Crime and private investment in urban neighborhoods

Crime and private investment in urban neighborhoods

Accepted Manuscript Crime and Private Investment in Urban Neighborhoods Johanna Lacoe , Raphael W. Bostic , Arthur Acolin PII: DOI: Reference: S0094...

NAN Sizes 0 Downloads 145 Views

Accepted Manuscript

Crime and Private Investment in Urban Neighborhoods Johanna Lacoe , Raphael W. Bostic , Arthur Acolin PII: DOI: Reference:

S0094-1190(18)30088-3 https://doi.org/10.1016/j.jue.2018.11.001 YJUEC 3148

To appear in:

Journal of Urban Economics

Received date: Revised date:

17 July 2017 29 October 2018

Please cite this article as: Johanna Lacoe , Raphael W. Bostic , Arthur Acolin , Crime and Private Investment in Urban Neighborhoods, Journal of Urban Economics (2018), doi: https://doi.org/10.1016/j.jue.2018.11.001

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

Johanna Lacoea Raphael W. Bosticb Arthur Acolinc

May 2018

CR IP T

Crime and Private Investment in Urban Neighborhoods

CE

PT

ED

M

AN US

Abstract: The question of how best to improve neighborhoods that lag behind has drawn considerable attention from policy-makers, practitioners, and academics, yet there remains a vigorous debate regarding the best approaches to accomplish community development. This paper investigates the role crime policy plays in shaping the trajectory of neighborhoods. Much of the existing research on neighborhood crime was conducted in rising-crime environments, and the evidence was clear: high levels of crime have adverse effects on neighborhoods and resident quality of life. This study examines how private investment in neighborhoods in two cities – Chicago and Los Angeles – changed as the incidence of neighborhood crime changed during the 2000s, a period when crime was declining city-wide in both places. Using detailed blockface-level data on the location of crime and private investments between 2006 and 2011, the analysis answers the question: Do changes in crime affect private development decisions? The results show that private investment, as represented by building permits, decreases on blocks where crime increases in the past year. We also find that the relationship between crime and private investment is not symmetric – private investment appears to only be sensitive to crime in rising crime contexts. The result is present in both cities, and robust to multiple definitions of crime and the elimination of outliers and the main commercial district. These results suggest that crimereduction policies can be an effective economic development tool, but only in certain neighborhoods facing specific circumstances.

AC

Key words: crime; investment; neighborhoods JEL classification: R a

Mathematica Policy Research, 505 14th St., Suite 800, Oakland, CA 94612. Corresponding author, contact at [email protected]. b Federal Reserve Bank of Atlanta, contact at [email protected]. C University of Washington, College of Built Environments. 3950 University Way NE Seattle, WA 98105, contact at [email protected].

ACCEPTED MANUSCRIPT

1. Introduction The question of how best to improve neighborhoods that lag behind has drawn considerable attention from policy-makers, practitioners, and academics. Many strategies have been proposed and tried, including subsidies, changes to tax law, investment in education, and improvements in

CR IP T

physical infrastructure. These have yielded a bevy of studies and commentary, and there remains a vigorous debate regarding the best approaches to accomplish community development (Bostic, 2014).

The topic of interest for this paper is whether crime policy can help shape the trajectory of

AN US

neighborhoods and the people that live in them. Significant research efforts have aimed to

understand neighborhood crime patterns in cities across the United States and how crime affects neighborhood quality, including housing values and resident health and well-being (Taylor, 1995

M

for example). Much of this research was conducted in rising-crime environments, and the evidence was clear: high levels and elevating rates of crime in a neighborhood have adverse

ED

effects on neighborhood and individual quality of life. This finding begs a question: is the dynamic relationship between crime and neighborhoods

PT

symmetric? That is, if increasing crime has negative effects, does declining crime have positive

CE

effects that lead to measurable change? If so, this would suggest that policies focused on reducing crime could be an effective and potentially important economic development tool. One

AC

version of this argument has been proposed, namely the “broken windows” strategy that posits that vigorous enforcement of relatively minor infractions sparks a dynamic that leads to acrossthe-board reductions in crime and eventually increases in neighborhood quality (Kelling and Wilson, 1982). Empirical studies of this theory have yielded only mixed results, however (Harcourt and Ludwig, 2006; Corman and Mocan, 2005). So the question remains unresolved.

