Å
Å /,3Å /,3Å /,3Å /,3Å /,3Å /,3Å Å Å Å Å Å Å !NNUALÅ0RECIPITATIONÅ#UTOFFÅ 4OPÅÅ0RECIPITATIONÅ!GGREGATIONÅÅ &RACTIONÅ9RSÅ &RACTIONÅ9RSÅ &RACTIONÅ9RSÅ 1UARTERLYÅ -ONTHLYÅ $AYÅ-AXÅ 4OPÅÅ 4OPÅÅ 4OPÅÅ Å Å Å Å 3OIL "ASEDÅ&LOODING Å Å Å $EPENDENTÅ6ARIABLEÅ Å Å Å Å
Å
Å
Å
Å
Å −Å Å Å Å Å Å Å −Å Å ಥÅ ಥÅ ಥ
Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å #ROSS 3ECTIONÅ #ROSS 3ECTIONÅ #ROSS 3ECTIONÅ #ROSS 3ECTIONÅ #ROSS 3ECTIONÅ #ROSS 3ECTIONÅ
XXIIIÅ
NU
RI P
T
.OTESÅ!UTHORSjÅCALCULATIONSÅWITHÅCITYÅDATAÅASÅDESCRIBEDÅINÅTHEÅ$ATAÅ!PPENDIXÅ4HEÅUNITÅOFÅOBSERVATIONÅISÅAÅCITYÅFORÅTHEÅSAMPLEÅOFÅLARGESTÅÅSELF GOVERNINGÅ CITIESÅ INÅ THEÅ 5NITEDÅ 3TATESÅ ASÅ OFÅ Å 4HEÅ TABLEÅ REPORTSÅ /,3Å ESTIMATESÅ OFÅ EQUATIONÅ Å INÅ THEÅ TEXTÅ WITHÅ -ANAGERÅ REPLACEDÅ BYÅ THEÅ INDICATEDÅDEPENDENTÅVARIABLEÅ3TANDARDÅERRORSÅCLUSTEREDÅATÅTHEÅCITYÅLEVELÅINÅPARENTHESESÅ&ORÅCOLUMNSÅ ÅPRECIPITATIONÅSHOCKSÅAREÅDEFINEDÅASÅ FRACTIONÅ OFÅ YEARSÅ WITHÅ ANNUALÅ PRECIPITATIONÅ INÅ THEÅ TOPÅ XÅ PERCENTÅ OFÅ THEÅ NATIONALÅ PRECIPITATIONÅ DISTRIBUTION Å WHEREÅ XÅ Å Å Å Å &ORÅ COLUMNSÅ ÅPRECIPITATIONÅSHOCKSÅAREÅDEFINEDÅASÅFRACTIONÅOFÅYEARSÅWITHÅ YÅPRECIPITATIONÅAGGREGATIONÅINÅTHEÅTOPÅÅPERCENTÅOFÅTHEÅNATIONALÅPRECIPITATIONÅ DISTRIBUTION ÅWHEREÅ YÅ ÅQUARTERLY ÅMONTHLY ÅANDÅTHEÅANNUALÅMAXIMUMÅOFÅ DAYÅCUMULATIVEÅPRECIPITATIONÅ ÅINDICATESÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅ LEVEL Å
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL ÅANDÅ
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVELÅ
/BSERVATIONSÅ
Å2Å Å0ERIODÅ
-EDIANÅ0RECIPITATIONÅ
#ENTURYÅ0RECIPITATIONÅ3HOCKSÅ Å
Å
-ODELÅ
4!",%Å!Å0RECIPITATIONÅ3HOCKSÅANDÅ&LOODINGÅ#ROSS 3ECTIONÅ
ACCEPTED MANUSCRIPT
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
/,3Å Å
PT
/,3Å Å
CE
AC
/,3Å Å
ED
MA
XXIV
RI P
SC
NU
T
.OTESÅ!UTHORSjÅCALCULATIONSÅWITHÅCITYÅDATAÅASÅDESCRIBEDÅINÅTHEÅ$ATAÅ!PPENDIXÅ4HEÅUNITÅOFÅOBSERVATIONÅISÅAÅCITY YEARÅFORÅTHEÅSAMPLEÅOFÅLARGESTÅÅSELF GOVERNINGÅCITIESÅINÅTHEÅ5NITEDÅ3TATESÅASÅOFÅÅ4HEÅRESTRICTEDÅ CITYÅSAMPLEÅCONTAINSÅONLYÅCITIESÅWITHÅAÅFLOODÅBEFOREÅÅANDÅATÅLEASTÅONEÅFLOODÅDURINGÅ Å3TANDARDÅERRORSÅCLUSTEREDÅATÅTHEÅCITYÅLEVELÅINÅPARENTHESESÅ4HEÅTABLEÅREPORTSÅ/,3ÅESTIMATESÅOFÅȕÅ FROMÅEQUATIONÅ ÅINÅTHEÅTEXTÅWITHÅ -ANAGERÅ REPLACEDÅBYÅ 2IVER "ASEDÅ&LOODING ÅANDÅCITYÅFIXEDÅEFFECTSÅINCLUDEDÅASÅADDITIONALÅCONTROLSÅ 0RECIPITATIONÅ 3HOCKÅ VARIABLESÅ AREÅBINARY ÅTAKINGÅAÅVALUEÅOFÅONEÅIFÅ ANÅEXTREMEÅPRECIPITATIONÅEVENTÅOCCURSÅINÅTHEÅRELEVANTÅDATAÅFREQUENCYÅUSINGÅTHEÅNATIONALÅÅCUTOFFÅFORÅTHEÅRELEVANTÅMEASUREÅ ÅINDICATESÅSIGNIFICANCEÅATÅTHEÅ ÅPERCENTÅLEVEL Å
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL ÅANDÅ
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVELÅ
/,3Å Å
Å /,3Å /,3Å /,3Å /,3Å /,3Å /,3Å /,3Å Å Å Å Å Å Å Å Å Å 0RECIPITATIONÅ!GGREGATIONÅ Å !NNUALÅ 1UARTERLYÅ -ONTHLYÅ $AYÅ-AXÅ Å Å Å Å Å Å Å Å Å Å Å Å Å $EPENDENTÅ6ARIABLEÅ Å2IVER "ASEDÅ&LOODINGÅ Å Å Å Å Å Å Å Å Å Å Å Å 0RECIPITATIONÅ3HOCK Å
Å Å Å Å
Å Å
Å
Å Å Å
Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å 9EARÅ&IXEDÅ%FFECTSÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ #ITYÅ&IXEDÅ%FFECTSÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ -ISSINGÅ)MPUTATIONÅ NOÅFLOODÅ MISSINGÅ NOÅFLOODÅ MISSINGÅ MISSINGÅ NOÅFLOODÅ MISSINGÅ MISSINGÅ NOÅFLOODÅ MISSINGÅ MISSINGÅ $ATAÅ&REQUENCYÅ 9EARÅ 9EARÅ 9EARÅ 9EARÅ 1UARTERÅ 9EARÅ 9EARÅ -ONTHÅ 9EARÅ 9EARÅ -ONTHÅ 0ERIODÅ Å Å Å Å Å Å Å Å Å Å Å #ITIESÅ Å Å Å Å Å Å Å Å Å Å Å /BSERVATIONSÅ Å Å Å Å Å Å Å Å Å Å Å
-ODELÅ
4!",%Å!Å0RECIPITATIONÅ3HOCKSÅANDÅ&LOODINGÅ4IMEÅ3ERIESÅ
ACCEPTED MANUSCRIPT
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
Å
Å
ED
Å
NU
nÅ Å nÅ Å Å Å Å 9ESÅ 9ESÅ 9ESÅ !BOVEÅKÅ Å Å Å
MA
nÅ Å nÅ Å Å Å Å 9ESÅ 9ESÅ .OÅ !BOVEÅKÅ Å Å Å
/,3Å /,3Å Å Å Å Å LOG.ON 0OLICEÅ%MPLÅPERÅ#AP Å
PT
CE
AC
/,3Å /,3Å Å Å Å Å LOG0OLICEÅ#IVILIANSÅPERÅ#AP Å Å nÅ nÅ Å Å nÅ nÅ Å Å Å Å Å Å Å Å 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ 9ESÅ !BOVEÅKÅ !BOVEÅKÅ Å Å Å Å Å Å
/,3Å /,3Å Å Å Å Å LOG#ITYÅ3PENDINGÅPERÅ#AP Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ 9ESÅ !BOVEÅKÅ !BOVEÅKÅ Å Å Å Å Å Å
Å Å Å
XXV
RI P
SC
T
.OTESÅ!UTHORSjÅCALCULATIONSÅWITHÅ ÅCITYÅDATAÅASÅDESCRIBEDÅINÅTHEÅ$ATAÅ!PPENDIXÅ4HEÅUNITÅOFÅOBSERVATIONÅISÅAÅCITY YEARÅ4HEÅSAMPLEÅISÅTHEÅ LARGESTÅ Å SELF GOVERNINGÅ CITIESÅ INÅ THEÅ 5NITEDÅ 3TATESÅ ASÅ OFÅ Å THATÅ HAVEÅ ATÅ LEASTÅ Å RESIDENTSÅ INÅ Å 4HEÅ TABLEÅ REPORTSÅ /,3Å ESTIMATESÅ OFÅ EQUATIONÅ Å INÅ THEÅ TEXTÅ AUGMENTEDÅ WITHÅ THEÅ INDICATEDÅ VARIABLESÅ ANDÅ INTERACTIONÅ TERMSÅ 3TANDARDÅ ERRORSÅ CLUSTEREDÅ ATÅ THEÅ CITYÅ LEVELÅ REPORTEDÅ INÅ PARENTHESESÅ $EMOGRAPHICÅ CONTROLSÅ 0OPULATIONÅ $ENSITYÅ &RACTIONÅ .ON 7HITEÅ &RACTIONÅ #OLLEGEÅ 'RADUATEÅ -EDIANÅ (OUSEHOLDÅ )NCOMEÅ )NDICATORÅ VARIABLESÅFORÅMISSINGÅVALUESÅADDEDÅ ÅINDICATESÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL Å
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL ÅANDÅ
ÅSIGNIFICANCEÅATÅTHEÅ ÅPERCENTÅLEVELÅ Å Å
Å9EARÅ&IXEDÅ%FFECTSÅ #ITYÅ&IXEDÅ%FFECTSÅ $EMOGRAPHICSÅ Å#ITYÅ0OPULATIONÅ 0ERIODÅ #ITIESÅ /BSERVATIONSÅ
-ANAGERÅÑÅ%LECTIONÅ Å %LECTIONÅ
-ANAGERÅ
$EPÅ6ARIABLEÅ ÅÅ
-ODELÅ
4!",%Å!Å#ITYÅ%LECTIONSÅANDÅ/THERÅ0OLICIESÅ
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT Appendix References Bedient, P.B., W.C. Huber, and B. Vieux (2008) Hydrology and Floodplain Analysis, Prentice Hall, Upper Saddle River: NJ.
T
Civil Service Assembly (1938) "Civil Service Agencies in the United States: A 1937 Cen-
RI P
sus," Pamphlet No. 11, January.
Federal Bureau of Investigation, Law Enforcement Management and Administrative Sta-
SC
tistics.
Federal Bureau of Investigation, Uniform Crime Reports. Ferreira, F. and J. Gyourko (2009) "Do Political Parties Matter? Evidence From U.S.
NU
Cities," Quarterly Journal of Economics 124(1): 399-422.
Fishback, P.V., W.C. Horrace, and S. Kantor (2005) "Did New Deal Grant Programs
MA
Stimulate Local Economies? A Study of Federal Grants and Retail Sales During the Great Depression," Journal of Economic History 65(1): 6-71. Fishback, P.V., W.C. Horrace, and S. Kantor (2006) "The Impact of New Deal Expen-
ED
ditures on Mobility During the Great Depression," Explorations in Economic History 43(2): 179-222.
PT
George, L.M. and J. Waldfogel (2006) "The New York Times and the Market for Local Newspapers," American Economic Review 96(1): 435-447.
CE
International County/City Management Association, Municipal Year Book. International County/City Management Association, Salaries of Municipal Officials. Newspaper Enterprise Association, World Almanac and Book of Facts.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
U.S. Census Bureau (1905) "The Statistics of Cities Having a Population of Over 25,000: 1902 and 1903," Bulletin 20. U.S. Census Bureau (1906) "The Statistics of Cities Having a Population 8,000 to 25,000: 1903," Bulletin 45. U.S. Census Bureau, Census of Population. U.S. Census Bureau, Census of Governments. U.S. Geological Service, National Water Information System. Williams, C.N., M.J. Menne, R.S. Vose, and D.R. Easterling (2006) "U.S. Historical Climatology Network Daily Temperature, Precipitation, and Snow Data," Oak Ridge National Laboratory.
xxvi
ACCEPTED MANUSCRIPT Highlights
T
RI P
SC
NU
x
MA
x
ED
x
PT
x
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
CE
x
AC
S0047-2727(15)00215-7 doi: 10.1016/j.jpubeco.2015.12.011 PUBEC 3639
To appear in:
Journal of Public Economics
Received date: Revised date: Accepted date:
30 May 2014 17 December 2015 22 December 2015
Please cite this article as: Vlaicu, Razvan, Whalley, Alexander, Hierarchical Accountability in Government, Journal of Public Economics (2016), doi: 10.1016/j.jpubeco.2015.12.011
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Razvan Vlaicuy
RI P
T
Hierarchical Accountability in Government Alexander Whalleyz
NU
SC
December 2015
Abstract
MA
This paper studies a setting where a relatively uninformed voter holds a policymaker accountable through an informed intermediary. In equilibrium the voter uses the intermediary to insulate the policymaker from pandering incentives when the voters policy
ED
expertise is low or the policymakers congruence is high. The voter can thus enjoy the bene ts of bureaucratic expertise without forfeiting electoral responsiveness. We examine the models predictions using U.S. city-level data, and nd that hierarchically-
PT
accountable managers reduce popular city employment, and adjust it more exibly, than electorally-accountable mayors. The estimated incentive e¤ects are smaller in
CE
cities with high voter expertise and larger during election years, and are robust to instrumentation by precipitation shocks that inuenced early 20th century manager government adoptions for reasons obsolete today.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
JEL Classi cation: D72, D73, H70. Keywords: pandering, accountability, incentives, city manager.
We would like to thank Jim Davies, Allan Drazen, Fernando Ferreira, Price Fishback, Lisa George, Gabriele Gratton, Alex Hirsch, Shawn Kantor, Ethan Kaplan, Ken Shotts, Gergely Ujhelyi, Richard Van Weelden, Emanuel Vespa, Joel Waldfogel, John Wallis, and seminar participants at Florida State University, IIES, New York University, Stanford University, Stockholm University, University of Chicago, University of Houston/Rice University, University of Illinois at Urbana, University of Pittsburgh, Canadian Public Economics Conference, Econometric Society NAWM, and Princeton Political Economy Conference. Erin Moody, Krystal Tapia, Michael Wall, and Marko Zivanovic provided excellent research assistance. y Northwestern University, Kellogg School of Management, 2001 Sheridan Road, MEDS 5th Floor, Evanston, IL 60208-2001, USA. Email: [email protected]. z University of California - Merced, School of Social Sciences, Humanities & Arts, 5200 North Lake Road, Merced, CA 95343, USA; and NBER, 1050 Massachusetts Ave., Cambridge, MA 02138, USA. E-mail: [email protected].
1
ACCEPTED MANUSCRIPT 1
Introduction
The accountability of government o¢cials to the general public is a fundamental principle of democratic governance. Yet, this principles most direct manifestation electoral
T
accountability while a useful safeguard may nevertheless introduce distortions in the poli-
RI P
cymaking process. Having to periodically face a public whose policy expertise is limited gives electorally-accountable o¢cials incentives to disregard their private expertise and adopt pop-
SC
ular policies not necessarily in the public interest by engaging in pandering (Canes-Wrone, Herron, and Shotts 2001) or electoral manipulation (Rogo¤ 1990). Delegating policymaking instead to unaccountable technocrats may allow expertise to be used without fear of
NU
electoral retribution but runs the risk that technocrats may pursue private goals that can disconnect policymaking from the public interest (Maskin and Tirole 2004). An important
MA
question then is how to resolve this tension between expertise and responsiveness, that is, how to allow public o¢cials expertise to be expressed in policymaking while also preserving accountability to the general public.
ED
In this paper we focus on a setting where the policymaker is accountable to the voter through an intermediary. Policymakers such as prime ministers, city managers, and school
PT
district superintendents cannot be directly removed by voters. However, they remain accountable to the voter in the sense that they can be replaced at will by a popularly elected
CE
intermediary, e.g., legislature, city council, school board. Following contract theory terminology we refer to this type of principal-intermediary-agent relationship as hierarchical accountability.1
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
At rst sight, holding a policymaker accountable through an intermediary seems to disconnect policymaker choices from voter preferences. If the voter is fully informed he can, nevertheless, exploit the intermediarys reelection motivation to control policymaker moral hazard just as e¤ectively as the voter could directly (Persson, Roland, and Tabellini 1997).2 In an incomplete information environment, however, the voter also faces an adverse selection problem which weakens electoral accountability by producing pandering incentives. Even if the voters best interest is to limit policymaker pandering, a commitment problem renders him unable to do so. Ex ante an uninformed voter may prefer to insulate a better-informed policymaker. Ex post, however, he would rather replace an unpopular policymaker because 1
About half of U.S. cities are run by city managers. "If the manager is not responsive to the governing body, it has the authority to terminate the manager at any time." (ICMA 2007, p. 2). Other forms of indirect accountability grant the policymaker a xed term, i.e., top regulators, central bank governors. 2 One caveat is possible policymaker-intermediary (executive-legislature) collusion. In that case, the executive would have more discretion to engage in rent-seeking.
2
ACCEPTED MANUSCRIPT unpopular policies signal the policymaker may have di¤erent preferences.3 When accountability to an imperfectly-informed voter distorts the policymakers incentives in this way, delegating policymaker accountability to an informed intermediary seems
T
justi able. The case for delegation is less clear-cut, however, if the intermediarys prefer-
RI P
ences may also di¤er from the voters. In that case the voter needs to provide incentives for the intermediary to act in the voters interest. Under what conditions can an informed intermediary help the voter insulate the policymaker from pandering incentives?
SC
To understand behavior in this setting we rst develop a hierarchical agency model (voterintermediary-policymaker model) where the voter is uncertain about the optimal policy but
NU
leans toward a "popular" policy. The intermediary is a political expert, i.e., informed about the policymakers type, and the policymaker is a policy expert, i.e., informed about the optimal policy. The models key insight is that the voter can credibly use the intermediary
MA
to insulate the policymaker from popular pressure when pandering is relatively detrimental to the voter. That happens when the voters policy expertise is low or the policymakers
ED
congruence, i.e., alignment with voter preferences, is high. In these cases the voter is ex ante better o¤ retaining an unpopular policymaker and the intermediary allows him to commit to do so ex post. This is because an unpopular policymaker is still more likely congruent,
PT
so retaining an unpopular policymaker signals intermediary congruence. Hierarchical accountability thus o¤ers the voter the exibility to promote policymaking expertise without
CE
forfeiting electoral responsiveness. Voters can credibly and optimally switch between erring on the side of expertise and erring on the side of responsiveness when their informational and political environment changes.4
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
In our empirical application we study policymaking by U.S. city managers, hierarchicallyaccountable o¢cials with the same major policy responsibilities as electorally-accountable mayors, i.e., writing the budget and hiring personnel. While manager government provides a natural measure of hierarchical accountability, how to measure pandering behavior is less clear. We propose that pandering behavior can be detected empirically in policy issues that satisfy three conditions: (i) is a primary (not secondary) policy issue over which the policymaker has jurisdictional control, (ii) the optimal policy is state-contingent, and (iii) the public has a clear stance on this issue. We argue that police o¢cer employment satis es 3
Besley and Smart (2007) and Smart and Sturm (2013) notice the commitment issue in the context of scal rules and term limits, respectively. 4 In contract theory and corporate nance, hierarchical agency models study optimal incentives through wage contracts (e.g., Strausz 1997, Park 2000). Our main theory result echoes the nding in this literature that under asymmetric information an intermediary allows the principal to commit to a broader range of incentive structures for the agent.
3
ACCEPTED MANUSCRIPT these requirements as (i) crime has consistently ranked among the top two local policy issues in Gallup surveys of local attitudes since 1959 (see Gallup 2000) and crime prevention among the top two city services with high "resident sensitivity to quality" (Levin and Tadelis 2010);
T
city executives have juristictional control over public safety, a substantial local budget item,
RI P
(ii) the probability of crime uctuates with economic and social conditions, changing the optimal policy response, and (iii) due to constant voter concern over public safety, demand for police o¢cers by relatively uninformed voters can be expected to persist even when the
SC
probability of crime is low; in contrast, civilian police employment, e.g., administrators, dispatchers, as well as other city employee categories, should not elicit such a clear popular
NU
preference. Across a number of speci cations we nd that on average managers employ 814% fewer police o¢cers per capita than mayors but a comparable number of police civilians and non-police employees per capita.
MA
A central challenge to estimating institutional e¤ects is that institutions may be endogenous to policymaking, for instance through unobserved voter preferences (Aghion, Alesina,
ED
and Trebbi 2004). To address potential endogeneity in accountability form we propose an instrument for manager government. The instrument is based on the observation that before the 1936 Flood Control Act transferred ood prevention from local governments to the fed-
PT
eral Army Corps of Engineers cities often responded to ood-related infrastructure crises by adopting manager government because it facilitated the ascension of engineers into the top
CE
executive o¢ce. We document that pre-1936 precipitation shocks inuenced early switches to manager government for reasons obsolete today and nd that the city employment patterns noted above also appear in this IV setting.5
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
We further explore the theory models incentive mechanisms and nd that policy volatility is higher in manager governments because managers can more often, i.e., when insulated, act on their expertise to adjust the policy to the stochastic state. We also propose a measure of voter expertise and nd that the manager-mayor o¢cer employment di¤erential decreases in cities a¤ected by the crack epidemic, where voters should be more aware of the state of crime. It is also more pronounced in election years, when the theory model predicts incentives should be sharper. We note that these patterns cannot be fully explained by alternative mechanisms, such as pure patronage motivations or policymaker type selection. Our paper relates to the literature on the career concerns of expert policymakers. Maskin and Tirole (2004) show how relying on unaccountable technocrats, e.g., unelected bureaucrats, tenured judges, eliminates pandering incentives but makes removing noncongruent 5 To our knowledge this is the rst instrument for manager government in the literature, if we exclude Baqirs (2002) use of lags of city institutions as instruments for current city institutions.
4
ACCEPTED MANUSCRIPT policymakers harder; thus there is a tradeo¤ between expertise and responsiveness in policymaking. Similarly, media providers improve the monitoring of public o¢cials, but face incentives to be yes men themselves (Ashworth and Shotts 2010); allowing decisionmaking
T
behind closed doors (Fox 2007), or imposing term limits (Smart and Sturm 2013), both
RI P
reduce the electoral pressure on congruent policymakers to be popular, but also allow noncongruent ones to make suboptimal policies; candidate competition reduces pandering, but encourages anti-pandering (Kartik, Squintani, and Tinn 2013). Our theory shows, by con-
SC
trast, that hierarchical accountability a¤ords the voter the exibility to optimally choose between insulating congruent policymakers, on the one hand, and providing pandering in-
NU
centives to noncongruent policymakers, on the other. The voter can thus enjoy the bene ts of bureaucratic expertise without forfeiting electoral responsiveness. Our contribution to the empirical literature is two-fold. First, we provide an empirical
MA
measure of pandering behavior grounded in the logic of the theoretical pandering literature and use it to quantify how pandering incentives respond to informational and electoral
ED
factors. In the previous literature pandering has been measured by looking at how U.S. presidents budgetary proposals respond to public opinion surveys of citizen spending preferences (Canes-Wrone and Shotts 2004), how judges with reelection concerns or facing stronger com-
PT
petition issue more punitive sentences (Huber and Gordon 2004, Gordon and Huber 2007), or how elected judges sentencing behavior follows the ideological preferences of their con-
CE
stitutents (Lim 2013). Second, we contribute to the literature on U.S. city managers which has largely focused on di¤erences in public spending from city mayors. Coate and Knight (2011) survey this literature, note that results have been mixed, and provide new evidence
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
that managers outspend mayors.6 A smaller literature has looked at other policy outcomes: Managers are more likely to privatize city services (Levin and Tadelis 2010), more exibly adjust tax revenues (Vlaicu and Whalley 2011), and reduce full-time city employment (Enikolopov 2014). In contrast, our paper documents di¤erences in policymaking incentives for popular and neutral policies, and the mechanisms behind them. Also, while the previous literature has treated city government form as exogenous, we o¤er an instrumental variable strategy that addresses the di¢cult issue of government form endogeneity. The rest of the paper is organized as follows. Section 2 presents the model. Section 3 provides historical background and introduces the data. Section 4 presents the empirical strategy and results. Section 5 concludes. 6
Coate and Knights (2011) citizen-candidate model attributes di¤erences in spending not to incentives, but to voters electing di¤erent councilmember types. Other papers (Baqir 2002, MacDonald 2008) found no spending di¤erential between managers and mayors.
5
ACCEPTED MANUSCRIPT 2
Incentives Model
To gain insight into how hierarchical accountability operates we extend a standard principalagent model of electoral accountability by introducing an asymmetrically informed intermedi-
T
ary into the voter-policymaker relationship. The goal is to understand how being accountable
RI P
through an informed intermediary a¤ects the policymakers incentives. In what follows, hierarchical accountability is de ned as a setting where the policymaker
SC
can be removed at any time by an intermediary that in turn can be replaced by the voter periodically in an election. Electoral accountability, by contrast, refers to a setting where
NU
the policymaker can be directly replaced by the voter periodically in an election. Setup. At time t a policy choice xt needs to be made from the set of policy alternatives
MA
f0; 1g: One can think of xt = 1 as a high level of the policy, and xt = 0 as a low level. The policys optimality is contingent on the state of the world st 2 f0; 1g prevailing in that period, namely the optimal policy is the one that matches the state. The state is i.i.d. across
ED
periods. Let P fst = 1g denote the probability of the high state. There are three kinds of players: voter, intermediary, and policymaker. The voter gets
PT
a unit of payo¤ when the policy matches the state, zero otherwise: v(xt ; st ) = I fxt = st g. Here I is an indicator function. Notice that, as long as voter information is limited to the state prior ; the voter strictly prefers the high policy xt = 1 if and only if the high state is
CE
more likely, > 21 . In this case we say that the high policy is the popular policy, borrowing Maskin and Tiroles (2004) terminology. In this informational context, voter expertise can be measured by how far is from a half.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
The voter delegates policymaking to a policymaker (P ) and delegates policymaker accountability to an intermediary (I). These two agents are both policy-motivated and o¢cemotivated. Their policy motivation depends on their type, biased or congruent: Pt ; It 2 fB; Cg. For each type, the preferred policy gives him a payo¤ of one, and he receives zero otherwise. A congruent type is fully aligned with the voter, wanting policy xt = 1 in state st = 1 and policy xt = 0 in state xt = 0: A biased type prefers policy xt = 0 with probability in both states. One may think of the biased type wanting to match an "alternative" state that is uncorrelated with the true state st : This payo¤ structure is similar to the one adopted by Fox (2007), who assumes that the biased type always prefers policy xt = 0 in both states. An agent obtains his policy payo¤s as long as the agent is in o¢ce; out of o¢ce an agents payo¤ is normalized to zero.7 7
A previous version of the paper assumed the policymakers type could be either congruent, always
6
ACCEPTED MANUSCRIPT This setting captures the intuition that an unpopular policy signals an unappealing policymaker type. For instance, in a world where the voter is concerned about crime, hiring too few police o¢cers signals that the policymaker is soft on crime, as opposed to tough on
T
crime; or, in a world where the voter is concerned about ination, a low interest rate signals
RI P
that the policymaker is an ination dove, as opposed to an ination hawk. To see this, note that a popular voter preference toward the high policy, i.e., > 21 ; coincides with the biased policymaker leaning toward the low policy, and vice-versa. Since the biased type prefers
SC
the low policy more than half the time, and the congruent type less than half the time, an unpopular policy signals policymaker bias toward the unpopular low policy.
NU
The timing of the in nite-horizon game is as follows. (I) Every period, the policymaker observes the state st 2 f0; 1g and chooses between the unpopular/popular policy: xt 2 f0; 1g: (II) Every period, the intermediary observes the policy and policymaker type (xt ; Pt ) and
MA
decides whether to replace/retain the policymaker: yt 2 f0; 1g: (III) Every other period the voter, having observed two periods of choices [(xt 1 ; yt 1 ); (xt ; yt )] ; but not types Pt ; It
ED
or the current state st ; decides to replace/reelect the intermediary: zt 2 f0; 1g: An agent
exiting at t is succeeded by an agent (challenger) whose type is a new draw from the type distribution. An exiting agent cannot run for o¢ce again.8
PT
For expositional simplicity we assume that reelection incentives a¤ect only election-period behavior (cf. Rogo¤ 1990, Shi and Svensson 2006); in the Theory Robustness section below
CE
we discuss an extension that relaxes this assumption. Let ( jt )t=1;2;::: be two independent sequences of i.i.d. preference shocks, for j = P; I; where jt 2 fB; Cg and P jt = C = j
< P < 1:9 An agents type is either last periods or the current periods preference shock: jt 2 jt 1 ; jt and P jt = jt 1 = :
is the congruent types frequency. Assume
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
1 2
Assume 0 < < 1: These assumptions imply: First, that an agents type is correlated over time between two adjacent periods, but uncorrelated between two non-adjacent periods. In particular, an agents type at t + 1 (non-election period) is correlated with his type at t (election period), but uncorrelated with his type at t 1 (non-election period).10 Second, that although policymaker and intermediary types are initially independent, correlation between preferring the state-matching policy, or dissonant, always preferring the policy that mismatches the state (see also Maskin and Tirole 2004). The analysis is similar and available upon request. 8 A key feature of hierarchical accountability is that the policymaker can be replaced between elections. A city manager, for example, serves at the city councils pleasure. His "job tenure is only secure until the next council meeting" (Stillman 1977, p. 664). The models timing captures this feature by assuming that the intermediary can replace the policymaker in the middle of an electoral term. 9 Using a structural approach, Aruoba, Drazen, and Vlaicu (2015) estimate that over half of incumbent U.S. state governors are congruent with the publics preferences. 10 The type persistence probability is Pfjt = jt 1 g = Pfjt = jt 1 ; jt 1 = jt 1 g = (1 ) ; for any t:
7
ACCEPTED MANUSCRIPT their types can nevertheless develop during the game through the intermediarys equilibrium retention decisions. The policymaker cares about policies he himself chooses. The intermediary cares about Formally, lifetime utility for a
T
the choices of the policymaker he is keeping in o¢ce.
RI P
policymaker at the beginning of an electoral term, i.e., at t 1; is u(xt 1 ; st 1 jPt 1 ) + Q P1 i t+1 u(xi ; si jPi ) i =t y 1 ; for an intermediary it is [u(xt 1 ; st 1 jIt 1 )+ u(xt ; st jIt )]+ i=t Qi P1 2i 2t+2 I I ; s j ; s j [u(x ) + u(x )] 2i t+1 2i t+2 2i t+1 2i t+2 2i t+1 2i t+2 =t z2 t : Here is a i=t
SC
time discount factor, with 0 < < 1:
As is standard, our policymaker is a policy expert who knows whether the popular policy
NU
is optimal (the state st ). The intermediary, on the other hand, is a political expert, in the sense of Gailmard and Jenkins (2009), who knows the policymakers preferences (his type Pt ). This second feature seems reasonable in settings where the intermediary has frequent contact
MA
with the policymaker. That is probably the case for a city council since the policymaker they choose, the city manager, has to attend all city council meetings and sometimes committee
ED
meetings. The voter, by contrast, knows neither the popular policys optimality (moral hazard) nor the policymakers preferences (adverse selection). Thus, this model features two forms of the classic delegation tradeo¤: by delegating policymaking to a policy expert the
PT
voter may bene t from more informed policy choices but may su¤er from the policymakers bias; by delegating policymaker accountability to a political expert the voter may bene t
CE
from more informed retention decisions but may su¤er from the intermediarys bias. We denote voter posterior beliefs about agent types by jt (Xt ; Yt ) P jt+1 = CjXt ; Yt ;
where j = P; I and (Xt ; Yt ) = (x ; y )t =1 is the public history at time t; i.e., policy choices
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
and retention decisions up to time t. We refer to voter posterior beliefs about an agents type as that agents "reputation." Equilibrium. The game described above is a principal-intermediary-agent model with incomplete information. We solve for its Perfect Bayesian Equilibrium, namely: (i) policymaker, intermediary, and voter strategies (xt ; yt ; zt ) that are sequentially rational; and (ii) voter beliefs Pt ; It that are consistent with agent strategies. We add the requirement
that voter strategies be robust to a small fraction of non-strategic agents, implying that in a pooling equilibrium o¤-equilibrium path beliefs reect sincere, i.e., preference-based, behavior; see Maskin and Tirole (2004) for details on this re nement. We restrict attention to pure strategies. Throughout this section we assume that the discount factor is large
enough that both agent types would rather override their current preferred policy for the sake of remaining in o¢ce; the equilibrium will thus characterize agents incentives, i.e., 8
ACCEPTED MANUSCRIPT extrinsic motivation. In the Discussion section below we also consider the case where agents have no career concerns and instead follow their preferences, i.e., intrinsic motivations; that equilibrium captures behavior driven by selection.
T
Since the voter can only replace the policymaker through the intermediary, his problem that following a two-period electoral term [t
RI P
is to reelect that intermediary which seems more in tune with his preferences. We show 1; t], after the voter has observed choices
SC
[(xt 1 ; yt 1 ); (xt ; yt )], the incumbent intermediarys equilibrium public reputation It (Xt ; Yt ) increases whenever ^ It (xt ; yt ) P It+1 = Cj (xt ; yt ) ; It+1 = It = It I : In words, the
voter predicts future congruence based on election-period behavior (xt ; yt ) conditional on
NU
the intermediarys current type persisting into the next period. Since the voters reelection decision at t depends only on election-period variables (xt ; yt ), in non-election periods the intermediary is free to retain only policymakers of his own type; this makes the intermediarys
MA
type in future non-election-periods t + 1; t + 3; and so on, matter to the voter. The following proposition states the implications of this voter calculation for policymaker
ED
incentives. For contrast we also include the case of no intermediary, i.e., electoral accountability, which is the mainstay of the current literature. Proposition 1 For popular policy issues > 12 : (i) Hierarchical accountability either
PT
insulates the policymaker pre-election, if P > ; or creates policymaker pandering incen-
tives pre-election, otherwise; (ii) Electoral accountability always creates policymaker pan-
CE
dering incentives pre-election; (iii) Agents follow their preferences post-election under both accountability forms, e.g., an intermediary retains only policymakers of his own type. Proof. See Online Appendix A.1.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Under electoral accountability the voter uses the policy signal contained in the policy choice xt to learn about the policymakers type. Since > 21 ; the high policy xt = 1 is the popular policy and the biased type leans toward xt = 0. Thus a popular policy is a signal of policymaker congruence, and an unpopular policy is a signal of policymaker bias. The voters best response then is to reelect only after the popular policy. This creates a pandering incentive for both policymaker types, i.e., needing to choose the popular policy regardless of its optimality. Hierarchical accountability, on the other hand, provides the voter an additional signal: the retention signal contained in the intermediarys retention decision yt : Because the intermediary prefers to retain only policymakers of his own type, retention is a signal of intermediary congruence if the policymaker is more likely congruent. Thus, because for a popular policymaker that is always the case, as his reputation Pt (Xt ; Yt ) improves from the prior P > 21 ;
9
ACCEPTED MANUSCRIPT retaining a popular policymaker signals intermediary congruence, and replacing a popular policymaker signals intermediary bias. For an unpopular policymaker the voters reaction is more subtle because an unpopular
T
policymaker, while less appealing than the challenger, may or may not be more likely con-
RI P
gruent, and that is what the intermediarys reputation hinges on. Whether an unpopular policymaker is believed more likely congruent turns out to depend on the relative magni tudes of P and , namely whether P (1 ) > 1 P . This says that the frequency of
SC
congruent policymakers wanting the unpopular policy exceeds the frequency of biased policymakers wanting the unpopular policy. This happens if the state is su¢cienly likely to be
NU
low, or the policymakers prior reputation is su¢ciently good. In such cases, retaining an unpopular policymaker, while never a sequentially rational action for the voter, since the voter expects him to do worse than a challenger, nevertheless signals intermediary congruence;
MA
conversely, replacing an unpopular policymaker signals intermediary bias. Consequently, the voter will reelect an intermediary that retains an unpopular policymaker.
ED
The key implication for incentives is that, when the retention signal dominates the policy signal, i.e., P larger than , the voter gives the intermediary incentives to retain the policymaker regardless of the policymakers policy choices, thus endogenously insulating the
PT
policymaker. Conversely, if the policy signal dominates the retention signal, then retaining an unpopular policymaker signals a biased intermediary. In this case the voter gives
CE
the intermediary incentives to retain only popular policymakers, which in turn gives the policymaker the incentive to pander. Policymaker insulation yields ex ante election-period voter welfare equal to P 1 + 1 P [ (1 ) + (1 ) ] whereas under pandering voter welfare is P + 1 P :
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
The pandering equilibrium thus worsens the high types incentives, but may improve the low types incentives. The bene t of insulation is then P (1 ), while the potential cost is 1 P [2 (1 ) ] ; which is no larger than 1 P . It follows that when the voter plays an insulation equilibrium, that is, when P (1 ) > 1 P , or P > ;
voter welfare always improves. In that sense, the hierarchical structure o¤ers the voter the exibility to switch between insulation and pandering to optimally induce, through the intermediary, the types of policymaker incentives that best serve the voters interest.11 One may argue that another way in which hierarchically-accountable policymakers di¤er 11
The two types of intermediary strategies, leading to either insulation or pandering, are consistent with case studies of U.S. city councils that distinguish between council strategies that insulate the manager, termed "blind faith," versus council strategies that transmit popular preferences, termed "political." See Stillman (1977).
10
ACCEPTED MANUSCRIPT from electorally-accountable policymakers is that they have weaker motivations to deliver political patronage to various constituencies. While patronage could potentially a¤ect popular policies, such as police protection or low taxation, it also typically inuences neutral
T
policies such as in-kind transfers or city jobs that may not elicit a clear popular preference. indi¤erent between the policies xt = 1 and xt = 0. Proposition 2 For neutral policy issues =
1 2
RI P
Neutral policy issues can be modeled by setting equal to a half, making the voter ex ante policymaking outcomes do not vary
SC
with the accountability form in election periods or non-election periods. Proof. See Online Appendix A.1.
NU
Intuitively, a balanced state prior silences the policy signal. Both policymaker types, congruent and biased, would choose either policy half the time. The voter can now credibly insulate an electorally-accountable policymaker from pandering incentives since the voter
MA
cannot link either policy to a particular type. He also nds it welfare-enhancing to do so, as P (1 ) > 1 P [2 (1 ) ] ; making voter welfare under insulation dominate
ED
voter welfare under pandering. For hierarchically-accountable policymakers the equilibrium is also insulating because the retention signal is always stronger than the policy signal, since
P > = 12 . The implication is that pandering incentives are absent on neutral issues for
PT
both accountability forms.
The key comparative statics implications for policymaker incentives, based on the analysis
CE
above, can be stated as follows.
Proposition 3 Hierarchical accountability weakens the policymakers pre-election incentives to pander to the voter, compared to electoral accountability, and results in more equilib-
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
rium matching of the policy to the stochastic state, when: the relevant policy issue elicits a popular preference, and the voters policy expertise on this issue is low or the policymakers congruence P is high. This result implies incentive-driven di¤erences in both the mean and the variance of policy choices between the two accountability forms. Hierarchically-accountable policymakers on average reduce the prevalence of popular policies, and adjust policy more exibly in response to changes in the state, relative to electorally-accountable policymakers. Intuitively the presence of the intermediary allows the voter to credibly insulate the policymaker when pandering is relatively detrimental to the voter. Policymaker insulation is possible when voter expertise is low or policymaker congruence P is high. In these cases the voter is ex ante better o¤ retaining both popular and unpopular policymakers, although ex post the voter would replace an unpopular policymaker. The intermediary provides a commitment de-
11
ACCEPTED MANUSCRIPT vice for the voter to implement his ex ante preferences. This is because when P > >
1 2
an
unpopular policymaker is still more likely congruent, so retaining an unpopular policymaker signals intermediary congruence. Thus, it is sequentially rational for the voter to reelect
T
an intermediary that retained an unpopular policymaker, even though it is not sequentially
RI P
rational for the voter to retain an unpopular policymaker himself. This insulating equilibrium results in less popular and more volatile policy choices than the pandering equilibrium
SC
prevalent under electoral accountability.
Theory Robustness. The theoretical results above are robust to a setup where the voter factors into his reelection decision both election-period and non-election-period policy (cf.
NU
Martinez 2009); see Online Appendix A.2 for a solution to the model under this relaxed assumption. We nd that election-period pandering is driven by the following policymaker
MA
intertemporal tradeo¤: choosing a popular policy against preferences at the beginning of the term results in a sure loss, whereas postponing it for the end of the term results in a potential loss, i.e., because the election periods state may be such that the policymaker
ED
actually prefers the popular policy. The same logic applies to the intermediary. In Online Appendix A.3 we also develop a version of the model that allows for variation
PT
in pandering intensity. The main insight of that model is that the pandering probability is equal to the discounted reelection probability premium obtained for choosing the popular
CE
policy. This premium shrinks monotonically as the policy signal becomes weaker. In the limit, as the state prior approaches 12 ; the policy signal gradually fades out, and the policymakers incentive to pander becomes arbitrarily close to zero. In this sense, the strict
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
separation we made above between popular vs. neutral issues can be seen as a more contin uous characteristic of policy issues as varies in the interval 21 ; 1 . Discussion. Our theoretical results show how the incentives of hierarchically-accountable policymakers may depend on their policy, informational, and political environment; observationally, they predict a negative election-period policy di¤erential in popular policies between hierarchical and electoral accountability, when di¤erential, when
1 2
1 2
< < P ; but no election-period policy
< P < ; or on neutral policies, i.e., close to a half. The propositions
also imply that in the rst case policy outcomes are more volatile under hierarchical accountability because this institutional setting gives the policymaker more freedom to adjust the policy to the stochastic state. The model can also be solved under the assumption that the accountability forms primary role is not to provide incentives but to select certain types of policymakers; see Online 12
ACCEPTED MANUSCRIPT Appendix A.4 for the details of this variant of the model. The key insight of the selection model is that policymaker selection through an informed intermediary is more precise, since the intermediary is better informed, and more frequent, since it can also occur between
T
elections, but may back re if intermediary selection by the voter is weak.
RI P
This tradeo¤ notwithstanding, the timing of policy choices does yield an unambiguous prediction. (a) Under the incentives model, a popular policy di¤erential should arise in election years because policymakers change their behavior pre-election to increase their chances
SC
to stay in o¢ce; see Proposition 1. (b) Under the pure selection model, a popular policy di¤erential could be observed in both non-election and election years; electoral selection
NU
produces its e¤ects post-election only, while hierarchical selection produces its e¤ects both post-election and pre-election. Since selection increases the prevalence of congruent policymakers, and the popular policy is more likely optimal, better selection should be associated
MA
with more popular policies. Thus, the election-year popular policy di¤erential is negative under an incentives model, but positive under a selection model. The non-election-year ; i.e., depends on whether intermediary con-
ED
popular policy di¤erential has the sign of I
gruence dominates voter expertise. The di¤erent predictions of the incentives and selection
PT
models allows us to distinguish between the two models empirically.