1

ACCEPTED MANUSCRIPT

This paper seeks to contribute to this literature using a different approach. In this study, we examine how private investment in neighborhoods in two cities – Chicago and Los Angeles – changed as the incidence (and rate) of crime in those neighborhoods changed during the 2000s, a period when crime was declining city-wide in both places. Using detailed block-level data on the

CR IP T

location of crime and private investments, the analysis tries to answer three questions. First, we seek evidence of whether changes in crime appear to affect private development decisions. The answer is yes. Private investment, as represented by building permit purchases, increases on blocks where crime falls in the previous year. The magnitude of the relationship varies across the

AN US

two cities, but the finding is robust. Second, we explore whether permit activity is influenced by both increases and decreases in crime. The answer is no, the relationship between crime and private investment is only present in rising crime contexts. Third, we explore whether some

M

neighborhoods benefit more from changes in crime. The answer here is also yes. We find that the effects of a change in crime are largest in neighborhoods with the highest original levels of

ED

crime. Taken together, these results suggest that crime-reduction policies can be an effective economic development tool, but only in certain neighborhoods facing specific circumstances.

PT

The analysis leverages point-specific crime and building permit data to apply a difference-in-

CE

difference approach to estimate the association between annual blockface-level crime and subsequent building activity. The inclusion of blockface fixed effects and controls for census

AC

tract-level time trends allows for a comparison of building activity on blockfaces that are within the same tract but are home to differing levels of crime. We test the robustness of the central findings by estimating models using threshold levels of crime, controlling for baseline crime rates, removing outlier observation and dense city centers, applying alternative estimators, and exploring variation in the impact estimates by crime type and permit type.

2

ACCEPTED MANUSCRIPT

The paper is organized as follows. First, we provide background on what is known about the relationship between neighborhoods and crime. Next, we discuss the empirical analysis, by first introducing the data and methodology and then reviewing our results. We close with some concluding thoughts about how changes in crime influence the level of private investment in a

CR IP T

neighborhood. 2. Background

While much has been written trying to explain why crime occurs and how it can be deterred (see, for example, Becker (1968) and Levitt (2004)), this research is more closely linked to the

AN US

literature on the relationships that crime has with urban economic activity. One strand of this work examines how crime affects household and business location decisions. At the individual level, research has found that high crime rates in central city neighborhoods cause families to

M

migrate out of neighborhoods, with the evidence suggesting that families with children and people with more education are particularly sensitive to crime (Katzman, 1980; Cullen and

ED

Levitt, 1999; Boggess and Hipp, 2010). The converse does not appear to be true, however. Declining crime rates have not been found to induce in-migration by those living in the suburbs,

CE

O’Regan, 2010).

PT

though they do appear to limit the extent of out-migration from urban neighborhoods (Ellen and

Regarding business location decisions, Rosenthal and Ross (2010) is one of a small number

AC

of studies that has considered how crime affects such decisions. Consistent with expectations, they find that a business’ choice of a site is negatively associated with violent crime. Bollinger and Ihlanfeldt (2003) and Abadie and Dermisi (2008) similarly find that business activity is reduced in neighborhoods where crime rates or the perceived threat of crime is higher.

3

ACCEPTED MANUSCRIPT

Other research along these lines has focused on crime’s impact on property values. The many studies looking at this question have found a consistent negative relation between crime and value. Taylor (1995) finds that housing prices are lower in communities with higher crime. Similarly, Gautier, Siegmen and Vuuren (2009) show that a highly publicized murder was

CR IP T

associated with reductions in value of property owned by households with cultural and ethnic affinity to the murderer. The perception and fear of a heightened likelihood of crime has also been shown to affect values. For example, studies have shown that property values were lower in communities in which a registered sex offender lived (Pope, 2008). Neighborhood crime has

AN US

been found to reduce commercial property values, one potential proxy for economic activity (Lens and Meltzer, 2016).