Empirical Application
CE
3
Our theoretical results for policymaker incentives, summarized in Proposition 3 above, imply di¤erential policy patterns according to the accountability form. Our empirical application
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
examines popular (police o¢cer employment) and neutral (police civilian employment, nonpolice employment) policy issues in U.S. cities where the executive is either hierarchicallyaccountable (manager) or electorally-accountable (mayor). Assuming >
1 2
seems reason-
able for policy issues such as police o¢cer employment. This makes the high policy xt = 1 of high o¢cer employment the popular policy option. Stucky (2005) reviews the literature on how the publics crime probability prior, also known as "fear of crime" in the criminology literature, a¤ects popular preference for police.12 12
Gallups annual Crime Survey asks the question "Is there more crime in your area than there was a year ago, or less?" Since the survey started in 1973 the percentage of respondents who say "More" exceeded those that say "Less" except for three years 1998, 2000, 2001 (see Gallup 2010). In this context, assuming the state is i.i.d. across periods means that past policy choices do not a¤ect the voters current crime probability prior. This seems to be a good approximation if police have only a short-term e¤ect on committed crime (actual crime), while the probability of crime (latent crime) is driven by more fundamental forces like economic inequality, economic growth, and race relations.
13
ACCEPTED MANUSCRIPT City managers operate under hierarchical accountability because they can be removed at any time by the city council, whose composition can in turn be changed by voters in local elections. City mayors operate under electoral accountability because they can be replaced
T
by voters periodically in local elections. The testable hypotheses can be stated as follows.
RI P
(H1) Managers are less likely than mayors to adopt a popular policy. Neutral policy issues do not exhibit this mean policy di¤erential.
(H2) Managers are more likely than mayors to make changes to popular policies. Neutral
SC
policy issues do not exhibit this policy volatility di¤erential.
(H3) The negative mean di¤erential on popular issues should decrease in voter expertise.
NU
(H4) The negative mean di¤erential on popular issues should be more pronounced in election years.
MA
In what follows we examine the evidence for these predictions. Historical Background. Early U.S. city executives were appointed by state governors or city councils. After major cities like Boston and St. Louis started to popularly elect their
ED
mayors in 1822, electoral accountability for city executives became nearly universal. By the end of the 19th century city politics had became dominated by "machine" politicians
PT
drawing their support from workers and immigrants, and often using illegal means to stay in o¢ce. This crisis in city government accountability sparked an urban reform movement that
CE
coincided with broader social reforms taking place during the Progressive Era (roughly 18901930). Initially reformers targeted cities electoral systems: from district-based partisan to at-large non-partisan elections. The goal was to dilute the power of minorities and parties
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
on which the machines thrived. The Progressives, who wanted not only cleaner government but also more e¢cient government, were later inspired by the "e¢ciency movement"s success with the new corporate form in business enterprises, i.e., CEO accountable to shareholders through a board of directors, and sought to apply this model to city governance. Manager government, rst experimented with in the small city of Staunton, Virginia in 1908, attained broad recognition when the National Municipal League picked it as their recommended government form in the 1915 edition of the Model City Charter. During the 1910s and 1920s most major cities, including New York City, were debating switching to a manager govenment. Despite strong opposition from incumbent mayors and party bosses manager government advanced steadily.13 Knoke (1982) attributes successful switches to manager to strong 13 A number of 87 cities adopted manager government between 1913-1918, another 153 between 1918-1923, and 84 more between 1923-1928 (Judd and Swanstrom 2010).
14
ACCEPTED MANUSCRIPT business interests, weak unions, high population mobility, small immigrant population, and small city size. As these factors are likely to persist and independently a¤ect policy today, empirically identifying the policy e¤ects of manager government seems challenging.
T
Manager government, however, also lled a growing need for technical expertise at the
RI P
top of city government, a need felt more acutely in times of crisis. To mitigate the e¤ects of natural disasters the new technologically-intensive public infrastructure, e.g., levees and bridges, had to be e¤ectively managed. The city of Dayton, Ohio provides an illustration.
SC
In March 1913 after days of heavy rainfall the Great Miami River overowed the citys levees causing a ood that destroyed over 20,000 homes. In the immediate aftermath local leaders
NU
sought to rebuild the ood control system with a large public works campaign. Expediency dictated the adoption of manager government so that an engineer could be appointed to lead the reconstruction e¤ort.
MA
Subsequent crises such as the Great Mississippi Flood of 1927 and the Northeast Flood of 1936 resulted in substantial losses across multiple local jurisdictions and helped swing the
ED
balance toward federal takeover of ood control from local authorities. In 1936 Congress passed the Flood Control Act (FCA) that moved responsibility for ood prevention and management to the Army Corps of Engineers. The hundreds of miles of levees and 375 major
PT
reservoirs constructed by the Army Corps of Engineers after 1936 signi cantly weakened the link between heavy precipitation and the incidence of oods.
CE
That adopting a manager government had been a local response to ood hazards before 1936 suggests an IV identi cation strategy: using precipitation shocks during the local ood control era (1900-1936) to isolate exogenous variation in manager government.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Data. Our sample consists of all U.S. cities with year-1900 Census population over 17,500 residents. This sample selection criterion is not a¤ected by government form choice, since manager reforms do not occur until the small city of Staunton, Virginia (pop. 7,289 in 1900) starts experimenting with it in 1908. After dropping Washington DC, since it has a federally-appointed city government until 1973, a number of 248 present-day cities satisfy this criterion. The sample period for our panel is 1960-2000.14 Our government form data comes from ICMAs Municipal Year Book. We point out two features of government form variation in our sample. First, the maps in Figures 1 and 2 display little geographic clustering in manager government. Second, Figure 3 shows that most manager charter adoptions in our sample occur before 1960. In fact, the majority of 14
Two intervening annexations and one merger slightly alter the sample: Pittsburgh, PA annexed Allegheny, PA in 1907; Omaha, NE annexed South Omaha, NE in 1915; and West Hoboken, NJ merged with Union Hill, NJ to form Union City, NJ in 1925.
15
ACCEPTED MANUSCRIPT adoptions occurred before the federal takeover of ood control in 1936, and only ten occurred after 1960. We measure local policymaking behavior using city employment during 1960-2000. This
T
policy area has the attractive feature that it allows us to distinguish popular from neu-
RI P
tral policy issues by disaggregating police employment into o¢cer and civilian employees, according to the distinction made in the FBIs Uniform Crime Reports.15 One may wonder how much inuence city managers have over the hiring of city employees.
SC
One way to examine this question is by looking at city managers "de jure" personnel powers as established in the charters of manager cities. Another approach is to gauge city managers
NU
"de facto" powers over city employment by estimating manager xed e¤ects. In terms of "de jure" powers, a standard handbook of "model" city charters prescribes under the section "Powers and Duties of the City Manager" that "The city manager shall
MA
appoint and, when necessary for the good of the service, suspend or remove all city employees and appointive administrative o¢cers provided for, by or under this charter, except as
ED
otherwise provided by law, this charter or personnel rules adopted pursuant to this charter. The city manager may authorize any administrative o¢cer subject to the managers direction and supervision to exercise these powers with respect to subordinates in that o¢cers
PT
department, o¢ce or agency." (Kemp 2003, p. 33) (our emphases added). In terms of "de facto" powers, Online Appendix Table A1 shows estimates of manager xed e¤ects for police
CE
and non-police city employees in the manager cities in our sample, using the methodology of Bertrand and Schoar (2003). If city managers have di¤erent employment policy stances and are e¤ective at implementing them, then we should nd that manager xed e¤ects capture
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
statistically signi cant variation in employment. Indeed, for all three employment categories manager xed e¤ects are highly signi cant, producing increases in the adjusted R2 for police o¢cers as high as 7:8 percentage points. We construct our instrument using 1900-2000 weather reports from the U.S. Historical Climatology Networks Daily Temperature, Precipitation, and Snow Data. This dataset contains daily readings for temperature extremes, rainfall, and snowfall collected from weather stations throughout the U.S. We match a city to the closest weather station based on geographic coordinates.16 Our main precipitation measure is the yearly sum of rainfall and
15
Census of Governments city employment data has the advantage that it distinguishes between full-time and part-time city employees, but distinguishes between police o¢cers and civilians only starting in 1977. 16 The U.S. has 126 weather stations reporting in 1900. The median distance to the closest weather station for our sample cities is 47 miles. As the opening of new stations could be related to changes in local weather or local economy we keep every city matched to the same station over time.
16
ACCEPTED MANUSCRIPT density-adjusted snowfall at the station level.17 We de ne an extreme precipitation event as a year when precipitation exceeds the 99th percentile of the national 20th century yearly precipitation distribution. Our cross-sectional measure of precipitation shocks for a given
T
city is the frequency of extreme precipitation events in a given period, referred to below (1900-1936) are referred to as LFC precipitation shocks.
RI P
as precipitation shocks. For instance, precipitation shocks in the local ood control period In addition to these key variables, we analyzed an extensive set of geographic, demo-
SC
graphic, economic, and crime variables. The Data Appendix (see Online Appendix A.6) provides the complete list of variables, with details about their measurement and sources.
NU
Table 1 reports descriptive statistics for our major variables by government form. Panel A shows that manager cities employ on average about 21% fewer o¢cers per capita than mayor cities, and virtually the same number of civilians per capita. Interestingly, Panel B
MA
suggests that before the advent of manager government there were no major di¤erences in either type of police employment. Other statistically signi cant di¤erences are: manager
ED
cities have on average higher reported crime, higher fraction of cleared crime (Panel A), higher incidence of Progressive institutions and Republican mayors (most manager cities maintain an honorary mayor position), lower non-police employment per capita (Panel C),
PT
and higher fraction college graduates (Panel D). Figure 4 provides a rst look at how manager government a¤ected police employment
CE
historically. Manager governments employ fewer police o¢cers from 1960 onward, but not in 1900. In Panel B we divide cities based on whether they experienced LFC precipitation shocks and nd that cities hit by these shocks have lower o¢cer employment after 1960, with
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
the di¤erence attenuating over time. Managers and Popular Policy. We rst estimate a simple model of the form: log(Employmenti;t ) = 1 M anageri;t + 02 xi;t +
t
+ i;t
(1)
where Employmenti;t is city employees per capita for city i in year t, M anageri;t is a dummy variable indicating manager form of city government, xi;t is a vector of controls,
t s
are year
xed e¤ects, and i;t is the error term. The coe¢cient 1 measures the conditional percentage di¤erence in mean employment between manager and mayor cities. According to hypothesis (H1) we expect 1 < 0 for police o¢cer employment and 1 = 0 for neutral employment categories. 17 We adjust snowfall for water density by dividing it by ten, as suggested by the U.S. Department of Agriculture. See http://www.ak.nrcs.usda.gov/snow/data.
17
ACCEPTED MANUSCRIPT This model allows us to account for measurable city geographic, demographic, economic, and political characteristics, and to control for national trends a¤ecting local police employment. As government form changes infrequently during 1960-2000 and the observations are
T
unlikely to be independent within a city we cluster the standard errors at the city level.
RI P
Table 2 presents OLS estimates of the o¢cer employment di¤erential. Accounting for year and division xed e¤ects (the U.S. has nine Census divisions)18 column (1) shows that manager governments employ 11:9% fewer o¢cers per capita. Adding demographic
SC
controls to account for voter preferences in column (2) shows that the point estimate remains negative and highly statistically signi cant.19 Manager government may be related to other
NU
institutional and political factors that could a¤ect policy. The descriptive statistics in Table 1 reect the historical fact that manager reform often came on the heels of other urban reforms: non-partisan ballots, at-large elections, and civil service rules. The literature has
MA
sometimes packaged these other reforms together with manager reform, distinguishing only between "traditional" and "reformed" cities (Stucky 2005). Manager cities may also di¤er
ED
in media penetration and police unionization, which may a¤ect the voter expertise () and policymaker congruence ( P ). Adding these institutional and political controls in column (3) does little to alter the estimate. Finally, in column (4) manager remains negatively
PT
correlated with o¢cer employment when controlling for city spending. In our theory model hierarchical accountability performs an informational role in the
CE
sense that it more often allows the policymaker to act on his private expertise. It is then useful to compare the magnitudes in Table 2 with impacts of information on government policy estimated in prior work. For instance, Strombergs (2004) estimated 0.201 elasticity
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
of federal unemployment relief spending with respect to radio penetration (see his Table II column IV) would imply a 20.1% upper bound, in absolute value, on the o¢cer di¤erential, corresponding to a change from zero radio penetration to full penetration. All the estimates of the o¢cer di¤erential in Table 2 are within this range. Endogeneity. Even if the OLS speci cations above control for all relevant confounds, estimates may still be biased by reverse causality. Despite the infrequency of actual changes in accountability form (e.g., 0.8% of 1,420 U.S. cities had changed their government form between 1980-90 according to Table 1 in Baqir 2002) city charters are endogenous by virtue 18
The U.S. Census divisions are: New England, North Atlantic, South Atlantic, East North Central, East South Central, West North Central, West South Central, Mountain, and Paci c. 19 Voter or special interest preferences can a¤ect institutional choice. Morelli and Van Weelden (2013) argue that pandering can depend on how divisive an issue is for the electorate and Aghion, Alesina, and Trebbi (2004) discuss how the distribution of voter preferences may a¤ect the choice of institutions.
18
ACCEPTED MANUSCRIPT of being subject to revision by popular referendum. Measurement error is another potential source of bias in OLS estimates leading to attenuation. Because manager cities typically maintain an honorary mayor, city clerks in these cities sometimes mistakenly report a mayor
T
form of government on ICMA survey forms (see Coate and Knight 2011).
RI P
To address concerns with bias in the OLS estimates we develop an IV approach. Our strategy is based on public infrastructure crises triggering early 20th century switches to manager government to facilitate the ascension of engineers into the top executive o¢ce. Floods
SC
caused by extreme precipitation before the 1936 Flood Control Act were one such relevant crisis; see Historical Background above. We thus instrument for manager government using
NU
the frequency of extreme precipitation events in the local ood control era (1900-1936), in short, LFC precipitation shocks. Formally, the rst stage model is:
MA
M anageri;t = 1 LF C_P recipitation_Shocksi + 02 wi;t +
t
+ i;t
(2)
where LF C_P recipitation_Shocksi is the frequency of precipitation shocks in the local
ED
ood control era for city i, wi;t is a vector of controls,
t s
are year xed e¤ects, and i;t is
the error term. The controls include Century_P recipitation_Shocksi , the frequency of city
PT
precipitation shocks during the 20th century, and M edian_P recipitationi , median annual city precipitation during the 20th century. These variables help make the exclusion restriction
CE
credible as previous research found that climate a¤ects economic growth (Dell, Jones, and Olken 2012), crime (Jacob, Lefgren, and Moretti 2007), conict (Miguel, Satyanath, and Sergenti 2004), and the origins of trust (Durante 2010).20
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Because we control for century precipitation shocks, the identi cation comes from the timing of precipitation shocks (pre vs. post 1936) and not simply their occurence. Our identi cation assumptions are that, conditional on typical local precipitation patterns: (i) cities hit by precipitation shocks in the LFC era have the same average unobserved characteristics as spared cities, and (ii) LFC precipitation shocks a¤ect police employment decades later only through their e¤ect on government form. In the Online Appendix we provide supportive evidence for these conditions. In support of instrument validity and the exclusion restriction we nd that: (i) LFC precipitation shocks are not correlated with early city characteristics (Online Appendix Table A2 Panel A); (ii) LFC precipitation shocks are not related to other present-day city institutions (Online Appendix Table A2 Panel B), and (iii) in contrast to 20
In a related IV strategy using country-level data Bruckner and Ciccone (2011) exploit the fact that in non-democratic societies the cost of popular opposition to an authoritarian regime is lower during times of economic distress, making negative rainfall shocks correlated with democracy.
19
ACCEPTED MANUSCRIPT LFC precipitation shocks, federal ood control (FFC) precipitation shocks (1936-1960) are not related to police o¢cer employment today (Online Appendix Table A3). Trends in city managers educational backgrounds provide additional support for our
T
identi cation strategy. Progressive Era city managers were disproportionately engineers.
RI P
Later in the century, however, the manager profession bacame dominated by individuals trained in public administration (Stillman 1977). More than 95 percent of city managers were engineers in 1918, 77 percent in 1934, and only 18 percent in 1977. In other words, LFC
SC
precipitation shocks are related to manager government for reasons largely obsolete today. Table 3 presents IV estimates of the o¢cer employment di¤erential controlling for typical
NU
local precipitation patterns, together with OLS counterparts. Columns (1),(2) control for year xed e¤ects and typical precipitation patterns. Columns (3),(4) add water proximity controls, e.g., swamps, rivers, to account for the fact that residents of cities close to water
MA
may implement private solutions to ood risk. Columns (5),(6) add division xed e¤ects. The rst stage shows a strong relationship between LFC precipitation shocks and present-
ED
day manager government: the F -statistic always exceeds the critical value of 10 below which nite-sample weak-instrument bias could be a concern (Bound, Jaeger, and Baker 1995). In the second stage we nd signi cantly lower o¢cer employment in manager cities. The larger
PT
IV point estimates relative to the baseline OLS results suggest that measurement error in government form might be present in our sample (Coate and Knight 2011); another factor
CE
could be unobserved voter preferences, e.g., more conservative cities are both more likely to adopt manager government and have a tougher stance on crime by demanding more police. Overall the IV results uphold the substantive conclusions derived from the OLS estimates.21
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
If the costs of changing government form in response to precipitation shocks are heterogeneous, the identi ed e¤ects will be local to the subset of responsive cities and potentially di¤erent from the population-wide treatment e¤ect. For example, Acemoglu, Robinson, and Torvik (2013) argue that voters dismantle exogenously imposed checks and balances when special interests are strong. One way to explore this issue is to examine the sensitivity of the IV estimates to changes in instrument construction. If the e¤ects of government form are heterogeneous and highly speci c to cities induced to change government form by our proposed instrument, alternative instrument de nitions would likely change the magnitudes of the estimated e¤ects. Online Appendix Tables A4 and A5 show that the IV results are by and large robust to altering the instrument de nition along several dimensions, strengthening their external validity. 21 Baqir (2002) and Whalley (2013) nd little evidence that related forms of institutional endogeneity contaminate OLS results in the context of U.S. local governments.
20
ACCEPTED MANUSCRIPT Using a di¤erent sample, a di¤erent period, and exploiting switches to manager government Coate and Knight (2011) nd that managers spend more than mayors. We nd that manager governments employ fewer police o¢cers. To reconcile these seemingly incongruous
T
results we examine total city government spending and police department spending data in
RI P
Online Appendix Table A6. Columns (1),(2) replicate Coate and Knights (2011) nding in our sample; accounting for endogeneity in column (3) weakens the manager-mayor total spending di¤erence although the point estimate remains positive and sizable. Columns
SC
(4),(5) imply little di¤erence in police spending per capita while column (6) shows smaller police spending by managers, with the estimate not statistically signi cant.
NU
Online Appendix Table A7 reports sensitivity checks for our preferred IV speci cation and two related OLS speci cations. We examine robustness to sample alterations, alternative inference procedures, and alternative dependent variable scaling; the estimates seem quite
MA
robust and inferences from both OLS and IV continue to hold. As one might expect, the manager e¤ects are somewhat larger in bigger cities, model (2), and prior to the federal
ED
intervention in local policing through the COPS program, model (4). Our hypothesized causal mechanism from LFC precipitation shocks to government form runs through ood-related infrastructure crises, yet direct measures of ood damage or ood
PT
intensity at the beginning of the twentieth century are sparse due to local management of ood control before 1936. We nevertheless provide two ways to make this connection. The
CE
rst uses a cross-sectional soil-based measure of ood incidence developed by U.S. Department of Agriculture scientists. The second uses a time-varying measure of ooding based on riverow readings as monitored by the U.S. Geological Service. We develop alternative
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
measures of precipitation shocks, based on di¤erent critical volume thresholds and di¤erent time aggregation, in order to check which ones are more tightly linked with these ooding proxies. The results are in Online Appendix Tables A8 and A9. Overall, we nd signi cant correlations between extreme precipitation and ooding. While there are tradeo¤s between the di¤erent precipitation measures, the one based on the top 1% volume threshold and annual aggregation seems to perform best with our yearly dataset. Other Employment Categories. Table 4 reports OLS and IV estimates of managermayor di¤erences in city employment other than police o¢cers. The rst three columns look at civilians hired in the police department, e.g., dispatchers, administrators, forensic sta¤. This can be considered a neutral policy issue since voters likely have no clear ex ante preference over high or low police civilian employment. The OLS estimates are positive and closer to zero than the negative estimates in the police o¢cer regressions, and not 21
ACCEPTED MANUSCRIPT statistically signi cant. The IV estimate is also positive, and somewhat larger, possibly reecting a substitution response between o¢cer and civilian employment, although still not signi cantly di¤erent from zero. While the lower number of o¢cers in manager cities may
T
be consistent with managers weaker patronage motivations (Enikolopov 2014) the lack of a
RI P
similar pattern for civilians seems to indicate that other types of incentives also matter. The last three columns consider non-police employees, e.g., re ghters, sanitation workers, assessors. We note that for 65.7% of our sample cities public education is provided
SC
through independent school districts; these cities have little direct control over teacher employment which falls under the jurisdiction of the school board and the school districts
NU
superintendent. The OLS estimates for non-police employees are similar in magnitude to the ones for o¢cers in Table 2, although less statistically signi cant; the IV estimate is substantially smaller. These negative point estimates may reect weaker pandering incentives
MA
for some of these employment categories, or may be consistent with a patronage explanation; we explore this issue further below in Table 5 and Online Appendix Table A10.
ED
Policy Volatility. In the insulation equilibrium of our model the manager is free to adjust the policy to match the stochastic state. The mayor, in contrast, is politically constrained by
PT
public pressure to pick the popular policy regardless of the state. This implies that popular policies should be more volatile in manager cities (hypothesis H2). Table 5 reports OLS and
CE
IV estimates of manager-mayor di¤erences in policy volatility.22 In columns (1)-(3) we nd a positive and statistically signi cant di¤erence in volatility for the popular policy, namely police o¢cer employment per capita. Columns (4)-(6) show that
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
the managers higher propensity to adjust o¢cers is not due to a higher crime rate volatility in manager cities. As we assume that the probability of crime (latent crime) is exogenous to government form, the di¤erences in committed crime (actual crime) volatility estimated here suggest that police employment a¤ects committed crime di¤erentially by government form. In the bottom half of the table we nd no signi cant di¤erences in volatility for police civilians per capita or non-police employees per capita. The estimates have for the most part a negative sign, opposite to the pattern for o¢cer volatility, and are closer to zero. The contrasting coe¢cient patterns for popular vs. neutral policy issues provide support for the types of incentives highlighted by our model. 22
volatility in a time-series variable yi;t as the log of relative deviations from mean, We measure PT 1 PT 1 log yi;t T t=1 yi;t = T t=1 yi;t . As extreme outliers may represent measurement error we trim the sample by removing the top and bottom 1% extreme values. For ease of comparison we use the 1975-2000 period when crime data is available.
22
ACCEPTED MANUSCRIPT Informational Mechanism. In our theory model a popular policy di¤erential strictly speaking only exists for low-voter-expertise high-policymaker-congruence cities where voters can credibly commit to insulate the policymaker. In Table 6 we examine (H3) which states
T
that the policy di¤erential for popular issues should decrease in the voters policy expertise.
RI P
We use the "crack epidemic" that started to a¤ect U.S. cities in the early 1980s as a shock to voter expertise measured by the voters crime probability prior (). The emergence of crack cocaine and associated media attention increased the match between the popular
SC
policy of high police and the optimal policy, as it heightened public awareness of crime in urban areas.23 Importantly for identi cation the geographic spread of crack was driven
NU
by proximity to a few entry points and other topographic conditions, rather than by local enforcement activities (Evans, Garthwaite, and Moore 2012). Based on Proposition 1 we expect the increase in the voters crime probability prior due to the crack epidemic to produce
MA
a manager pandering equilibrium, leading to a reduction in the negative manager-mayor o¢cer di¤erential toward zero. In Table 6 we add to equation (1) an interaction of M anageri;t
ED
with this proxy for the informational mechanism. The results in columns (1)-(3) show that the crack epidemic shrinks the negative o¢cer di¤erential, consistent with a pandering
PT
equilibrium in manager cities with high voter expertise. Electoral E¤ects. Our incentives model predicts that pandering is strongest pre-election
CE
because the voter rationally relies more heavily on recent information about policy choices. We measure election proximity by coding the year before the election date as an election year.24 To estimate electoral e¤ects we add to equation (1) the election indicator
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Electioni;t and the interaction M anageri;t Electioni;t , as well as city xed e¤ects. In this model M anager measures the manager-mayor policy di¤erential in non-election years, while M anager + M anagerElection measures the manager-mayor policy di¤erential in election years. Hypothesis (H4) states that the election year employment di¤erential is more negative than the non-election year di¤erential for popular issues, meaning M anagerElection < 0 for o¢cers. Table 7 column (1) shows that the interaction coe¢cient is negative, consistent with (H4), yet somewhat statistically weak. To increase precision we notice from the robustness 23
See Blenko (1993). Economists and other social scientists have provided evidence that the crack epidemic increased crime. For example, Grogger and Willis (2000) conclude that "Our results indicate that, in the absence of crack cocaine, the crime rate in 1991 would have remained below its previous peak in the early 1980s." (p. 528). 24 Election dates are set in the city charter; special elections outside the regular election cycle, say to replace a mayor resigning before the end of the term, are infrequent and we ignore them. For manager government we use mayoral-council elections. We cross-checked elections data against multiple sources to resolve inconsistencies we found among sources.
23
ACCEPTED MANUSCRIPT analysis, Online Appendix Table A7 Model (2), that mean policy di¤erentials are greater in larger cities.25 Columns (2)-(4) therefore restrict the sample to the subset of 174 cities with population over 50,000 in 1960. The negative interaction coe¢cient from column (1)
T
reappears in columns (2)-(3) and is now also statistically signi cant at conventional levels.
RI P
The employment cycle could be confounded by a spending cycle although in Online Appendix Table A10 we nd neither a di¤erential spending cycle nor a di¤erential employment cycle in other employment categories. The further addition of government spending in colThe election-year di¤erential is
0:008
0:013; or
SC
umn (4) leaves the estimate of M anagerElection and its standard error virtually unchanged. 0:8 percentage points and statistically
NU
negative, while the non-election-year di¤erential is
2:1 percentage points, and statistically
zero. Note also that the election-year di¤erential is not driven by fewer election-year o¢cers in manager cities, but by more election-year o¢cers in mayor cities, namely 0:9% more.
MA
These results are consistent with the predictions of our incentives model as opposed to a selection model by preference type which would predict a positive election-year di¤erential
Conclusion
PT
4
ED
(see Section 2 and Online Appendix A.4).
This paper sheds light on the incentive e¤ects of an accountability form that is pervasive at
CE
various levels of governments. Our principal-intermediary-agent model gives informational asymmetries a central role and shows that hierarchical accountability creates either insulation or pandering incentives depending on informational and political factors. It implies
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
that in a setting with government-voter expertise asymmetries this structure rather than hindering democratic accountability may in fact increase the exibility of voter control over policymaking incentives, by reducing pandering when it is detrimental to voter welfare. The theory guides our empirical analysis of U.S. city managers in a salient area of local policy, namely public safety. The estimates for the main institutional e¤ect and its mechanisms are consistent with the theoretically characterized incentives. For simplicity we assumed that informational asymmetries are exogenous to the institutional environment, yet motivations for acquiring information by the voter or the policymaker (Stephenson 2011) may depend on the accountability form: on one hand, the voter might be less interested in acquiring information if he cannot directly remove the policymaker; on the 25
Levitt (1997) also argues that electoral cycles should be more pronounced in larger cities. He exploits the correlation between election years and police hiring in a panel of 59 large U.S. cities to identify the e¤ect of police on crime.
24
ACCEPTED MANUSCRIPT other hand, a hierarchically-accountable policymaker, who enjoys more policy discretion, may acquire more policy information. Thus, the informational asymmetry may be more severe under hierarchical accountability. Empirically, a more stringent test of the model
T
would be to verify that voter and intermediary strategies, in addition to policy outcomes,
RI P
adjust to the informational and political environment as shown here to provide optimal incentives for a hierarchically-accountable policymaker. This requires more detailed data on council elections and manager retention (see Enikolopov 2014 for an e¤ort in this direction).
SC
Finally, one could apply our framework to study the incentives of indirectly-accountable policymakers with xed terms. We conjecture that in that setting voters have less inuence
CE
PT
ED
MA
NU
over policymaker incentives since the retention signal would be weaker.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
25
ACCEPTED MANUSCRIPT FIGURE 1: Manager Cities, 1960
MA
NU
SC
RI P
T
Notes: Authors’ calculations for the sample of the 248 largest self-governing cities as of 1900. Manager government locations are plotted based on 1960 form of government. The map reflects state jurisdictional boundaries.
CE
PT
ED
FIGURE 2: Mayor Cities, 1960
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Notes: Authors’ calculations for the sample of the 248 largest self-governing cities as of 1900. Mayor government locations are plotted based on 1960 form of government. The map reflects state jurisdictional boundaries.
26
ACCEPTED MANUSCRIPT FIGURE 3: Timing of Manager Charter Adoptions
1900
NU
0
1
SC
Frequency 2 3
RI P
T
4
5
1920
1940
1960
1980
2000
Year
MA
Notes: Authors’ tabulations using adoption data from the Municipal Year Book, newspapers, and city charters for the sample of the 248 largest selfgoverning cities as of 1900. Year of manager government adoption is based on year of charter approval by voters.
PT
ED
FIGURE 4: Police Employment, by Indicated Cohort
CE
Panel B: Officers Per Capita; by LFC Precipitation Shocks .4
1
Panel A: Officers Per Capita; by Government Form
.2 Mean Residuals -.2 0 -.6
1900
1920
1940
1960
1980
2000
1920
1940
1960
1980
2000
Notes: Authors’ calculations for the sample of the 248 largest self-governing cities as of 1900. Panel B plots the mean residual from a regression of log(Police Officers Per Capita) on Century Precipitation Shocks and Median Precipitation.
27
1900
Year
Year
LFC Precipitation Shocks No LFC Precipitation Shocks
-.4
Mean log(Officers Per Capita) .2 .4 .6 .8
AC
Manager 1960 Mayor 1960
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
ACCEPTED MANUSCRIPT TABLE 1: Descriptive Statistics, By Government Form Manager Cities mean std. dev. (1) (2)
Mayor Cities mean std. dev. (3) (4)
mean diff. p-value (5)
0.53 0.24 227 0.43 32.07 0.11
2.25 0.39 198 0.31 81.00 0.25
1.08 0.36 213 0.44 39.27 0.15
0.000 0.890 0.576 0.266 0.003 0.005
1.10 0.06 0.05
0.46 0.06 0.04
0.935 0.074 0.022
0.50 0.43 0.46 0.45 1.88 2,681 12.47
0.000 0.000 0.886 0.004 0.610 0.548 0.040
RI P
1.86 0.38 204 0.37 92.96 0.28
SC
Police Officers per Capita Police Civilians per Capita Police Dept. Spending per Capita Fraction Officers Unionized (1968) Crime Rate (1975-2000) Fraction Crimes Cleared (1975-2000)
T
Panel A: Police Department, 1960-2000
Panel B: Police Department, 1900
Police Officers per Capita Police Civilians per Capita Arrests per Capita
0.34 0.11 0.05
MA
NU
1.10 0.08 0.07
Panel C: City Government, 1960-2000
PT
ED
Non-Partisan Elections (1960) Fraction At-Large Seats (1960) Early Civil Service (1960) Incumbent Mayor Republican Local Newspaper Sales (1990-2000) City Gov. Spending per Cap. Non-Police Empl. per Cap. (1972-2000)
0.85 0.83 0.69 0.47 0.42 2,115 15.62
0.35 0.34 0.46 0.50 1.06 2,584 11.69
0.53 0.55 0.70 0.27 0.51 2,208 18.59
Panel D: City Demographics, 1960-2000
CE
Population Population Density Fraction Non-White Fraction College Graduate Household Income Observations
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
146,198 184,417 5,280 19,744 0.19 0.15 0.15 0.08 28,653 7,477 3,622
252,521 712,626 6,556 8,837 0.20 0.20 0.13 0.08 28,241 7,498 6,546
0.074 0.732 0.900 0.020 0.487 10,168
Notes: Authors’ calculations with 1960-2000 city data, unless otherwise indicated. The unit of observation is a city-year for the sample of 248 largest, as of 1900, self-governing cities in the United States. Each entry in columns (1)-(4) presents the mean/std. dev. for the indicated variable and column (5) has pvalues for the hypothesis that the difference between the means in columns (1) and (3) is zero, using standard errors clustered at the city level. The number of observations reflects the maximum sample size across all the reported variables. All monetary values are expressed in 2000$.
28
ACCEPTED MANUSCRIPT TABLE 2: Government Form and Popular Policy Model:
SC
NU
Year Fixed Effects Division Fixed Effects Demographics Institutional Fiscal Period Cities Observations
RI P
T
Manager
OLS OLS OLS OLS (1) (2) (3) (4) Dependent Variable = log(Police Officers per Capita) í0.119*** í0.083*** í0.069*** í0.076*** (0.033) (0.027) (0.030) (0.029) Yes Yes Yes Yes Yes Yes Yes Yes No Yes Yes Yes No No Yes Yes No No No Yes 1960-2000 1960-2000 1960-2000 1960-2000 248 248 248 248 9,974 9,974 9,974 9,974
CE
PT
ED
MA
Notes: Authors’ calculations with 1960-2000 city data as described in the Data Appendix. The unit of observation is a city-year for the sample of the largest 248 self-governing cities in the United States as of 1900. The table reports OLS estimates of ȕ1 from equation (1) in the text. Standard errors clustered at the city level in parentheses. Demographic controls: Population Density; Fraction Non-White; Fraction College Graduate; Median Household Income. Institutional controls: Incumbent Mayor Republican; NonPartisan Elections 1960; Fraction At-Large Seats 1960; Early Civil Service; Police Unionization 1968; Local Newspaper Sales 1990-2000. Fiscal controls: log(City Government Spending per Capita). Indicator variables for missing values added. * indicates significance at the 10 percent level, ** significance at the 5 percent level, and *** significance at the 1 percent level.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
29
ACCEPTED MANUSCRIPT TABLE 3: Government Form and Popular Policy: Endogeneity
[p-value]
OLS IV OLS IV OLS (2) (3) (4) (5) (6) Dependent Variable = log(Police Officers per Capita) í0.284** í0.143*** í0.270** í0.098*** í0.394* í0.086*** (0.140) (0.035) (0.127) (0.031) (0.207) (0.032) First Stage: Dependent Variable = Manager 11.675*** – 13.676*** – 9.875*** – (2.851) (2.566) (2.676) 16.76 – 28.40 – 13.62 – [0.0001] [0.0003] [0.0000]
Year Fixed Effects Weather Water Division Fixed Effects Period Cities Observations
Yes Yes No No 1960-2000 248 9,974
Model:
Yes Yes No No 1960-2000 248 9,974
Yes Yes Yes No 1960-2000 248 9,974
NU
F-stat
MA
LFC Precip. Shocks
SC
RI P
T
Manager
IV (1)
Yes Yes Yes No 1960-2000 248 9,974
Yes Yes Yes Yes 1960-2000 248 9,974
Yes Yes Yes Yes 1960-2000 248 9,974
CE
PT
ED
Notes: Authors’ calculations with 1960-2000 city data as described in the Data Appendix. The unit of observation is a city-year for the sample of largest 248 self-governing cities in the United States as of 1900. The first panel reports estimates of ȕ1 in equation (1) in the text. The second panel reports estimates of D1 in equation (2) in the text. Standard errors clustered at the city level in parentheses. Fstatistic corresponds to the excluded instrument LFC Precipitation Shocks with p-values in brackets. Weather controls: Century Precipitation Shocks; Median Precipitation. Water controls: number of Lakes; number of Swamps; one indicator for location on Coast; three indicators for Small Rivers, Medium Rivers, Large Rivers. * indicates significance at the 10 percent level, ** significance at the 5 percent level, and *** significance at the 1 percent level.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
30
ACCEPTED MANUSCRIPT TABLE 4: Government Form and Other Employment
NU
Year Fixed Effects Weather Water Division Fixed Effects Period Cities Observations
T
Manager
OLS OLS IV (4) (5) (6) log(Non-Police Empl. Per Capita) í0.083 í0.104* í0.036 (0.068) (0.056) (0.361) Yes Yes Yes Yes Yes Yes Yes Yes Yes No Yes No 1972-2000 1972-2000 1972-2000 246 246 246 6,702 6,702 6,702
RI P
Dep. Var. =
OLS OLS IV (1) (2) (3) log(Police Civilians per Capita) 0.082 0.015 0.162 (0.070) (0.070) (0.253) Yes Yes Yes Yes Yes Yes Yes Yes Yes No Yes No 1960-2000 1960-2000 1960-2000 248 248 248 9,850 9,850 9,850
SC
Model:
CE
PT
ED
MA
Notes: Authors’ calculations with city data as described in the Data Appendix. The unit of observation is a city-year for the sample of the largest 248 self-governing cities in the United States as of 1900. Columns (1)-(6) report estimates of ȕ1 from equation (1) in the text. Standard errors clustered at the city level reported in parentheses. Weather controls: Century Precipitation Shocks; Median Precipitation. Water controls: number of Lakes; number of Swamps; one indicator for location on Coast; three indicators for Small Rivers, Medium Rivers, Large Rivers. * indicates significance at the 10 percent level, ** significance at the 5 percent level, and *** significance at the 1 percent level.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
31
ACCEPTED MANUSCRIPT TABLE 5: Government Form and Volatility
Model: Dep. Var. =
Manager Cities Observations Year Fixed Effects Weather Water Division Fixed Effects Period
OLS OLS IV (7) (8) (9) log(Police Civil. Variation Coef.) –0.112 –0.088 –0.325 (0.076) (0.083) (0.259) 247 247 247 6,310 6,310 6,310 Yes Yes Yes Yes Yes Yes Yes Yes Yes No Yes No 1975-2000 1975-2000 1975-2000
T
RI P
Cities Observations
OLS OLS IV (4) (5) (6) log(Crime Rate Variation Coef.) í0.147*** í0.149*** í0.455** (0.056) (0.056) (0.205) 248 248 248 6,024 6,024 6,024 OLS OLS IV (10) (11) (12) log(Non-Police Variation Coef.) í0.090 í0.071 0.002 (0.097) (0.090) (0.270) 246 246 246 5,850 5,850 5,850 Yes Yes Yes Yes Yes Yes Yes Yes Yes No Yes No 1975-2000 1975-2000 1975-2000
SC
Manager
NU
Dep. Var. =
OLS OLS IV (1) (2) (3) log(Police Officer Variation Coef.) 0.210*** 0.148** 0.507* (0.070) (0.067) (0.288) 246 246 246 6,336 6,336 6,336
MA
Model:
CE
PT
ED
Notes: Authors’ calculations with 1975-2000 city data as described in the Data Appendix. The unit of observation is a city-year. The sample is of the largest 248 self-governing cities in the United States as of 1900. Columns (1)-(6) report estimates of ȕ1 from equation (1) in the text. Standard errors clustered at the city level reported in parentheses. The dependent variable is calculated as log(|yit – average[yit]|/average[yit]), for the yit observations between the bottom 1 percentile and the top 1 percentile of its sampling distribution. Weather controls: Century Precipitation Shocks; Median Precipitation. Water controls: number of Lakes; number of Swamps; one indicator for location on Coast; three indicators for Small Rivers, Medium Rivers, Large Rivers. * indicates significance at the 10 percent level, ** significance at the 5 percent level, and *** significance at the 1 percent level.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
32
ACCEPTED MANUSCRIPT TABLE 6: Government Form and Popular Policy: Mechanisms OLS (1)
Model: Dep. Variable =
OLS (3)
RI P
T
Manager
OLS (2)
log(Police Officers per Capita) –0.096*** –0.087*** –0.095*** (0.026) (0.031) (0.031) 0.041* 0.047** 0.050** (0.023) (0.023) (0.023) –0.030 –0.036 –0.034 (0.028) (0.035) (0.035) Yes Yes Yes Yes Yes Yes Yes Yes Yes No Yes Yes No No Yes 1960-2000 1960-2000 1960-2000 248 248 248 9,974 9,974 9,974
Manager × Crack Epidemic
SC
Crack Epidemic
MA
NU
Year Fixed Effects Division Fixed Effects Demographics Institutional Fiscal Period Cities Observations
CE
PT
ED
Notes: Authors’ calculations with 1960-2000 city data as described in the Data Appendix. The unit of observation is a city-year for the sample of largest 248 self-governing cities in the United States as of 1900. Columns (1)-(3) report OLS estimates of equation (1) in the text augmented with the indicated variables and interaction terms. Standard errors clustered at the city level reported in parentheses. Demographic controls: Population Density; Fraction Non-White; Fraction College Graduate; Median Household Income. Institutional controls: Incumbent Mayor Republican; Non-Partisan Elections 1960; Fraction At-Large Seats 1960; Early Civil Service; Police Unionization 1968; Local Newspaper Sales 1990-2000. Fiscal controls: log(City Government Spending per Capita). Indicator variables for missing values added. * indicates significance at the 10 percent level, ** significance at the 5 percent level, and *** significance at the 1 percent level.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
33
ACCEPTED MANUSCRIPT TABLE 7: Government Form and Popular Policy: Electoral Effects Model: Dep. Variable =
OLS (2)
OLS (3)
OLS (4)
log(Police Officers per Capita) –0.078 –0.064 –0.045 –0.008 (0.056) (0.084) (0.048) (0.057) –0.009* –0.015** –0.013*** –0.013*** (0.005) (0.007) (0.005) (0.005) 0.006 0.010** 0.009** 0.009** (0.004) (0.004) (0.004) (0.004) – – – –0.005 (0.007) – – – 0.005 (0.004) Yes Yes Yes Yes Yes Yes Yes Yes No No Yes Yes 1960-2000 1960-2000 1960-2000 1960-2000 248 174 174 174 9,974 7,033 7,033 7,033 All Above 50k Above 50k Above 50k
Manager × Election Election Manager × log(City Gov. Spend. per Cap.)
NU
log(City Gov. Spend. per Cap.)
SC
RI P
T
Manager
OLS (1)
MA
Year Fixed Effects City Fixed Effects Demographics Period Cities Observations 1960 City Population
CE
PT
ED
Notes: Authors’ calculations with 1960-2000 city data as described in the Data Appendix. The unit of observation is a city-year. In column (1) the sample is the largest 248 self-governing cities in the United States as of 1900. In the remaining columns the sample is the 174 cities that also have at least 50,000 residents in 1960. The table reports OLS estimates of equation (1) in the text augmented with the indicated variables and interaction terms. Standard errors clustered at the city level reported in parentheses. Demographic controls: Population Density; Fraction Non-White; Fraction College Graduate; Median Household Income. Indicator variables for missing values added. * indicates significance at the 10 percent level, ** significance at the 5 percent level, and *** significance at the 1 percent level.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
34
ACCEPTED MANUSCRIPT References Acemoglu, D., J. Robinson, and R. Torvik (2013) "Why Do Voters Dismantle Checks and Balances?" Review of Economic Studies 80(3): 845-875.
T
Acemoglu, D., G. Egorov, and K. Sonin (2012) "A Political Theory of Populism," Quar-
RI P
terly Journal of Economics 128(2): 771-805.
Aghion, P., A. Alesina, and F. Trebbi (2004) "Endogenous Political Institutions," Quar-
SC
terly Journal of Economics 119(2): 565-611.
Aruoba, B., A. Drazen, and R. Vlaicu (2015) "A Structural Model of Electoral Accountability," NBER working paper 21151.
NU
Ashworth, S. and K.W. Shotts (2010) "Does Informative Media Commentary Reduce Politicians Incentives to Pander?" Journal of Public Economics 94(11-12): 838-847.
MA
Baqir, R. (2002) "Districting and Government Overspending," Journal of Political Economy 110(6): 1318-1354.
Bertrand, M. and A. Schoar (2003) "Managing with Style: The E¤ect of Managers on
ED
Firm Policies," Quarterly Journal of Economics 118 (4): 1169-1208. Besley, T. and M. Smart (2007) "Fiscal Restraints and Voter Welfare," Journal of Public
PT
Economics 91(3-4): 755-773.
Blenko, S. (1993) Crack and the Evolution of Anti-Drug Policy, Westport, CT: Green-
CE
wood Press.