An additional important consideration is whether any observed relationships hold equally

M

across all types of neighborhoods. One might expect to see systematic variation in any crimeinvestment relationships according to particular neighborhood characteristics because

ED

correlations have been found between crime and certain neighborhood characteristics. For example, Krivo and Peterson (1996) found that extremely disadvantaged areas, defined using an

PT

index of distress and disadvantage, have higher levels of violent crime. Similarly, evidence

CE

suggests that crime dynamics in Latino neighborhoods are different from those in other neighborhoods. Boggess and Hipp (2010) find that stable Latino neighborhoods are insulated to

AC

some extent against violence. Similarly, Burchfield and Silver (2013) interpret their results as suggesting that crime in Latino neighborhoods is less a function of collective efficacy, and thus a weaker signal of neighborhood trajectory, than in other places. To the extent that private investors are aware or perceive that there are different crime dynamics in neighborhoods, they may be more or less sensitive to a given change in crime in a neighborhood.

4

ACCEPTED MANUSCRIPT

The current research is also linked to a literature that seeks to understand the dynamics of neighborhood change. Rosenthal (2008) shows that it is fairly common for neighborhood change to result in a reordering of status, such that some neighborhoods improve their status relative to others (urban renewal) while others fall in relative status (urban decline). Rosenthal points to

CR IP T

filtering mechanisms and how a community’s demographic mix evolves over time as key drivers of these dynamics. Other research examines the process of gentrification, and seeks to identify its catalysts. A set of studies explores which neighborhoods are likely to gentrify and highlights a number of economic and demographic factors that are important (Martin and Bostic, 2003;

AN US

McKinnish, Walsh, and White, 2010). Our research provides information on crime’s role in these changes, with a particular emphasis on urban renewal possibilities.

The results of this work can also inform the question of whether policy can induce the urban

M

renewal and urban decline documented in Rosenthal (2008). There has been a renewed recognition of the importance of this question given the findings in recent research that a

ED

neighborhood’s economic condition is a major determinant in its residents’ long-run prosperity and their quality of life. For example, Chetty, et al. (2016) finds that children that move out of a

PT

high poverty neighborhood when young are more likely to attend college, have higher incomes,

CE

and have lower probabilities of being single parents. Finding ways to improve neighborhoods can potentially have broad welfare implications.

AC

As summarized briefly in Bostic (2014), the evidence regarding the effectiveness in policy triggering positive neighborhood change is mixed. For example, some analyses of enterprise zone programs, which are a relatively popular tax expenditure program, find they have been effective in generating elevated levels of economic activity while others find only limited effects (Papke, 1993; Bostic and Prohofsky, 2006; Kolko and Neumark, 2010). Research on zoning and

5

ACCEPTED MANUSCRIPT

land use policies suggests that they can impact economic development potential, though much of this research has emphasized the negative aspects of the restrictions this zoning generally introduces (Breuckner, 2007; Knaap, 1985; Glaeser, Gyourko, and Saks, 2005). In addition, evidence suggests that investment in public infrastructure enhances economic performance at

CR IP T

various geographic levels, though the magnitude of effects appears to vary with time, industry and geography (Aschauer, 1989; Munnell, 1990; Morrison and Schwartz, 2006).

Along these lines, our work has implications regarding the potential efficacy of policies aimed at reducing crime in neighborhoods with the goal of sparking economic development.

AN US

Most of the scholarly effort on this issue has focused on the question of whether the strategy of vigorously enforcing laws targeting minor infractions, such as broken windows, are effective, with the results being surprisingly inconclusive (Harcourt and Ludwig, 2006; Corman and

M

Mocan, 2005). Indeed, the current research is the only study we are aware of that has studied the effects on private investment empirically.