Bound, J., D.A. Jaeger, and R.M. Baker (1995) "Problems with Instrumental Variables Estimation when the Correlation Between the Instruments and the Endogenous Explanatory
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Variable is Weak," Journal of the American Statistical Association 90(430): 443-450. Bruckner, M. and A. Ciccone (2011) "Rain and the Democratic Window of Opportunity," Econometrica 79(3): 923-947. Canes-Wrone, B., M. Herron, and K. Shotts (2001) "Leadership and Pandering: A Theory of Executive Policymaking," American Journal of Political Science 45(3): 532-550. Canes-Wrone, B. and K. Shotts (2004) "The Conditional Nature of Presidential Responsiveness to Public Opinion," American Journal of Political Science 48(4): 690-706. Coate, S. and B. Knight (2011) "Government Form and Public Spending: Theory and Evidence from U.S. Municipalities," AEJ: Economic Policy 3(3): 82-112. Dell, M., B. Jones, and B. Olken (2008) "Climate Shocks and Economic Growth: Evidence from the Last Half Century," AEJ: Macroeconomics 4(3): 66-95. Durante, R. (2010) "Risk, Cooperation, and the Economic Origins of Social Trust: An Empirical Investigation," working paper, Sciences Po. 35
ACCEPTED MANUSCRIPT Enikolopov, R. (2014) "Politicians, Bureaucrats, and Targeted Redistribution," Journal of Public Economics 120: 74-83. Evans, W., C. Garthwaite, and T.J. Moore (2012) "The White/Black Educational Gap,
T
Stalled Progress, and the Long Term Consequences of the Emergence of Crack Cocaine
RI P
Markets," NBER working paper 18437.
Fox, J. (2007) "Government Transparency and Policymaking," Public Choice 131(1): 23-44.
SC
Gailmard, S. and J. Jenkins (2009) "Agency Problems, the 17th Amendment, and Representation in the Senate," American Journal of Political Science 53(2): 324-342.
NU
Gallup, Inc. (2000) "Crime Tops List of Americans Local Concerns," by Lydia Saad. http://www.gallup.com/poll/2800/crime-tops-list-americans-local-concerns.aspx. Gallup, Inc. (2010) "Americans Still Perceive Crime as on the Rise," by Je¤rey M. Jones.
MA
http://www.gallup.com/poll/144827/americans-perceive-crime-rise.aspx. Gordon, S.C. and G.A. Huber (2007) "The E¤ect of Electoral Competitiveness on In-
ED
cumbent Behavior," Quarterly Journal of Political Science 2(2): 107-138. Grogger, J. and M. Willis (2000) "The Emergence of Crack Cocaine and the Rise in Urban Crime Rates," The Review of Economics and Statistics 82(4): 519-529.
PT
Huber, G.A. and S.C. Gordon (2004) "Accountability and Coercion: Is Justice Blind when It Runs for O¢ce?" American Journal of Political Science 48(2): 247263.
CE
ICMA (2007) Council-Manager Form of Government, Washington, DC: ICMA Press. Jacob, B., L. Lefgren, and E. Moretti (2007) "The Dynamics of Criminal Behavior," Journal of Human Resources 42(3): 489-527.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Kartik, N., F. Squintani, and K. Tinn (2013) "Information Revelation and Pandering in Elections," working paper, Columbia University. Kemp, R.L. (ed.) (2003) Model Government Charters: A City, County, Regional, State, and Federal Handbook, McFarland Press. Knoke, D. (1982) "The Spread of Municipal Reform: Temporal, Spatial, and Social Dynamics," American Journal of Sociology, 87(6): 1314-1339. Levin, J. and S. Tadelis (2010) "Contracting for Government Services: Theory and Evidence from U.S. Cities," Journal of Industrial Economics 58(3): 507-541. Levitt, S. (1997) "Using Electoral Cycles in Police Hiring to Estimate the E¤ect of Police on Crime," American Economic Review 87(3): 270-290. Lim, C. (2013) "Preferences and Incentives of Appointed and Elected Public O¢cials: Evidence from State Trial Court Judges," American Economic Review 103(4): 1360-1397.
36
ACCEPTED MANUSCRIPT MacDonald, L. (2008) "The Impact of Government Structure on Local Public Expenditures," Public Choice 136(3-4): 457-473. Martinez, L. (2009) "A Theory of Political Cycles," Journal of Economic Theory 144(3):
T
11661186.
RI P
Maskin, E. and J. Tirole (2004) "The Politician and the Judge: Accountability in Government," American Economic Review 94(4): 1034-1054.
Miguel, E., S. Satyanath, and E. Sergenti, (2004) "Economic Shocks and Civil Conict:
SC
An Instrumental Variables Approach," Journal of Political Economy 112(4): 725-753. Morelli, M. and R. Van Weelden (2013) "Ideology and Information in Policymaking,"
NU
Journal of Theoretical Politics 25 (3): 412-439.
Park, C. (2000) "Monitoring and Structure of Debt Contracts," Journal of Finance 55(5): 2157-95.
MA
Persson, T., G. Roland and G. Tabellini (1997) "Separation of Powers and Political Accountability," Quarterly Journal of Economics 112: 1163-1202.
ED
Rogo¤, K. (1990) "Equilibrium Political Budget Cycles," American Economic Review 80(1): 21-36.
Shi, M. and Svensson, J. (2006) "Political Budget Cycles: Do They Di¤er Across Coun-
PT
tries and Why?" Journal of Public Economics 90(8-9): 1367-1389. Smart, M. and D. Sturm (2013) "Term Limits and Electoral Accountability," Journal of
CE
Public Economics 107: 93-102.
Stephenson, M. (2011) "Information Acquisition and Institutional Design," Harvard Law Review 124: 1422-1483.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Stillman, R.J. (1977) "The City Manager: Professional Helping Hand, or Political Hired Hand?" Public Administration Review 37(6): 659-670. Strausz, R. (1997) "Delegation of Monitoring in a Principal-Agent Relationship," Review of Economic Studies 64(3): 337-357. Stromberg, D. (2004) "Radios Impact of Public Spending," Quarterly Journal of Economics 119(1): 189-221. Stucky, T.D. (2005) "Local Politics and Police Strength," Justice Quarterly 22: 139-169. Vlaicu, R. and A. Whalley (2011) "Do Housing Bubbles Generate Fiscal Bubbles? Evidence from California Cities," Public Choice 149(1-2): 89-108. Whalley, A. (2013) "Elected Versus Appointed Policymakers: Evidence from City Treasurers," Journal of Law and Economics 56(1): 39-81.
37
0DQXVFULSW &OLFNKHUHWRYLHZOLQNHG5HIHUHQFHV
Online Appendix (Not for Publication) Section A.1 contains formal proofs of Propositions 1-2 in the text. Sections A.2-A.3 provide
T
theory robustness results. Section A.4 solves a selection model. Section A.5 contains a
RI P
discussion of supplementary empirical results. Section A.6 is the Data Appendix containing data definitions and sources, followed by supplementary Tables A1-A10.
SC
A.1 Incentives Model
NU
∞ Assumption 1 Given per-period i.i.d. preference shocks η jt t=1 such that η jt ∈ {B, C} with P η jt = C = γ j , agent j’s type at t + 1 is θjt+1 ∈ η jt , η jt+1 with P θjt+1 = η jt = λ. β[1+β(1−π)] 1−β 2
MA
Assumption 2 The discount factor β satisfies: Proposition 1 For popular policy issues π > P
1 2
> 1.
: (i) Hierarchical accountability either in-
sulates the policymaker pre-election, if γ > π, or creates policymaker pandering incentives
ED
pre-election, otherwise; (ii) Electoral accountability always creates policymaker pandering incentives pre-election; (iii) Agents follow their preferences post-election under both account-
PT
ability forms, e.g., an intermediary retains only policymakers of his own type. Proof of Proposition 1.
We first establish that voter beliefs about agent types are
CE
based on current period actions. To see this, note that at t + 1 incumbent j’s type may be based on either the current or a past preference shock: γ jt (Xt , Yt ) ≡ P θjt+1 = C| (Xt , Yt ) = P θjt+1 = C| (Xt , Yt ) , θjt+1 = η jt+1 (1 − λ)+P θjt+1 = C| (Xt , Yt ) , θjt+1 = θjt = η jt λ (1 − λ)+ P θjt+1 = C| (Xt , Yt ) , θjt+1 = η jt , θjt = η jt−1 λ2 . When θjt+1 = θjt = η jt , earlier observations
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
ACCEPTED MANUSCRIPT
(Xt−1 , Yt−1 ) are generated by preference types η jt−1 and earlier, which are independent of η jt . Thus, P θjt+1 = C| (Xt , Yt ) , θjt+1 = θjt = η jt = P θjt+1 = C| (xt , yt ) , θjt+1 = θjt = η jt . When θjt+1 6= θjt , the period t + 1 type is independent of period t type, since θjt+1 , θjt ∈ j η t+1 , η jt , η jt+1 , η jt−1 , η jt , η jt−1 . Thus, P θjt+1 = C| (Xt , Yt ) , θjt+1 6= θjt = γ j . Using these observations, it follows that γ jt (Xt , Yt ) = γˆ jt (xt , yt ) λ (1 − λ) + γ j [1 − λ (1 − λ)] . One can also write:
γ jt (Xt , Yt ) − γ j = λ (1 − λ) γˆ jt (xt , yt ) − γ j
(1)
Under electoral accountability, i.e., no intermediary and every other period it is the voter who decides whether to reelect the policymaker, the voter reelects the incumbent policymaker at the end of electoral term [t − 1, t] iff γˆ Pt (xt ) ≥ γ P . Thus, at t − 1 the incumbent follows his preferences. Suppose the equilibrium has separating election-period strategies: xt (st |C) 6= xt (st |B) for some state st . Then, either γˆ Pt (1) > γ P > γˆ Pt (0) i
ACCEPTED MANUSCRIPT or γˆ Pt (0) > γ P > γˆ Pt (1) , so the voter’s best response is to condition reelection on the election-period policy xt that improves the policymaker’s reputation. By Assumption IIa both types of policymaker prefer to override their preferences and comply at t, because the
RI P
T
resultant expected payoff is bounded below by the low type’s future expected payoff, namely P∞ 2τ −1 . Therefore, policymaker strategies in an election [1 + β (1 − π)] = β[1+β(1−π)] τ =1 β 1−β 2 period are pooling in both states: xt (st |C) = xt (st |B) for st = 0, 1. With pooling electionperiod strategies off-equilibrium path beliefs are pinned down by policymaker preferences,
SC
due to the assumption of a small fraction of non-strategic agents. Off-path, preference-based, policymaker equivalent reputations would be γˆ Pt (1) =
NU
γ P (1−π) . Thus pooling on xt = 0 cannot be an γ P (1−π)+(1−γ P )π P P γ t (0) = γ , which makes the probability of reelection
γP π γ P π+(1−γ P )(1−π)
> γ P > γˆ Pt (0) =
equilibrium because γ Pt (1) > γ P and
for xt = 1 no smaller than for xt = 0,
so each type would sometimes choose xt = 1 in election periods. The unique equilibrium then
MA
has election-period pooling on xt = 1, supported by equilibrium voter beliefs γ Pt (1) = γ P and γ Pt (0) < γ P .
Under hierarchical accountability the voter reelects the incumbent intermediary at the
ED
end of electoral term [t − 1, t] iff γˆ It (xt , yt ) ≥ γ I . Thus, at t − 1 the incumbent intermediary follows his preferences, namely retains only policymakers of his own type. Then,
PT
the policymaker also follows his preferences at t − 1. Suppose the equilibrium has separating election-period strategies: yt (xt |C) 6= yt (xt |B) for some policy xt . Then, either
CE
γˆ It (xt , 1) > γ I > γˆ It (xt , 0) or γˆ It (xt , 0) > γ I > γˆ It (xt , 1) , so given xt the voter’s best response is to condition reelection on the retention decision yt that improves the intermediary’s reputation. Overriding the preferred retention decision at t in order to get reelected is preferred
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
by both intermediary types because that yields positive future payoffs whereas following preferences yields zero. Thus, yt (xt |C) = yt (xt |B) for xt = 0, 1, and election-period strategies cannot be separating. With pooling election-period strategies off-equilibrium path beliefs are pinned down by agent preferences, due to the assumption of a small fraction of nonstrategic agents. Off-path, preference-based, intermediary equivalent reputations would be γ I (1−γ P )(1−π) γI γP π I γˆ It (1, 1) = γ I γ P π+(1−γ ˆ It (1, 0) = γ I (1−γ P )(1−π)+(1−γ I )γ P π (< γ I ); γˆ It (0, 1) = I )(1−γ P )(1−π) (> γ ); γ γ I (1−γ P )π γ I γ P (1−π) I ; γ ˆ (0, 0) = . Therefore, if xt = 1 off-path beliefs I P I P I P t γ γ (1−π)+(1−γ )(1−γ )π γ (1−γ )π+(1−γ I )γ P (1−π) can only support intermediary pooling on retaining the policymaker. If xt = 0 the form of the equilibrium strategies depends on whether γ P (1−π) ≶ 1 − γ P π, or γ P ≶ π. When the left-hand side is larger, then γˆ It (0, 1) > γ I > γˆ It (0, 0) and so the resulting off-path beliefs can
only support intermediary pooling on retaining the policymaker. Otherwise, the resulting off-path beliefs can only support intermediary pooling on replacing the policymaker. The
ii
ACCEPTED MANUSCRIPT unique equilibrium can thus take two forms: (i) in election period t the intermediary retains regardless of xt ; or (ii) in election period t the intermediary retains iff xt = 1. In the first type of equilibrium the policymaker follows his preferences in an election period; in the second
T
type of equilibrium, by the argument in the previous paragraph and Assumption IIa, both
Proposition 2 For neutral policy issues π =
1 2
RI P
policymaker types choose the popular policy in an election period.
policymaking outcomes do not vary with
SC
the accountability form in election periods or non-election periods. Proof of Proposition 2. The claim follows by plugging π =
1 2
into the voter’s equilibrium
beliefs in the Proof of Proposition 1. Under electoral accountability election-period policy xt
NU
does not signal policymaker type θPt since γˆ Pt (1) = γˆ Pt (0) = γ P . Assuming that when indifferent between the incumbent and the challenger the voter reelects the incumbent, the voter
MA
reelects the incumbent policymaker regardless of policy; alternatively one could assume the voter plays the equilibrium with the highest voter welfare, which is the insulating equilibrium since γ P + 1 − γ P 2π (1 − π) > π. This insulates the policymaker in an election period, so
ED
xt (st |C) = st , and xt (st |B) = 1 with probability (1 − π) and xt (st |B) = 0 with probability
π. Under hierarchical accountability the retention signal is informative, ruling out separatγP >
1 2
PT
ing election-period intermediary strategies. Since only the case γ P > π is now possible, as = π, based on Proposition 1 only the isulating equilibrium can be supported, thus
CE
the unique pooling equilibrium features policymaker election-period insulation.
A.2 Theory Robustness I: Incentive Dynamics
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
In this section we explore the robustness of the incentives model results in Appendix A.1 by relaxing the constraint implicit in Assumption 1 that reelection incentives affect only electionperiod behavior (as in Rogoff 1990, Shi and Svensson 2006). Here we use the following assumption that allows reelection incentives to affect both election and non-election period behavior (as in Martinez 2009). ∞ Assumption 1(I) Given per-electoral-term i.i.d. preference shocks η j2τ −1 τ =1 such that η j2τ −1 ∈ {B, C} with P η j2τ −1 = C = γ j , agent j’s type at [t + 1, t + 2] is θjt+1 = θjt+2 ∈ j η t−1 , η jt+1 with P θjt+2 , θjt+1 = η jt−1 = λ. Assumption 2(I) The discount factor β satisfies:
β(1−π) 1−β 1 2
> 1.
: (i) Hierarchical accountability ei ther insulates the policymaker pre-election, if γ (1 − π)2 > 1 − γ P π 2 , or creates policy-
Proposition 1(I) For popular policy issues
π> P
iii
ACCEPTED MANUSCRIPT maker pandering incentives pre-election with probability γ I (1 − π) + 1 − γ I π, otherwise;
an intermediary retains only a policymaker of his own type.
T
(ii) Electoral accountability creates policymaker pandering incentives pre-election with proba bility γ P (1−π)+ 1 − γ P π; (iii) Agents follow their preferences post-election; in particular,
RI P
Proof of Proposition 1(I). For electoral accountability the voter reelects the incumbent
SC
policymaker at the end of electoral term [t − 1, t] iff γˆ Pt (xt−1 , xt ) ≥ γ P . To see this, note γ jt (Xt , Yt ) ≡ P θjt+1 = C| (Xt , Yt ) = P θjt+1 = C| (Xt , Yt ) , θjt+1 = η jt+1 (1 − λ) + P{θjt+1 = C| (Xt , Yt ) , θjt+1 , θjt = η jt−1 }λ (1 − λ) + P θjt+1 = C| (Xt , Yt ) , θjt+1 , θjt = η jt−1 , η jt−3 λ2 . If
MA
NU
θjt+2 , θjt+1 = θjt , θjt−1 = η jt−1 , earlier observations (Xt−2 , Yt−2 ) are generated by preference types η jt−3 and earlier, which are independent of η jt−1 . Thus, P θjt+1 = C| (Xt , Yt ) , θjt+1 , θjt = η jt−1 = P θjt+1 = C| (xt−1 , yt−1 , xt , yt ) , θjt+1 = θjt = η jt−1 . When θjt+1 = 6 θjt , the period t+1 type is in dependent of period t type, since θjt+1 , θjt ∈ η jt+1 , η jt−1 , η jt+1 , η jt−3 , η jt−1 , η jt−3 . Thus, P θjt+1 = C| (Xt , Yt ) , θjt+1 6= θjt = γ j . Using these observations, it follows that γ jt (Xt , Yt ) = γˆ jt (xt−1 , yt−1 , xt , yt ) λ (1 − λ) + γ j [1 − λ (1 − λ)] . One can also write:
ED
γ jt (Xt , Yt ) − γ j = λ (1 − λ) γˆ jt (xt−1 , yt−1 , xt , yt ) − γ j
(2)
PT
Suppose in equilibrium policymaker types play separating strategies. Then, because π > 21 , there is one pair of policies (xt−1 , xt ) that is chosen more often by the biased type. The
CE
voter’s best response is to not reelect when observing that policy pair. In a state pair (st−1 , st ) where the biased type prefers that policy pair, forgoing it at t yields at least P∞ τ β(1−π) τ =1 β (1 − π) = 1−β , whereas forgoing reelection yields 1 at t. Thus, by Assumption
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
2(I), policymaker strategies have to be pooling. In a pooling equilibrium off-equilibrium path voter beliefs are pinned down by policymaker preferences, due to the assumption of a small fraction of non-strategic agents. Then, γˆ Pt (1, 1) =
γˆ Pt (0, 1) = γ P ; γˆ Pt (0, 0) =
γ P (1−π)2 γ P (1−π)2 +(1−γ P )π 2
γ P π2 (> γ P π 2 +(1−γ P )(1−π)2
γ P ); γˆ Pt (1, 0) =
(< γ P ). Assuming that when indifferent be-
tween the incumbent and the challenger the voter reelects the incumbent, the voter reelects when at least one of the last two observed policies is popular. A reelection-seeking policymaker will then choose the popular policy at least once. When st−1 is such that the incumbent prefers the popular policy, he chooses xt−1 = 1 and then xt according to his period t preferences. When st−1 is such that the incumbent prefers the unpopular policy, choosing xt−1 = 1 and then xt according to period t preferences yields β during the term whereas choosing xt−1 according to period t − 1 preferences and then xt = 1 yields 1 + βπ for a congruent type and 1 + β (1 − π) for a biased type. Thus, for both types it is optimal to comply later in the term. Pandering in election period t occurs when the popular iv
ACCEPTED MANUSCRIPT policy went against the policymaker’s period t − 1 preferences, that is with probability γ P (1 − π) + 1 − γ P π. By Assumption 2(I) at t both policymaker types prefer to over-
RI P
T
ride their election-period preferences and comply with the voter’s reelection rule, because P 2τ −1 it yields ∞ {π (1 + β) + (1 − π) (1 + βπ)} = β[1+βπ(2−π)] for a congruent type and τ =1 β 1−β 2 P∞ 2τ −1 β [1+β (1−π 2 )] for a biased type. {(1 − π) (1 + β) + π [1 + β (1 − π)]} = τ =1 β 1−β 2 Under hierarchical accountability, the voter reelects the incumbent intermediary at the
end of electoral term [t − 1, t] iff γˆ It [(xt−1 , yt−1 ), (xt , yt )] ≥ γ I . See equation (2). >
1 , 2
Be-
given (xt−1 , xt ), there is one pair of retention decisions (yt−1 , yt ) that is
SC
cause γ
P
chosen more often by the biased type, which makes the voter not reelect when observ-
NU
ing that pair of intermediary reactions. But then the biased type would not choose that course of action in the first place because pursuing reelection yields a positive expected
MA
payoff. Thus, intermediary strategies have to be pooling. In a pooling equilibrium offequilibrium path voter beliefs are pinned down by intermediary preferences, due to the assumption of a small fraction of non-strategic agents. Then, after two popular policies γ I γ P π2
γ I γ P π 2 +(1−γ I )(1−γ P )(1−π)2
ED
γˆ It [(1, 1), (1, 1)] = 2
γ I (1−γ P ) (1−π)2
(> γ I ); γˆ It [(1, 0), (1, 1)] = γ I ; γˆ It [(1, 0), (1, 0)] =
(< γ I ); one popular and one unpopular policy γ I (1−γ P )2 (1−π)2 +(1−γ I )(γ P )2 π 2 γ I γ P π(1−π) I ˆ It [(0, 0), (1, 0)] = γˆ It [(1, 1), (0, 1)] = γ I γ P π(1−π)+(1−γ I )(1−γ P )(1−π)π (> γ ); γ
γˆ It [(0, 1), (1, 1)] =
CE
PT
γˆ It [(1, 0), (0, 0)] = 2 γ I (1−γ P )γ P π 2 γ I (1−γ P ) (1−π)π I I (> (< γ ); γ ˆ [(0, 0), (1, 1)] = 2 2 t γ I (1−γ P )γ P π 2 +(1−γ I )γ P (1−γ P )(1−π)2 γ I (1−γ P ) (1−π)π+(1−γ I )(γ P ) π(1−π) 2 I P P γ (1−γ )γ (1−π) γ I ); γˆ It [(1, 0), (0, 1)] = γ I (1−γ P )γ P (1−π)2 +(1−γ I )γ P (1−γ P )π2 (< γ I ); after two unpopular policies γ I γ P (1−π)2 γˆ It [(0, 1), (0, 1)] = γ I γ P (1−π) (> γ I iff γ P (1 − π)2 > 1 − γ P π 2 iff γ P > 2 +(1−γ I )(1−γ P )π 2
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
p); γˆ It [(0, 0), (0, 1)] = γ I ; γˆ It [(0, 0), (0, 0)] =
2
γ I (1−γ P ) π 2
γ I (1−γ P )2 π 2 +(1−γ I )(γ P )2 (1−π)2
(> γ I iff γ P < p),
and we set γˆ It [(xt−1 , 1), (xt , 0)] = γˆ It [(xt−1 , 0), (xt , 1)] to preserve symmetry, even though [(xt−1 , 1), (xt , 0)] are not consistent with the preferences of either type of intermediary. Assuming that when indifferent between the incumbent and the challenger the voter reelects the incumbent, the voter reelects (i) after two popular policies, iff the intermediary retains at least once; (ii) after one popular and one unpopular policy, iff the intermediary retains after the popular policy, and (iii) after two unpopular policies, iff the intermediary retains at least once, when γ P > p, regardless of the intermediary’s reaction, when γ P (1 − π)2 > 1 − γ P π 2 , or iff the intermediary replaces at least once, when γ P (1 − π)2 < 1 − γ P π 2 . It remains to
show that the intermediary prefers to comply with the voter’s flexible reelection rule at the
end of the term rather than at the beginning of the term. When γ P (1 − π)2 > 1 − γ P π 2 the voter’s strategy is to reelect the intermediary if there
is at least one retention and, in the case of only one popular policy, the retention has to follow v
ACCEPTED MANUSCRIPT the popular policy. If the intermediary faces a policymaker of the same type, retaining in both periods is both consistent with intermediary preferences and guarantees reelection. If however the intermediary faces a policymaker of the opposite type, he could either retain at
T
the beginning of the term or at the end. Retaining at t − 1 and replacing at t produces a zero
RI P
payoff at t, because the policymaker will follow his preferences at t; replacing at t − 1 and retaining at t produces a positive expected payoff at t. Thus, the intermediary always retains at the end of the term. When γ P (1 − π)2 < 1 − γ P π 2 the voter’s strategy is to reelect the
SC
intermediary if there is at least one retention after popular or one replacement after unpopu-
lar, and in the case of only one popular policy, the retention has to follow the popular policy.
NU
If the intermediary faces a popular same type policymaker or an unpopular opposite type policymaker, acting on preferences in both periods results in reelection. In the opposite cases,
MA
the intermediary can either replace an unpopular same type policymaker at t−1 yielding γ P + β π + βγ P P θPt+1 = θIt+1 |θPt = θIt = C + β 1 − γ P P θPt+1 = θIt+1 +Vt+3 and 1 − γ P + β (1 − π) + βγ P P θPt+1 = θIt+1 + β 1 − γ P P θPt+1 = θIt+1 |θPt = θIt = B +Vt+3 for a con-
ED
gruent and biased type respectively, or retain at t − 1 and retain again at t iff he is pop ular yielding π + β P θPt+1 = θIt+1 |θPt = θIt = C + βπ + Vt+3 for a congruent type and (1 − π) + β P θPt+1 = θIt+1 |θPt = θIt = B + β (1 − π) + Vt+3 for a biased type; the inter-
PT
mediary can either retain a popular opposite type policymaker at t − 1 and replace at t,
yielding zero at t, or replace at t − 1 and retain at t iff the incoming policymaker is pop-
CE
ular, yielding a positive expected payoff at t. If the type persistence probability λ (1 − λ) is large enough, the strategy of complying with the voter’s reelection rule at the end of the term has a higher payoff for both intermediary types. The frequency of election-period pan dering then is γ I γ P (1 − π) + 1 − γ P (1 − π) + 1 − γ I γ P π + 1 − γ P π which equals γ I (1 − π) + 1 − γ I π.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Proposition 2(I) For neutral policy issues π =
1 2
policymaking outcomes do not vary with
the accountability form in an election or a non-election period. Proof of Proposition 2(I). The claim follows by plugging π =
1 2
into the voter’s equilib-
rium beliefs in the Proof of Proposition 1(I). Under electoral accountability policies (xt−1 , xt ) do not signal type θPt−1 . Assuming that when indifferent between the incumbent and the challenger the voter reelects the incumbent, the voter reelects the incumbent policymaker regardless of policies chosen in the electoral term [t − 1, t]; alternatively, one could assume that the voter plays the equilibrium with the highest voter welfare, which in this case is the insulating equilibrium, since (1 + β) γ P + 1 − γ P 2π (1 − π) > γ P [π (1 + β) + (1 − π) (1 + βπ)] + 1 − γ P π (1 − π + βπ). This insulates the policymaker throughout the electoral term. Unvi
ACCEPTED MANUSCRIPT der hierarchical accountability the retention signal is still informative, since γ P > 21 , ruling out separating equilibria. Since only the case γ P (1 − π)2 > 1 − γ P π 2 is now possible,
based on Proposition 1(I) the unique pooling equilibrium is policymaker insulation in both
RI P
T
halves of an electoral term.
A.3 Theory Robustness II: Variable Pandering Intensity
SC
Assumption 1(II) The process for types ( θPt , θIt ) is the same as in the Incentives Model in the main text and Online Appendix A.1. The policy payoffs of the congruent and biased types are as in the Incentives Model in the main text and Online Appendix A.1, with the
NU
modification that the preferred policy gives, instead of a payoff of 1, a payoff equal to a random fraction ϕ ˜ of an agent’s value of office Vθj , and that random fraction is uniformly
MA
distributed: ϕ ˜ ∼ U [0, 1] .
Assumption 2(II) The probability γ˜ j that the challenger is a congruent type is uniformly
ED
distributed: γ˜ j ∼ U [0, 1] .
Proposition 1(II) The claims of Propositions 1 and 2 continue to hold. Reelection prob-
PT
abilities satisfy 0 < rt (0) < rt (1) < 1. Moreover, when pre-election pandering occurs in equilibrium, the pandering probability pt is equal to the discounted reelection probability pre-
CE
mium, pt = β [rt (1) − rt (0)] , and is monotonically increasing in π, from pt = 0 when π = 12 , to pt < 1 when π → 1.
Proof of Proposition 1(II). Under electoral accountability, the voter reelects the in-
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
cumbent policymaker if and only if γ Pt (Xt ) ≥ γ˜ P , which becomes, using equation (1), λ (1 − λ) γˆ Pt (xt ) − γ P ≥ γ˜ P − γ P . Since γ˜ P ∼ U [0, 1] , reelection probabilities are rt (xt ) = γ P + λ (1 − λ) γˆ Pt (xt ) − γ P , for xt = 0, 1. We first show the existence of a pandering equilibrium on policy xt = 1 and then rule out other equilibria.
In an equilibrium with pandering on the popular policy xt = 1 the voter’s equilibrium beliefs are determined by γˆ Pt (1) = γ P (1−π) γ P (1−π)+(1−γ P )π
γ P [pt +(1−pt )π] γ P [pt +(1−pt )π]+(1−γ P )[pt +(1−pt )(1−π)]
> γ P > γˆ Pt (0) =
and equation (1). Thus, rt (1) > rt (0) . An incumbent policymaker faces a
tradeoff when the current preferred policy is not the most rewarded electorally, i.e., it is not policy xt = 1. In these cases he will override his current policy preference and pick the popular policy xt = 1 if and only if rt (1) βVθP ≥ ϕ ˜ VθP + rt (0) βVθP . Since ϕ ˜ ∼ U [0, 1] the incumbent policymaker picks policy xt = 1 with probability pt = β [rt (1) − rt (0)] . Using the derived expression for rt (xt ) , this can be rewritten as pt = βλ (1 − λ) γˆ Pt (1) − γˆ Pt (0) . vii
ACCEPTED MANUSCRIPT This last equation, together with the expressions for γˆ Pt (1) , γˆ Pt (0) above, give an implicit function in pt and π. Using the implicit function theorem: ∂ P ∂ P γˆ t (1) − ∂π γˆ t (0) βλ (1 − λ) ∂π ∂pt = ∂π 1 − βλ (1 − λ) ∂p∂ t γˆ Pt (1) where it can be easily verified that
∂ P γˆ ∂π t
∂pt ∂π
then γˆ Pt (1) =
> 0. Note also that when π =
1 2
RI P
T
(3)
∂ P γˆ (0) and ∂p∂ t γˆ Pt (1) < ∂π t γˆ Pt (0) = γ P and so pt = 0.
(1) > 0 >
0. Therefore,
SC
We now show that equilibria where rt (1) ≤ rt (0) can be ruled out. In the case rt (1) = rt (0) the incumbent policymaker acts on his current policy preference, implying that γˆ Pt (1) = (0) =
γ P (1−π) . γ P (1−π)+(1−γ P )π
But since γˆ Pt (1) > γ P > γˆ Pt (0) then, rt (1) > γ P >
NU
γP π , γˆ Pt γ P π+(1−γ P )(1−π)
rt (0) , a contradiction. In the case rt (1) < rt (0) , both incumbent types sometimes pander γP π , γˆ Pt γ P π+(1−γ P )(1−π)
(0) =
MA
on xt = 0. This implies γˆ Pt (1) =
γ P [pt +(1−pt )(1−π)] . γ P [pt +(1−pt )(1−π)]+(1−γ P )[pt +(1−pt )π]
Then, since γˆ Pt (1) > γ P > γˆ Pt (0) again rt (1) > γ P > rt (0) , a contradiction. Thus the unique equilibrium is the one that features pandering on the popular policy xt = 1.
ED
Under hierarchical accountability the equilibrium can be either insulation or pandering, depending on whether γ P > π. The proof closely follows the structure of the Proof for
PT
Proposition 1. In a pandering equilibrium, the intermediary plays the role of the voter in electoral accountability, so the policymaker’s pandering probability has the same properties
CE
as the ones derived above.
A.4 Selection Model
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Assumption 1(III) The process for types and the payoff structure are as in the Incentives Model in the main text and Online Appendix A.1. Assumption 2(III) The discount factor β satisfies: β = 0. Proposition 1(III) For popular policy issues π >
1 2
: (i) Hierarchical accountability per-
forms policymaker selection through intermediary hselection. The increase i in intermediary γI γP I I I congruence post-election is γ t+1 − γ = λ (1 − λ) γ I γ P +(1−γ I )(1−γ P ) − γ , when γ P > π, h i γI π I and γ It+1 − γ I = λ (1 − λ) γ I π+(1−γ − γ , when γ P < π. The increase in policymaker I )(1−π) h i γP γI P P P congruence is γ t+1 − γ = λ (1 − λ) γ P γ I +(1−γ P )(1−γ I ) − γ post-election; pre-election it is γ P γ It+1 P P P γ t+2 −γ = λ (1 − λ) γ P γ I +(1−γ P ) 1−γ I − γ ; (ii) Electoral accountability performs pol( t+1 ) t+1 icymaker selection post-election only. The increase in policymaker congruence post-election h i is γ Pt+1 − γ P = λ (1 − λ)
γP π γ P π+(1−γ P )(1−π)
− γP . viii
ACCEPTED MANUSCRIPT Proof of Proposition 1(III). By Assumption 2(III) both policymaker types as well as intermediary types follow their preferences rather than comply with a reelection or retention rule. As a consequence, their reputations rise and fall according to equation (1).
T
Under electoral accountability the voter reelects only policymakers that chose the pop-
RI P
ular policy in the most recent election period, by Proposition 1. Then, γ Pt+1 − γ P = i h γP π P . Under hierarchical accountabilλ (1 − λ) γˆ Pt (1) − γ P = λ (1 − λ) γ P π+(1−γ P )(1−π) − γ
ity the voter reelects only intermediaries that have retained the policymaker in the election
SC
period, if γ P > π, and reelects only intermediaries that have retained a popular policymaker and replaced an unpopular policymaker in the election period, if γ P < π. Intermediary selec-
MA
NU
tion happens after every election period. Then, for γ P > π a reelected reputai h intermediary’s I γI γP I I I tion improved by γ t+1 − γ = λ (1 − λ) γˆ t (xt , 1) − γ = λ (1 − λ) γ I γ P +(1−γ I )(1−γ P ) − γ I , γ I [γ P π+(1−γ P )π ] P I I I and for γ < π by γ t+1 − γ = λ (1 − λ) γ I [γ P π+(1−γ P )π]+γ I [γ P (1−π)+(1−γ P )(1−π)] − γ = h i γI π I λ (1 − λ) γ I π+(1−γ . Policymaker selection occurs every period as a result of the I )(1−π) − γ
ED
intermediary’s retention decision. Post-election the reputation goes up by h policymaker’s i P I γ γ P γ Pt+1 − γ P = λ (1 − λ) γˆ Pt (xt , 1) − γ P = λ (1 − λ) γ P γ I +(1−γ . The expression P )(1−γ I ) − γ P P P P for pre-election selection γ t+2 − γ = λ (1 − λ) γˆ t (xt+1 , 1) − γ follows by replacing γ I
PT
with γ It+1 . Thus γ Pt+1 > γ Pt+2 = γ P under electoral selection while γ Pt+1 < γ Pt+2 under hierarchical selection, since γ It+1 > γ I . Because π >
1 2
a congruent policymaker is more likely
CE
to choose the popular policy, creating a positive election-period popular policy differential between hierarchical and electoral accountability. Proposition 2(III) For neutral policy issues π >
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
1 2
: (i) Hierarchical accountability per-
forms policymaker selection through intermediaryhselection. The increase i in intermediary γI γP I I I congruence post-election is γ t+1 − γ = λ (1 − λ) γ I γ P +(1−γ I )(1−γ P ) − γ . The increase in h i γP γI P policymaker congruence is γ Pt+1 − γ P = λ (1 − λ) γ P γ I +(1−γ − γ post-election; preP )(1−γ I ) γP γI election it is γ Pt+2 − γ P = λ (1 − λ) γ P γ I +(1−γt+1 − γ P ; (ii) Electoral accountability P ) 1−γ I ( ) t+1 t+1 performs no policymaker selection pre-election or post-election, γ Pt+2 − γ P = γ Pt+1 − γ P = 0. Proof of Proposition 2(III). For electoral accountability the result follows by plugging in π=
1 2
into the electoral selection formula of Proposition 1(III). For hierarchical accountabilityi h
only the case γ P > π is now possible. Then γ It+1 − γ I = λ (1 − λ)
γI γP γ I γ P +(1−γ I )(1−γ P )
and the formulas for policymaker selection are those derived for Proposition 1(III).
ix
− γI
ACCEPTED MANUSCRIPT A.5 Supplementary Empirical Results Robustness. Online Appendix Table A7 reports sensitivity checks for our preferred IV specification and two related OLS specifications. Panel A checks the sensitivity of our results
T
to minor sample alterations: excluding extremely large/small cities, excluding the pre-1972
RI P
period, which was characterized by racial tensions, and the post-1994 period, when the federal government intervened more forcefully in local police employment through COPS
SC
grants, and dropping dependent variable outliers and cities far away from weather stations. Overall, the OLS and the IV results maintain their prior patterns. Online Appendix Table A7 Panel B varies the inference procedure. We examine three
NU
sets of alternative standard errors: clustering on both city and year, clustering on weather station (as this is the unit of observation for the weather variables), and accounting for
MA
spatial correlation (Conley 1999).1 Inferences from both the OLS and the IV estimates remain unchanged.
Thus far we have scaled the number of police employees by population, however a more
ED
relevant measure may be the crime level since police should be proportional to the extent of crime. A high officer-crime ratio may also better capture deviations from policy optimality,
PT
by providing a measure of "excess police." Online Appendix Table A7 Panel C presents estimates with employment scaled by two different crime measures: reported crime and
CE
cleared crime. In both OLS and IV specifications manager cities have significantly fewer officers per crime than mayor cities. Extreme Precipitation and Flooding. As the posited key determinant of city govern-
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
ment form is flood-related infrastructure crises, the ideal instrument would be based on the extent of flood damage in the local flood control era, 1900-1936. This data, however, is not available, not in least part because flood control in this period was handled by local authoritites. The next best approach is to produce an indirect measure of flooding, based on the much more readily available precipitation data. The existing literature on the link between extreme precipitation and flooding, however, has not reached a consensus on this question (Bedient, Huber, and Vieux 2008). To nevertheless validate our precipitation-based instrument we adopt two approaches. First, in Online Appendix Table A8 we use a cross-sectional measure of flooding, based 1
To implement Conley standard errors we allow for a spatial dependence of up to 0.5 degrees latitude and longitude, which corresponds to about 65 miles, quite close to the 58 miles mean distance of a city from the nearest weather station. We do not include year effects in this model as the unbalanced nature of the panel impedes the estimation of the Conley procedure. This slight change accounts for the difference between the point estimates in row (9) and previous ones.
x
ACCEPTED MANUSCRIPT on soil analysis by the USDA, to study correlations with several versions of our precipitationbased instrument. We find that the history of flooding in a location (revealed by the presence of flood-specific sediment deposited in the soil) is clearly related to the history of annual
T
precipitation shocks, variously defined, in the full cross section of cities. Overall, the top-1%
RI P
precipitation cutoff seems to perform best.
Second, in Online Appendix Table A9 we use a time-varying measure of flooding, based on USGS riverflow readings in locations throughout the U.S. This variable is not available
SC
for our full sample. However, we use it to study precipitation-flooding correlations, first by imputing missing observations as no flood, and second, by restricting the sample to
NU
reported observations. The imputation is based on the assumption that reported readings generally reflect higher-than-usual riverflow values. The correlations are positive for all methods of time aggregation, annual, quarterly, monthly, and 7-day max, and are more
MA
precisely estimated when using data at higher frequencies, e.g., quarterly, monthly. In the IV analysis in Online Appendix Table A5, the IV estimates based on quarterly or monthly
ED
LFC precipitation shocks are less precise, however (first stage F-statistics 5.03 and 6.38, compared to 28.40).
Putting these considerations in the balance, we choose the instrument based on annual
PT
aggregation for our preferred IV specification because it has (i) considerable validity in
CE
predicting flooding and (ii) a stronger first stage than the alternatives.
A.6 Data Appendix i) Police Protection
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Police Officers per Capita: Number of sworn police officers, per 1,000 residents. Sources: Uniform Crime Reports (1960-2000). Police Civilians per Capita: Number of civilian (non-sworn) police employees, per 1,000 residents. Sources: Uniform Crime Reports (1960-2000). Police Department Spending per Capita: Total police department expenditure, per resident, in 2000$. Sources: Census of Governments, City Government Finances (19602000). Fraction Officers Unionized 1968: Fraction of the police department unionized in 1968. Source: Municipal Year Book (1969). Crime Rate: Sum of violent and property crime rates. Violent crime rate is total number of murder, rape, and robbery crimes reported, per 1,000 residents. Property crime rate is
xi
ACCEPTED MANUSCRIPT total number of motor vehicle theft, larceny, burglary, and assault crimes reported, per 1,000 residents. Sources: Uniform Crime Reports (1975-2000). Fraction Crimes Cleared: Ratio of total offences cleared through arrest to total crimes
RI P
T
reported. Sources: Uniform Crime Reports (1975-2000). ii) City Characteristics in 1900 Sources: U.S. Census Bureau (1905, 1906).
SC
Police Officers per Capita: Number of sworn police officers, per 1,000 residents. Police Civilians per Capita: Number of civilian (non-sworn) employees, per 1,000 resi-
NU
dents.
Arrests per Capita: Number of police department arrests, per 1,000 residents.
MA
Sewers per Square Mile: Miles of sewers per square mile. Roads per Square Mile: Miles of paved roads per square mile.
ED
Population: Number of city residents. iii) City Government
PT
Manager: Dummy variable equal to 1 if the city has a manager form of government, and 0 otherwise. Sources: Municipal Year Book (1960-2000).
CE
Manager Name: First and last name of the city manager in manager cities. Sources: World Almanac (1960-2000), ICMA, city clerk offices. Election: Indicator variable equal to 1 in an election year, and 0 otherwise. If the election
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
takes place before July 31 the previous year is coded as an election year. Election years are coded based on mayor elections and in case no mayor office exists in a manager city, based council elections. Sources: Municipal Year Book (various years), World Almanac and Book of Facts (various years), www.ourcampaigns.com, city charters, newspaper articles. Non-Partisan Elections: Indicator variable equal to 1 if the city charter in effect in 1960 mandates non-partisan elections, and 0 otherwise. Source: Municipal Year Book (1960). Fraction At-Large Seats: Fraction of city council seats elected at-large in 1960. Source: Municipal Year Book (1960). Early Civil Service: Indicator variable equal to 1 if the city has a non-political civil service before 1937, and 0 otherwise. Source: Civil Service Assembly (1938). Incumbent Mayor Republican: Takes a value of 1 if the incumbent mayor’s party is Republican, and 0 otherwise. Most manager cities maintain a separate and honorary office of the mayor. Sources: Ferreira and Gyourko (2009). xii
ACCEPTED MANUSCRIPT Local Newspaper Sales: Average daily sales of local newspapers, per capita, 1990-2000. Source: George and Waldfogel (2006). City Government Spending Per Capita: Total city government expenditure, in thou-
T
sands of 2000$, per capita. Sources: U.S. Census Bureau, Census of Governments, City
RI P
Government Finances (1960-2000).
Non-Police Employees Per Capita: Total city government employees minus total police protection employees, per 1,000 residents. Sources: U.S. Census Bureau, Census of
SC
Governments (1972-2000).