ED

3. Data and Approach

The current research seeks to uncover relationships between declines in crime in

PT

neighborhoods and private investment and answer some of the questions raised above. To do

CE

this, we focus our analysis on Chicago and Los Angeles, two large vibrant cities that have very different land use patterns and housing stock (see table 1). Well known facts about the two cities

AC

are reflected in the descriptive characteristics: Chicago has much higher density per square mile than Los Angeles, and the population of Los Angeles is larger. The two cities also differ in terms of housing type: the majority of Los Angeles homes are either single family or have 5 or more units, while Chicago’s housing is fairly equally divided between single family, 2-4 unit, and 5 or more unit buildings. The cities have similarly low homeownership rates: approximately 46

6

ACCEPTED MANUSCRIPT

percent of housing units in Chicago were owner-occupied in 2010, while only 38 percent of units in Los Angeles were owner occupied. The cities are home to different populations as well. In Los Angeles, Hispanic residents make up almost half of the population, while they comprise less than one-third of the population in Chicago. Further, Chicago’s population is one-third Black, while

CR IP T

Black residents comprise less than 10 percent of the population in Los Angeles. By contrast, the level of poverty and median household income in the two cities are more similar. Exploring the similarities and differences in the relationship between neighborhood crime and private

investment across these two, diverse cities will provide more robust information about the

AN US

overall relationship than a case study of an individual city would allow. [Table 1 about here] 3.1 Data and sample

M

Our data are drawn from varied sources and over varied time periods, but we obtained the same information from both cities. In particular, we have location-specific data on crimes,

ED

building permits, and building code violations. In Los Angeles, the Los Angeles Police Department provided data on all reported crimes between 1999 and 2012. The city’s Department

PT

of Building and Safety provided data on building permits from 1999 to 2012 and service requests

CE

(which includes building code violations) from 2002 to 2013. For Chicago, the crime data include reported crimes from the Chicago Police Department between 2001 and 2012. The

AC

building permit and building code violation data were obtained from the Chicago Department of Buildings.

These data are quite detailed in the information they provide. The crime data indicate the

type of crime which we code using UCR categories as violent or property crime, with other

7

ACCEPTED MANUSCRIPT

crimes categorized as “public order.”1 The data also include the date the criminal activity occurred, and the location/address where it took place. For the data on building permits, which is our proxy for private investment, we have information on the address for the permit, the type of permit (new building, renovation, demolition), and the value of the work being done. The

CR IP T

building code violation data includes a breakdown by type of violation, such as having

overgrown or excessive vegetation, having trash or debris on private property, or operating a business from a house or garage. Inclusion of code violations in our analytical framework allows us to control for the level of disorder in a neighborhood, and distinguish disorder from crime.

AN US

Overall, the detail of our database allows us to gain a more nuanced understanding of the

relationship between crime and private investment, including whether any observed relationships vary by type of crime, permit, or code violation.

M

Figure 1 shows how crime has evolved from 2006 to 2011.2 The data show a broad general decline in the level of crime in both cities. Overall, crime in Los Angeles decreased by 19

ED

percent over the study period, and crime in Chicago decreased by 22 percent. Property crime declined in both cities, and violent crime declined in Los Angeles, but the violent crime trend in

PT

Chicago was fairly flat. The pace of investment also slowed in both cities over the time period.

CE

In Figure 2, the number of total permits filed in Los Angeles declined from 2007 to 2009, and then showed slight increases, however, the overall trend was a decline in permit activity of 26

AC

percent over the study period. There was a similar pattern in permit activity in Chicago, with the

1

Total crime includes all reported crimes. Violent crime includes murder/homicide, aggravated assault, rape, and robbery. Property crime includes burglary, theft, arson, and motor vehicle theft. Public order crimes include vandalism, disturbing the peace, and vagrancy. 2 As will be discussed more fully, we trace trends from 2006 to 2011 so that we can explore the possibility that crime effects operate with a lag.