NU
iv) Demographics
Population: Number of city residents, in thousands. Based on 1960, 1970, 1980, 1990, and
MA
2000 Census of Population numbers, linearly interpolated in intercensal years. Sources: U.S. Census Bureau, Census of Population (various years). Population Density: Number of city residents, in thousands, divided by square miles
ED
of city area. Population based on 1960, 1970, 1980, 1990, and 2000 Census of Population numbers, linearly interpolated in intercensal years. City area based on 1900 and 2000 Census
PT
numbers, linearly interpolated between 1900 and 2000. Sources: Authors’ calculations. Fraction Non-White: Fraction of city population that are non-white. Based on 1960, 1970, 1980, 1990, and 2000 Census of Population numbers, linearly interpolated in intercensal
CE
years. Sources: U.S. Census Bureau, Census of Population (various years). Fraction College Graduate: Fraction of city population that are college graduates. Based on 1960, 1970, 1980, 1990, and 2000 Census of Population numbers, linearly interpolated in
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
intercensal years. Sources: U.S. Census Bureau, Census of Population (various years). Household Income: Median household income of city residents, in 2000$. Based on 1960, 1970, 1980, 1990, and 2000 Census of Population numbers, linearly interpolated in intercensal years. Sources: U.S. Census Bureau, Census of Population (various years). v) Weather Sources: The U.S. Historical Climatology Network’s Daily Temperature, Precipitation, and Snow Data contains daily readings for precipitation, snowfall, and temperature extremes collected from weather stations throughout the U.S. We construct yearly variables based on this dataset. LFC Precipitation Shocks: Fraction of years in the Local Flood Control era (1900-1936) with annual precipitation in the top 1 percent of the national precipitation distribution. xiii
ACCEPTED MANUSCRIPT FFC Precipitation Shocks: Fraction of years in the Federal Flood Control era (1937-1960) with annual precipitation in the top 1 percent of the national precipitation distribution. Century Precipitation Shocks: Fraction of twentieth century (1900-2000) years with
T
annual precipitation in the top 1 percent of the national precipitation distribution.
RI P
Median Precipitation: Median annual precipitation from 1900-2000. Quarterly Precipitation Shocks: Fraction of years in the relevant era (LFC or Century) with quarterly precipitation in the top 1 percent of the national precipitation distribution.
SC
Monthly Precipitation Shocks: Fraction of years in the relevant era (LFC or Century) with monthly precipitation in the top 1 percent of the national precipitation distribution.
NU
7-Day Max Precipitation Shocks: Fraction of years in the relevant era (LFC or Century) with the annual maximum cumulative 7-day precipitation in the top 1 percent of the national
MA
precipitation distribution. vi) Water
ED
Sources: Fishback, Horrace, and Kantor (2005, 2006). The data are reported at the county level. We match it to our sample cities.
PT
Lakes: Number of lakes in the county.
Swamps: Number of swamps in the county. Coast: Dummy variable that indicates if county is located on the coast. to 20 counties.
CE
Small Rivers: Dummy variable that indicates if county has a river that goes through 11 Medium Rivers: Dummy variable that indicates if county has a river that goes through
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
21 to 50 counties.
Large Rivers: Dummy variable that indicates if county has a river that goes through more than 50 counties. vii) Flooding Soil-Based Flooding: Standardized average flood class for the county, based on contemporary surveys by U.S. Department of Agriculture soil scientists. Soil surveys measure the frequency of flood occurrence in locations across the United States. The surveys collect a soil profile for each location. When a flood occurs at a location a thin stratum of gravel, sand, silt, or clay is deposited by the floodwater. These strata in the soil profile produce estimates of flood frequency at the location. Flood classes: (1) Frequent floods (> 50 times per 100 years), (2) Occasional floods (5 to 50 times per 100 years), (3) Rare floods (1-5 times per xiv
ACCEPTED MANUSCRIPT 100 years) or (4) No measurable flood activity. We standardize the county-level averages to have mean zero and standard deviation one. Sources: Fishback, Horrace, and Kantor (2005, 2006).
T
River-Based Flooding: Peak streamflow, in cubic feet per second (cfs), captures the
RI P
maximum instantaneous discharge of a stream or river at a given location. Usually occurs at or near the time of maximum stage, i.e. when water levels are high enough to cause a flood. Based on monitoring station nearest to city downtown or the river. When missing
SC
data is imputed as no flood occurence, the variable is turned into a dummy that takes the value one in each year with a positive reading. Sources: U.S. Geological Service National
NU
Water Information System.
MA
viii) Mechanisms
Crack Epidemic: Dummy variable equal to 1 if crack cocaine arrived in the state in a prior year, based on cocaine-related deaths, and 0 otherwise. Sources:
CE
PT
ED
Moore (2012).
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
xv
Evans, Garthwaite, and
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 Å
Å Å
Å
Å Å
Å Å
Å
Å
Å
Å Å
MA
Å Å
Å
ED
Å Å
Å
Å
Å Å
Å
Å /,3Å /,3Å /,3Å /,3Å Å Å Å Å ÅLOG0OLICEÅ#IVILIANSÅ0ERÅ#AP Å Å ÅÅ Å Å Å Å nÅÅ nÅÅ nÅ Å nÅ Å Å Å Å ÅÅ nÅ Å nÅ Å Å Å Å Å Å Å ÅÅ 9ESÅ 9ESÅ 9ESÅ 9ES 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ 9ESÅ .OÅ 9ESÅ
PT
Å
CE
AC
/,3Å /,3Å /,3Å /,3Å Å Å Å Å ÅLOG0OLICEÅ/FFICERSÅPERÅ#AP Å Å ÅÅ Å Å Å Å nÅÅ nÅÅ nÅ Å nÅ Å Å Å Å ÅÅ nÅ Å nÅ Å Å Å Å Å Å Å ÅÅ 9ESÅ 9ESÅ 9ESÅ 9ES 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ 9ESÅ .OÅ 9ESÅ Å Å
Å
Å Å
Å
Å Å
Å
Å Å
Å
/,3Å /,3Å /,3Å /,3Å Å Å Å Å Å Å Å LOG.ON 0OLICEÅ%MPLÅ0ERÅ#APÅ Å Å Å Å Å nÅÅ nÅÅ nÅ Å nÅ Å Å Å Å ÅÅ nÅ Å nÅ Å Å Å Å Å Å Å ÅÅ 9ESÅ 9ESÅ 9ESÅ 9ES 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ 9ESÅ .OÅ 9ESÅ
XVIÅ
RI P
SC
NU
T
.OTESÅ!UTHORSjÅCALCULATIONSÅWITHÅ ÅCITYÅDATAÅASÅDESCRIBEDÅTHEÅ$ATAÅ!PPENDIXÅ4HEÅUNITÅOFÅOBSERVATIONÅISÅAÅCITY YEARÅ4HEÅFULLÅSAMPLEÅCONSISTSÅOFÅ THEÅÅCITIESÅOUTÅOFÅTOTALÅ ÅTHATÅUSEÅMANAGERÅGOVERNMENTÅATÅSOMEÅPOINTÅDURINGÅTHEÅSAMPLEÅPERIODÅ)NÅEACHÅOFÅTHEÅTHREEÅPANELSÅTHEÅFIRSTÅTWOÅCOLUMNSÅUSEÅ THEÅFULLÅAVAILABLEÅSAMPLEÅWHILEÅTHEÅLASTÅTWOÅCOLUMNSÅRESTRICTÅTHEÅSAMPLEÅTOÅTHOSEÅCITIESÅTHATÅHAVEÅATÅLEASTÅONEÅMANAGERÅSERVINGÅINÅATÅLEASTÅONEÅOTHERÅCITYÅINÅ THEÅSAMPLEÅ4HEÅ&%SÅAREÅBASEDÅONÅAÅCITYÅMANAGERjSÅFIRSTÅANDÅLASTÅNAMEÅ4HEREÅAREÅÅMANAGERSÅINÅTHISÅSAMPLE ÅOFÅWHICHÅÅSERVEDÅINÅATÅLEASTÅTWOÅCITIESÅ 4HEÅREPORTEDÅNUMBERÅ OFÅCONSTRAINTSÅFALLSÅ BELOWÅTHISÅMAXIMUMÅDUEÅTOÅ MISSINGÅDATAÅ ONÅTHEÅDEPENDENTÅVARIABLESÅ4HEÅ TABLEÅREPORTSÅ&Å TESTÅSTATISTICS ÅANDÅ CORRESPONDINGÅP VALUES ÅFORÅTHEÅJOINTÅSTATISTICALÅSIGNIFICANCEÅOFÅADDINGÅ-ANAGERÅ&IXEDÅ%FFECTSÅTOÅAÅREGRESSIONÅMODELÅTHATÅINCLUDESÅYEARÅANDÅCITYÅFIXEDÅEFFECTSÅ 3TANDARDÅERRORSÅAREÅCLUSTEREDÅATÅTHEÅCITYÅLEVELÅ
0ERIODÅ -ANAGERÅ#ITIESÅ /BSERVATIONSÅ
Å $EPÅ6ARÅ Å Å& STATÅ-ANAGERÅ&%ÅÅ P VALUEÅ-ANAGERÅ&%Å Å#ONSTRAINTSÅ !DJUSTEDÅ2Å Å9EARÅ&IXEDÅ%FFECTSÅ #ITYÅ&IXEDÅ%FFECTSÅ -ANAGERÅ&IXEDÅ%FFECTSÅ
-ODELÅ
4!",%Å!Å-ANAGERÅ&IXEDÅ%FFECTSÅ
ACCEPTED MANUSCRIPT
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
/,3Å Å Å
ED
RI P
SC
NU
MA
Å Å Å Å Å Å Å Å 0ANELÅ"ÅÅ#HARACTERISTICS Å0OST ,&#Å Å Å Å Å %ARLYÅ#IVILÅ -ANAGERÅ .ON 0ARTISANÅ &RACTIONÅ3EATSÅÅ !T ,ARGEÅ 3ERVICEÅ %LECTIONSÅ Å Å Å Å
Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ #ROSS 3ECTIONÅ #ROSS 3ECTIONÅ #ROSS 3ECTIONÅ #ROSS 3ECTIONÅ
CE
PT
Å
ÅÅ Å 9ESÅ 9ESÅ #ROSS 3ECTIONÅ
ÅÅ
Å Å ÅÅ
Å ಥÅ Å Å Å ÅÅ
LOG0OPUL Å
/,3Å Å Å
XVIIÅ
Å
T
.OTESÅ!UTHORSjÅCALCULATIONSÅWITHÅCITYÅDATAÅASÅDESCRIBEDÅINÅTHEÅ$ATAÅ!PPENDIXÅ4HEÅUNITÅOFÅOBSERVATIONÅISÅAÅCITYÅFORÅTHEÅSAMPLEÅOFÅLARGESTÅÅSELF GOVERNINGÅCITIESÅINÅTHEÅ5NITEDÅ3TATESÅASÅOFÅÅ4HEÅTABLEÅREPORTSÅ/,3ÅESTIMATESÅOFÅ α1 Å FROMÅEQUATIONÅ ÅINÅTHEÅTEXT ÅWITHÅ -ANAGERÅREPLACEDÅBYÅ THEÅ INDICATEDÅ VARIABLEÅ 3TANDARDÅ ERRORSÅ CLUSTEREDÅ ATÅ THEÅ CITYÅ LEVELÅ INÅ PARENTHESESÅ 7EATHERÅ CONTROLSÅ #ENTURYÅ 0RECIPITATIONÅ 3HOCKSÅ -EDIANÅ 0RECIPITATIONÅ7ATERÅCONTROLSÅNUMBERÅOFÅ,AKESÅNUMBERÅOFÅ3WAMPSÅONEÅINDICATORÅFORÅLOCATIONÅONÅ#OASTÅTHREEÅINDICATORSÅFORÅ3MALLÅ2IVERS Å-EDIUMÅ 2IVERS Å ,ARGEÅ2IVERSÅ ÅINDICATESÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL Å
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL ÅANDÅ
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅ LEVELÅ
7ATERÅ 0ERIODÅ
ÅÅ Å 9ESÅ 9ESÅ #ROSS 3ECTIONÅ
ÅÅ
,&#Å0RECIPITATIONÅ3HOCKSÅ Å
Å/BSERVATIONSÅ Å7EATHERÅ
Å Å ÅÅ
Å Å Å ÅÅ
ÅÅ $EPÅ6ARIABLEÅ ÅÅ
ÅÅ Å
Å/BSERVATIONSÅ
/,3Å Å Å
/,3Å /,3Å /,3Å Å Å Å Å Å Å 0ANELÅ!ÅÅ#HARACTERISTICS Å0RE ,&#Å LOG0OLICEÅ#IVÅ LOG!RRESTSÅPERÅ Å LOG3EWERSÅPERÅ LOG2OADSÅPERÅÅ 3QÅ-ILE Å PERÅ#AP Å #APITA Å 3QÅ-ILEÅ Å Å Å Å Å ಥÅ Å ಥ Å ಥÅ Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å /,3Å /,3Å /,3Å /,3Å
AC
Å ÅÅ $EPÅ6ARIABLEÅ ÅÅ LOG0OLICEÅ/FFÅ PERÅ#AP Å Å Å ,&#Å0RECIPITATIONÅ3HOCKSÅ ಥÅ
-ODELÅ
4!",%Å!Å)NSTRUMENTÅ#OVARIATEÅ"ALANCEÅANDÅ%XCLUSIONÅ2ESTRICTIONÅ
ACCEPTED MANUSCRIPT
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
Å
PT
CE
AC ED
MA
Å /,3Å /,3Å Å Å Å -ANAGERÅ Å
Å Å
Å Å Å ಥÅ ಥÅ Å Å Å 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ .OÅ Å Å Å Å Å Å /,3Å /,3Å Å Å Å Å LOG0OLICEÅ/FFICERSÅPERÅ#APITA Å Å ಥ
Å ಥ
Å Å Å ಥÅ Å Å Å Å 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ .OÅ Å Å Å Å Å Å
XVIIIÅ
RI P
SC
NU
T
.OTESÅ!UTHORSjÅCALCULATIONSÅWITHÅ ÅCITYÅDATAÅASÅDESCRIBEDÅINÅTHEÅ$ATAÅ!PPENDIXÅ4HEÅUNITÅOFÅOBSERVATIONÅISÅ AÅ CITY YEARÅ FORÅ THEÅ SAMPLEÅ OFÅ LARGESTÅ Å SELF GOVERNINGÅ CITIESÅ INÅ THEÅ 5NITEDÅ 3TATESÅ ASÅ OFÅ Å 4HEÅ TABLEÅ REPORTSÅ ESTIMATESÅ OFÅ α1 Å INÅ EQUATIONÅ Å INÅ THEÅ TEXTÅ INÅ COLUMNSÅ Å ANDÅ )6Å REDUCEDÅ FORMSÅ INÅ COLUMNSÅ Å 3TANDARDÅ ERRORSÅ CLUSTEREDÅ ATÅ THEÅ CITYÅ LEVELÅ INÅ PARENTHESESÅ 7EATHERÅ CONTROLSÅ #ENTURYÅ 0RECIPITATIONÅ 3HOCKSÅ -EDIANÅ 0RECIPITATIONÅ 7ATERÅ CONTROLSÅ NUMBERÅ OFÅ ,AKESÅ NUMBERÅ OFÅ 3WAMPSÅ ONEÅ INDICATORÅ FORÅ LOCATIONÅ ONÅ #OASTÅ THREEÅ INDICATORSÅ FORÅ 3MALLÅ 2IVERS Å -EDIUMÅ 2IVERS Å ,ARGEÅ 2IVERSÅ Å INDICATESÅ SIGNIFICANCEÅ ATÅ THEÅ Å PERCENTÅ LEVEL Å
Å SIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL ÅANDÅ
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVELÅ
7EATHERÅ 7ATERÅ $IVISIONÅ&IXEDÅ%FFECTSÅ 0ERIODÅ #ITIESÅ /BSERVATIONSÅ
Å9EARÅ&IXEDÅ%FFECTSÅ
&&#Å0RECIPITATIONÅ3HOCKSÅ
Å Å $EPÅ6ARIABLEÅ Å Å ,&#Å0RECIPITATIONÅ3HOCKSÅ
-ODELÅ
4!",%Å!Å)6Å2EDUCEDÅ&ORMÅANDÅ&ALSIFICATIONÅ
ACCEPTED MANUSCRIPT
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
Å
ED
RI P
SC
NU
MA
XIXÅ
Å Å
T
.OTESÅ!UTHORSjÅCALCULATIONSÅWITHÅ ÅCITYÅDATAÅASÅDESCRIBEDÅINÅTHEÅ$ATAÅ!PPENDIXÅ4HEÅUNITÅOFÅOBSERVATIONÅISÅAÅ CITY YEARÅ FORÅ THEÅ SAMPLEÅ OFÅ LARGESTÅ Å SELF GOVERNINGÅ CITIESÅ INÅ THEÅ 5NITEDÅ 3TATESÅ ASÅ OFÅ Å 4HEÅ TABLEÅ REPORTSÅ )6Å ESTIMATESÅOFÅȕÅ FROMÅEQUATIONÅ ÅINÅTHEÅTEXTÅINÅTHEÅFIRSTÅPANELÅANDÅFIRST STAGEÅ/,3ÅESTIMATESÅOFÅ α1 ÅINÅEQUATIONÅ ÅINÅ THEÅ TEXTÅ INÅ THEÅ SECONDÅ PANELÅ 3TANDARDÅ ERRORSÅ CLUSTEREDÅ ATÅ THEÅ CITYÅ LEVELÅ INÅ PARENTHESESÅ 7EATHERÅ CONTROLSÅ #ENTURYÅ 0RECIPITATIONÅ3HOCKSÅ -EDIANÅ 0RECIPITATIONÅ 7ATERÅ CONTROLSÅ NUMBERÅ OFÅ ,AKESÅ NUMBERÅ OFÅ 3WAMPSÅ ONEÅ INDICATORÅ FORÅ LOCATIONÅONÅ#OASTÅTHREEÅINDICATORSÅFORÅ 3MALLÅ2IVERS Å -EDIUMÅ2IVERS Å ,ARGEÅ2IVERSÅ ÅINDICATESÅSIGNIFICANCEÅATÅTHEÅÅ PERCENTÅLEVEL Å
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL ÅANDÅ
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVELÅ
PT
CE
Å )6Å )6Å )6Å )6Å Å Å Å Å ,&#Å%RAÅ$EFINITIONÅ 0RECIPITATIONÅ3HOCKÅ$EFINITIONÅ Å Å 4OPÅÅ$UMMYÅ 4OPÅÅ$UMMYÅ Å Å Å Å Å 3ECONDÅ3TAGEÅ$EPENDENTÅ6ARIABLEÅ ÅLOG 0OLICEÅ/FFICERSÅPERÅ#APITA Å ಥ Å ಥ
Å ಥ Å ಥ Å Å Å Å Å Å Å Å Å &IRSTÅ3TAGEÅ$EPENDENTÅ6ARIABLEÅ Å-ANAGERÅ Å
Å
Å
Å
Å Å Å Å Å Å Å Å Å ;=Å ;=Å ;=Å ;=Å Å Å Å Å Å 2EDUCEDÅ&ORMÅ$EPENDENTÅ6ARIABLEÅ ÅLOG 0OLICEÅ/FFICERSÅPERÅ#APITA Å Å Å Å ಥ Å ಥ
Å ಥ Å ಥ Å Å Å Å Å Å Å Å Å 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ .OÅ .OÅ .OÅ Å Å Å Å Å Å Å Å Å Å Å Å
AC
Å Å ÅÅ Å
Å ,&#Å0RECIPITATIONÅ3HOCKSÅ Å Å & STATISTICÅ ;P VALUE=Å Å Å ,&#Å0RECIPITATIONÅ3HOCKSÅ Å Å Å9EARÅ&IXEDÅ%FFECTSÅ 7EATHERÅ 7ATERÅ $IVISIONÅ&IXEDÅ%FFECTSÅ 0ERIODÅ #ITIESÅ /BSERVATIONSÅ
Å
Å
-ANAGERÅ
Å
-ODELÅ
4!",%Å!Å!LTERNATIVEÅ)NSTRUMENTÅ$EFINITIONSÅ)Å
ACCEPTED MANUSCRIPT
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
ED
Å
RI P
SC
NU
MA
XXÅ
T
.OTESÅ!UTHORSjÅCALCULATIONSÅWITHÅ ÅCITYÅDATAÅASÅDESCRIBEDÅINÅTHEÅ$ATAÅ!PPENDIXÅ4HEÅUNITÅOFÅOBSERVATIONÅISÅAÅ CITY YEARÅ FORÅ THEÅ SAMPLEÅ OFÅ LARGESTÅ Å SELF GOVERNINGÅ CITIESÅ INÅ THEÅ 5NITEDÅ 3TATESÅ ASÅ OFÅ Å 4HEÅ TABLEÅ REPORTSÅ )6Å ESTIMATESÅOFÅȕÅ FROMÅEQUATIONÅ ÅINÅTHEÅTEXTÅINÅTHEÅFIRSTÅPANELÅANDÅFIRST STAGEÅ/,3ÅESTIMATESÅOFÅ α1 ÅINÅEQUATIONÅ ÅINÅ THEÅ TEXTÅ INÅ THEÅ SECONDÅ PANELÅ 3TANDARDÅ ERRORSÅ CLUSTEREDÅ ATÅ THEÅ CITYÅ LEVELÅ INÅ PARENTHESESÅ 7EATHERÅ CONTROLSÅ #ENTURYÅ 0RECIPITATIONÅ3HOCKSÅ -EDIANÅ 0RECIPITATIONÅ 7ATERÅ CONTROLSÅ NUMBERÅ OFÅ ,AKESÅ NUMBERÅ OFÅ 3WAMPSÅ ONEÅ INDICATORÅ FORÅ LOCATIONÅONÅ#OASTÅTHREEÅINDICATORSÅFORÅ 3MALLÅ2IVERS Å -EDIUMÅ2IVERS Å ,ARGEÅ2IVERSÅ ÅINDICATESÅSIGNIFICANCEÅATÅTHEÅÅ PERCENTÅLEVEL Å
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL ÅANDÅ
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVELÅ
Å Å9EARÅ&IXEDÅ%FFECTSÅ 7EATHERÅ 7ATERÅ $IVISIONÅ&IXEDÅ%FFECTSÅ 0ERIODÅ #ITIESÅ /BSERVATIONSÅ
Å,&#Å0RECIPITATIONÅ3HOCKSÅ
PT
CE
)6Å Å
)6Å )6Å )6Å Å Å Å 0RECIPITATIONÅ!GGREGATIONÅ !NNUALÅ 1UARTERLYÅ -ONTHLYÅ $AYÅ-AXÅ Å Å Å Å Å 3ECONDÅ3TAGEÅ$EPENDENTÅ6ARIABLEÅ ÅLOG0OLICEÅ/FFICERSÅPERÅ#APITA Å ಥ
Å ಥÅ ಥ Å ಥÅ Å Å Å Å Å Å Å -ANAGERÅ Å &IRSTÅ3TAGEÅ$EPENDENTÅ6ARIABLEÅ Å Å
Å
Å
Å Å Å Å Å Å Å Å Å Å ;=Å ;=Å ;=Å ;=Å Å Å Å Å Å 2EDUCEDÅ&ORMÅ$EPENDENTÅ6ARIABLEÅ ÅLOG 0OLICEÅ/FFICERSÅPERÅ#APITA Å Å Å Å ಥ
Å ಥ
Å ಥ
Å ಥÅ Å Å Å Å Å Å Å Å 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ .OÅ .OÅ .OÅ Å Å Å Å Å Å Å Å Å Å Å Å
AC
Å Å ÅÅ Å
Å ,&#Å0RECIPITATIONÅ3HOCKSÅ Å Å & STATISTICÅ ;P VALUE=Å ÅÅ
Å
Å
-ANAGERÅ
Å
-ODELÅ
4!",%Å!Å!LTERNATIVEÅ)NSTRUMENTÅ$EFINITIONSÅ))Å
ACCEPTED MANUSCRIPT
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
Å
PT
CE
AC ED
Å /,3Å /,3Å )6Å Å Å Å ÅLOG#ITYÅ3PENDINGÅPERÅ#APITA Å ÅÅ Å
Å
Å Å Å Å Å Å Å Å 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ 9ESÅ .OÅ Å Å Å Å Å Å Å Å Å /,3Å /,3Å )6Å Å Å Å Å Å Å Å LOG0OLICEÅ$EPTÅ3PENDINGÅPERÅ#APITA Å Å Å ಥÅ Å Å Å Å Å Å 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ 9ESÅ .OÅ Å Å Å Å Å Å Å Å Å
XXIÅ
RI P
SC
NU
MA
T
.OTESÅ!UTHORSjÅCALCULATIONSÅWITHÅ ÅCITYÅDATAÅASÅDESCRIBEDÅINÅTHEÅ$ATAÅ!PPENDIXÅ4HEÅUNITÅOFÅOBSERVATIONÅISÅAÅCITY YEARÅFORÅTHEÅ SAMPLEÅOFÅLARGESTÅÅSELF GOVERNINGÅCITIESÅINÅTHEÅ5NITEDÅ3TATESÅASÅOFÅÅ4HEÅTABLEÅREPORTSÅ/,3ÅANDÅ)6ÅESTIMATESÅOFÅȕÅ FROMÅEQUATIONÅ ÅINÅTHEÅMAINÅTEXTÅWITHÅLOG%MPLOYMENT ÅREPLACEDÅBYÅTHEÅINDICATEDÅDEPENDENTÅVARIABLESÅ3TANDARDÅERRORSÅCLUSTEREDÅATÅTHEÅCITYÅLEVELÅINÅ PARENTHESESÅ 7EATHERÅ CONTROLSÅ #ENTURYÅ 0RECIPITATIONÅ 3HOCKSÅ -EDIANÅ 0RECIPITATIONÅ 7ATERÅ CONTROLSÅ NUMBERÅ OFÅ ,AKESÅ NUMBERÅ OFÅ 3WAMPSÅONEÅINDICATORÅFORÅLOCATIONÅONÅ#OASTÅTHREEÅINDICATORSÅFORÅ3MALLÅ2IVERS Å-EDIUMÅ2IVERS Å,ARGEÅ2IVERSÅ ÅINDICATESÅSIGNIFICANCEÅATÅ THEÅÅPERCENTÅLEVEL Å
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL ÅANDÅ
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVELÅ Å Å Å
7EATHERÅ 7ATERÅ $IVISIONÅ&IXEDÅ%FFECTSÅ 0ERIODÅ #ITIESÅ /BSERVATIONSÅ
Å9EARÅ&IXEDÅ%FFECTSÅ
-ANAGERÅ
Å $EPÅ6ARIABLEÅ ÅÅ
-ODELÅ
4!",%Å!Å'OVERNMENTÅ&ORMÅANDÅ0UBLICÅ3PENDINGÅ
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT 4!",%Å!Å'OVERNMENTÅ&ORMÅANDÅ0OPULARÅ0OLICYÅ2OBUSTNESSÅ Å -ODELÅ
CE
PT
ED
MA
NU
SC
RI P
T
/,3Å /,3Å )6Å Å Å Å ÅLOG0OLICEÅ/FFICERSÅPERÅ#APITA Å $EPÅ6ARÅ ÅÅ ÅÅ Å ಥ
Å Å "ASELINEÅFROMÅ4ABLEÅ ÅCOLSÅ ÅANDÅ Å ಥ
Å ಥ
Å Å Å Å Å Å Å Å 0ANELÅ!Å!LTERNATIVEÅ3AMPLESÅ Å Å$ROPÅ#ITIESÅWITHÅÅ0OPULATIONÅOVERÅMILÅ Å Å Å ಥ
Å ಥ
Å ಥ
Å Å Å Å Å$ROPÅ#ITIESÅWITHÅÅ0OPULATIONÅUNDERÅKÅ ಥ
Å ಥ
Å ಥ Å Å Å Å Å$ROPÅPRE Å0ERIODÅ ಥ
Å ಥ
Å ಥ Å Å Å Å Å Å$ROPÅPOST #/03Å0ROGRAMÅ0ERIODÅ ಥ
Å ಥ
Å ಥ
Å Å Å Å Å$ROPÅ/FFICERÅPERÅ#APITAÅ/UTLIERÅ/BSERVATIONSÅ ಥ
Å ಥ
Å ಥ
Å Å Å Å Å$ROPÅ&ARÅFROMÅ7EATHERÅ3TATIONÅ#ITIESÅ ಥ
Å ಥ
Å ಥÅ Å Å Å Å Å Å Å 0ANELÅ"Å!LTERNATIVEÅ3TANDARDÅ%RRORÅ#ONSTRUCTIONÅANDÅ)NFERENCESÅ Å Å#LUSTERÅONÅBOTHÅ#ITYÅANDÅ9EARÅ Å Å Å ಥ
Å ಥ
Å ಥ
Å Å Å Å Å#LUSTERÅONÅ7EATHERÅ3TATIONÅ ಥ
Å ಥ
Å ಥ
Å Å Å Å Å#ONLEYÅ3TDÅ%RRORSÅ.OÅ9EARÅ&IXEDÅ%FFECTS Å ಥ
Å ಥ
Å ಥ
Å Å Å Å Å Å Å Å 0ANELÅ#Å!LTERNATIVEÅ0OLICEÅ/FFICERÅ3CALINGÅ Å Å0ERÅ2EPORTEDÅ#RIMEÅ Å Å Å ಥ
Å ಥ
Å ಥ
Å Å Å Å Å0ERÅ#LEAREDÅ#RIMEÅ ಥ
Å ಥ
Å ಥ
Å Å Å Å Å Å9EARÅ&IXEDÅ%FFECTSÅ 9ESÅ 9ESÅ 9ESÅ 7EATHERÅ 9ESÅ 9ESÅ 9ESÅ 7ATERÅ 9ESÅ 9ESÅ 9ESÅ $IVISIONÅ&IXEDÅ%FFECTSÅ .OÅ 9ESÅ .OÅ
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
.OTESÅ !UTHORSjÅ CALCULATIONSÅ WITHÅ Å CITYÅ DATAÅ ASÅ DESCRIBEDÅ INÅ THEÅ $ATAÅ !PPENDIXÅ 4HEÅ UNITÅ OFÅ OBSERVATIONÅ ISÅ AÅ CITY YEARÅ FORÅ THEÅ SAMPLEÅ OFÅ LARGESTÅ Å SELF GOVERNINGÅ CITIESÅ INÅ THEÅ 5NITEDÅ 3TATESÅ ASÅ OFÅ Å4HEÅTABLEÅREPORTSÅESTIMATESÅOFÅȕÅINÅEQUATIONÅ ÅINÅTHEÅMAINÅTEXTÅ3TANDARDÅERRORSÅCLUSTEREDÅATÅTHEÅ CITYÅ LEVELÅ REPORTEDÅ INÅ PARENTHESESÅ EXCEPTÅ INÅ 0ANELÅ " Å 7EATHERÅ CONTROLSÅ #ENTURYÅ 0RECIPITATIONÅ 3HOCKSÅ -EDIANÅ0RECIPITATIONÅ7ATERÅCONTROLSÅNUMBERÅOFÅ ,AKESÅNUMBERÅOFÅ 3WAMPSÅONEÅINDICATORÅFORÅLOCATIONÅONÅ #OASTÅ THREEÅ INDICATORSÅ FORÅ 3MALLÅ 2IVERS Å -EDIUMÅ 2IVERS Å ,ARGEÅ 2IVERSÅ 3EEÅ MAINÅ TEXTÅ FORÅ DETAILEDÅ DESCRIPTIONSÅOFÅTHEÅSPECIFICATIONSÅINÅPANELSÅ! Å" ÅANDÅ#Å ÅINDICATESÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL Å
Å SIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL ÅANDÅ
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVELÅ
Å Å
XXIIÅ
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
ÅÅ
Å Å
Å
Å
PT
CE
AC
MA
Å
ED
Å
SC
Å
Å
Å /,3Å /,3Å /,3Å /,3Å /,3Å /,3Å Å Å Å Å Å Å !NNUALÅ0RECIPITATIONÅ#UTOFFÅ 4OPÅÅ0RECIPITATIONÅ!GGREGATIONÅÅ &RACTIONÅ9RSÅ &RACTIONÅ9RSÅ &RACTIONÅ9RSÅ 1UARTERLYÅ -ONTHLYÅ $AYÅ-AXÅ 4OPÅÅ 4OPÅÅ 4OPÅÅ Å Å Å Å 3OIL "ASEDÅ&LOODING Å Å Å $EPENDENTÅ6ARIABLEÅ Å Å Å Å
Å
Å
Å
Å
Å −Å Å Å Å Å Å Å −Å Å ಥÅ ಥÅ ಥ
Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å #ROSS 3ECTIONÅ #ROSS 3ECTIONÅ #ROSS 3ECTIONÅ #ROSS 3ECTIONÅ #ROSS 3ECTIONÅ #ROSS 3ECTIONÅ
XXIIIÅ
NU
RI P
T
.OTESÅ!UTHORSjÅCALCULATIONSÅWITHÅCITYÅDATAÅASÅDESCRIBEDÅINÅTHEÅ$ATAÅ!PPENDIXÅ4HEÅUNITÅOFÅOBSERVATIONÅISÅAÅCITYÅFORÅTHEÅSAMPLEÅOFÅLARGESTÅÅSELF GOVERNINGÅ CITIESÅ INÅ THEÅ 5NITEDÅ 3TATESÅ ASÅ OFÅ Å 4HEÅ TABLEÅ REPORTSÅ /,3Å ESTIMATESÅ OFÅ EQUATIONÅ Å INÅ THEÅ TEXTÅ WITHÅ -ANAGERÅ REPLACEDÅ BYÅ THEÅ INDICATEDÅDEPENDENTÅVARIABLEÅ3TANDARDÅERRORSÅCLUSTEREDÅATÅTHEÅCITYÅLEVELÅINÅPARENTHESESÅ&ORÅCOLUMNSÅ ÅPRECIPITATIONÅSHOCKSÅAREÅDEFINEDÅASÅ FRACTIONÅ OFÅ YEARSÅ WITHÅ ANNUALÅ PRECIPITATIONÅ INÅ THEÅ TOPÅ XÅ PERCENTÅ OFÅ THEÅ NATIONALÅ PRECIPITATIONÅ DISTRIBUTION Å WHEREÅ XÅ Å Å Å Å &ORÅ COLUMNSÅ ÅPRECIPITATIONÅSHOCKSÅAREÅDEFINEDÅASÅFRACTIONÅOFÅYEARSÅWITHÅ YÅPRECIPITATIONÅAGGREGATIONÅINÅTHEÅTOPÅÅPERCENTÅOFÅTHEÅNATIONALÅPRECIPITATIONÅ DISTRIBUTION ÅWHEREÅ YÅ ÅQUARTERLY ÅMONTHLY ÅANDÅTHEÅANNUALÅMAXIMUMÅOFÅ DAYÅCUMULATIVEÅPRECIPITATIONÅ ÅINDICATESÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅ LEVEL Å
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL ÅANDÅ
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVELÅ
/BSERVATIONSÅ
Å2Å Å0ERIODÅ
-EDIANÅ0RECIPITATIONÅ
#ENTURYÅ0RECIPITATIONÅ3HOCKSÅ Å
Å
-ODELÅ
4!",%Å!Å0RECIPITATIONÅ3HOCKSÅANDÅ&LOODINGÅ#ROSS 3ECTIONÅ
ACCEPTED MANUSCRIPT
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
/,3Å Å
PT
/,3Å Å
CE
AC
/,3Å Å
ED
MA
XXIV
RI P
SC
NU
T
.OTESÅ!UTHORSjÅCALCULATIONSÅWITHÅCITYÅDATAÅASÅDESCRIBEDÅINÅTHEÅ$ATAÅ!PPENDIXÅ4HEÅUNITÅOFÅOBSERVATIONÅISÅAÅCITY YEARÅFORÅTHEÅSAMPLEÅOFÅLARGESTÅÅSELF GOVERNINGÅCITIESÅINÅTHEÅ5NITEDÅ3TATESÅASÅOFÅÅ4HEÅRESTRICTEDÅ CITYÅSAMPLEÅCONTAINSÅONLYÅCITIESÅWITHÅAÅFLOODÅBEFOREÅÅANDÅATÅLEASTÅONEÅFLOODÅDURINGÅ Å3TANDARDÅERRORSÅCLUSTEREDÅATÅTHEÅCITYÅLEVELÅINÅPARENTHESESÅ4HEÅTABLEÅREPORTSÅ/,3ÅESTIMATESÅOFÅȕÅ FROMÅEQUATIONÅ ÅINÅTHEÅTEXTÅWITHÅ -ANAGERÅ REPLACEDÅBYÅ 2IVER "ASEDÅ&LOODING ÅANDÅCITYÅFIXEDÅEFFECTSÅINCLUDEDÅASÅADDITIONALÅCONTROLSÅ 0RECIPITATIONÅ 3HOCKÅ VARIABLESÅ AREÅBINARY ÅTAKINGÅAÅVALUEÅOFÅONEÅIFÅ ANÅEXTREMEÅPRECIPITATIONÅEVENTÅOCCURSÅINÅTHEÅRELEVANTÅDATAÅFREQUENCYÅUSINGÅTHEÅNATIONALÅÅCUTOFFÅFORÅTHEÅRELEVANTÅMEASUREÅ ÅINDICATESÅSIGNIFICANCEÅATÅTHEÅ ÅPERCENTÅLEVEL Å
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL ÅANDÅ
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVELÅ
/,3Å Å
Å /,3Å /,3Å /,3Å /,3Å /,3Å /,3Å /,3Å Å Å Å Å Å Å Å Å Å 0RECIPITATIONÅ!GGREGATIONÅ Å !NNUALÅ 1UARTERLYÅ -ONTHLYÅ $AYÅ-AXÅ Å Å Å Å Å Å Å Å Å Å Å Å Å $EPENDENTÅ6ARIABLEÅ Å2IVER "ASEDÅ&LOODINGÅ Å Å Å Å Å Å Å Å Å Å Å Å 0RECIPITATIONÅ3HOCK Å
Å Å Å Å
Å Å
Å
Å Å Å
Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å 9EARÅ&IXEDÅ%FFECTSÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ #ITYÅ&IXEDÅ%FFECTSÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ 9ESÅ -ISSINGÅ)MPUTATIONÅ NOÅFLOODÅ MISSINGÅ NOÅFLOODÅ MISSINGÅ MISSINGÅ NOÅFLOODÅ MISSINGÅ MISSINGÅ NOÅFLOODÅ MISSINGÅ MISSINGÅ $ATAÅ&REQUENCYÅ 9EARÅ 9EARÅ 9EARÅ 9EARÅ 1UARTERÅ 9EARÅ 9EARÅ -ONTHÅ 9EARÅ 9EARÅ -ONTHÅ 0ERIODÅ Å Å Å Å Å Å Å Å Å Å Å #ITIESÅ Å Å Å Å Å Å Å Å Å Å Å /BSERVATIONSÅ Å Å Å Å Å Å Å Å Å Å Å
-ODELÅ
4!",%Å!Å0RECIPITATIONÅ3HOCKSÅANDÅ&LOODINGÅ4IMEÅ3ERIESÅ
ACCEPTED MANUSCRIPT
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
Å
Å
ED
Å
NU
nÅ Å nÅ Å Å Å Å 9ESÅ 9ESÅ 9ESÅ !BOVEÅKÅ Å Å Å
MA
nÅ Å nÅ Å Å Å Å 9ESÅ 9ESÅ .OÅ !BOVEÅKÅ Å Å Å
/,3Å /,3Å Å Å Å Å LOG.ON 0OLICEÅ%MPLÅPERÅ#AP Å
PT
CE
AC
/,3Å /,3Å Å Å Å Å LOG0OLICEÅ#IVILIANSÅPERÅ#AP Å Å nÅ nÅ Å Å nÅ nÅ Å Å Å Å Å Å Å Å 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ 9ESÅ !BOVEÅKÅ !BOVEÅKÅ Å Å Å Å Å Å
/,3Å /,3Å Å Å Å Å LOG#ITYÅ3PENDINGÅPERÅ#AP Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å Å 9ESÅ 9ESÅ 9ESÅ 9ESÅ .OÅ 9ESÅ !BOVEÅKÅ !BOVEÅKÅ Å Å Å Å Å Å
Å Å Å
XXV
RI P
SC
T
.OTESÅ!UTHORSjÅCALCULATIONSÅWITHÅ ÅCITYÅDATAÅASÅDESCRIBEDÅINÅTHEÅ$ATAÅ!PPENDIXÅ4HEÅUNITÅOFÅOBSERVATIONÅISÅAÅCITY YEARÅ4HEÅSAMPLEÅISÅTHEÅ LARGESTÅ Å SELF GOVERNINGÅ CITIESÅ INÅ THEÅ 5NITEDÅ 3TATESÅ ASÅ OFÅ Å THATÅ HAVEÅ ATÅ LEASTÅ Å RESIDENTSÅ INÅ Å 4HEÅ TABLEÅ REPORTSÅ /,3Å ESTIMATESÅ OFÅ EQUATIONÅ Å INÅ THEÅ TEXTÅ AUGMENTEDÅ WITHÅ THEÅ INDICATEDÅ VARIABLESÅ ANDÅ INTERACTIONÅ TERMSÅ 3TANDARDÅ ERRORSÅ CLUSTEREDÅ ATÅ THEÅ CITYÅ LEVELÅ REPORTEDÅ INÅ PARENTHESESÅ $EMOGRAPHICÅ CONTROLSÅ 0OPULATIONÅ $ENSITYÅ &RACTIONÅ .ON 7HITEÅ &RACTIONÅ #OLLEGEÅ 'RADUATEÅ -EDIANÅ (OUSEHOLDÅ )NCOMEÅ )NDICATORÅ VARIABLESÅFORÅMISSINGÅVALUESÅADDEDÅ ÅINDICATESÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL Å
ÅSIGNIFICANCEÅATÅTHEÅÅPERCENTÅLEVEL ÅANDÅ
ÅSIGNIFICANCEÅATÅTHEÅ ÅPERCENTÅLEVELÅ Å Å
Å9EARÅ&IXEDÅ%FFECTSÅ #ITYÅ&IXEDÅ%FFECTSÅ $EMOGRAPHICSÅ Å#ITYÅ0OPULATIONÅ 0ERIODÅ #ITIESÅ /BSERVATIONSÅ
-ANAGERÅÑÅ%LECTIONÅ Å %LECTIONÅ
-ANAGERÅ
$EPÅ6ARIABLEÅ ÅÅ
-ODELÅ
4!",%Å!Å#ITYÅ%LECTIONSÅANDÅ/THERÅ0OLICIESÅ
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT Appendix References Bedient, P.B., W.C. Huber, and B. Vieux (2008) Hydrology and Floodplain Analysis, Prentice Hall, Upper Saddle River: NJ.
T
Civil Service Assembly (1938) "Civil Service Agencies in the United States: A 1937 Cen-
RI P
sus," Pamphlet No. 11, January.
Federal Bureau of Investigation, Law Enforcement Management and Administrative Sta-
SC
tistics.
Federal Bureau of Investigation, Uniform Crime Reports. Ferreira, F. and J. Gyourko (2009) "Do Political Parties Matter? Evidence From U.S.
NU
Cities," Quarterly Journal of Economics 124(1): 399-422.
Fishback, P.V., W.C. Horrace, and S. Kantor (2005) "Did New Deal Grant Programs
MA
Stimulate Local Economies? A Study of Federal Grants and Retail Sales During the Great Depression," Journal of Economic History 65(1): 6-71. Fishback, P.V., W.C. Horrace, and S. Kantor (2006) "The Impact of New Deal Expen-
ED
ditures on Mobility During the Great Depression," Explorations in Economic History 43(2): 179-222.
PT
George, L.M. and J. Waldfogel (2006) "The New York Times and the Market for Local Newspapers," American Economic Review 96(1): 435-447.
CE
International County/City Management Association, Municipal Year Book. International County/City Management Association, Salaries of Municipal Officials. Newspaper Enterprise Association, World Almanac and Book of Facts.
AC
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
U.S. Census Bureau (1905) "The Statistics of Cities Having a Population of Over 25,000: 1902 and 1903," Bulletin 20. U.S. Census Bureau (1906) "The Statistics of Cities Having a Population 8,000 to 25,000: 1903," Bulletin 45. U.S. Census Bureau, Census of Population. U.S. Census Bureau, Census of Governments. U.S. Geological Service, National Water Information System. Williams, C.N., M.J. Menne, R.S. Vose, and D.R. Easterling (2006) "U.S. Historical Climatology Network Daily Temperature, Precipitation, and Snow Data," Oak Ridge National Laboratory.
xxvi
ACCEPTED MANUSCRIPT Highlights
T
RI P
SC
NU
x
MA
x
ED
x
PT
x
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
CE
x
AC