8

ACCEPTED MANUSCRIPT

number of permits decreasing between 2007 and 2009, and then increasing slightly until 2011. The overall decrease in permit activity over the study period was 24 percent. [Figure 1 about here] [Figure 2 about here]

CR IP T

Citywide trends mask significant variation in investment activity and crime at the

neighborhood level. Figures 3 and 4 show how crime and building permits are distributed across blockfaces in Los Angeles, and Figures 5 and 6 show the same distributions for Chicago. The data show that the crime and building permits are both skewed geographically. In the vast

AN US

majority of cases, the typical blockface saw no crime or permits in a given year, with a small but measurable number of blockfaces experiencing relatively high levels of crime. The larger size of Los Angeles results in less concentration of crime than in Chicago, with large areas experiencing

M

very few crimes in a year. In both cities, property crime is far more common that violent crime, occurring 3 to 4 times as often. In terms of permits, more than 75 percent of all permits were for

ED

renovation. In Los Angeles, permits for new building were twice as common as permits for

[Figure 3 about here] [Figure 4 about here] [Figure 5 about here] [Figure 6 about here]

AC

CE

PT

demolition, while the frequency of these two permit types was comparable in Chicago.3

3.2 Estimation strategy

3

Blockfaces with outlier crime counts larger than 3 standard deviations are dropped from the analysis. In Los Angeles, the mean crime count for outlier observations is approximately 90 crimes in one year (ranging from 58-368 crimes), and in Chicago it is 105 crimes in one year (ranging from 59-427 crimes).

9

ACCEPTED MANUSCRIPT

Using these data, we ask if either the level of crime or changes in crime levels in a neighborhood is associated with subsequent changes in the number of building permits in that neighborhood. The unit of observation is the blockface, defined as a street segment between the two closest cross-streets. Analyzing crime trends at the local micro level – such as the blockface

CR IP T

– is critical given research identifying significant heterogeneity in crime trajectories on adjacent streets (Groff, Weisburd, and Yang, 2010). We test whether the number of crimes on a blockface affects the amount of investment that occurs on that blockface in subsequent periods. To capture time-varying characteristics of the surrounding neighborhood that likely also contribute to

AN US

investment decisions, we compare changes in permit activity on the blockface to changes in permit activity on nearby blockfaces – those within the same census tract – in the same year. Our specification includes fixed effects at the blockface level and for census tract by year. The

M

estimated equation is:

ED

In this specification, permitsbt is the annual number of permits pulled on a blockface; crimebt-i is the number of crimes lagged i years; violbt-1 is the number of building code violations in the

PT

previous year; γb is the blockface fixed effect; and δt is the year time trend for the census tract.

CE

The length of time between a change in crime levels in a neighborhood and a measurable change in investment, through the filing of a building permit, is not pre-determined. The effect of

AC

a change in crime could extend beyond the period immediately following when the crime occurs. For example, a murder in a neighborhood might change perceptions of that neighborhood that last for longer than one year. Alternatively, increases in non-violent crime over time may not have an immediate effect on the perception of the neighborhood as dangerous, but multiple years of elevated property crime rates may have that effect. As we have no prior as to what an

10

ACCEPTED MANUSCRIPT

appropriate lag structure should be, we examine this empirically. We include regressions of lagged 12-month counts of crime on building permits, including blockface fixed effects and tract*year fixed effects (Table 2). The data show that 1-year lags are most appropriate in both Chicago and Los Angeles, but that the lag relationships for the second and third years diverge in

CR IP T

the two cities. This analysis suggests that the effects of crime linger longer in Los Angeles than in Chicago, and by quite a substantial amount longer. The sources of such differences are a good subject for future research.

For consistency, in the analysis below we use two-year lags for both cities, which allows us

AN US

to explore persistence in the effect of crime changes over time, but still places a greater

analytical value on the most recent changes in crime. Future research will seek to better understand the underlying dynamics that drive the lag structure. It would also be worth

M

examining the temporal changes in crime and investment decisions more precisely, rather than focusing solely on the observed outcome of that decision (i.e., the issuance of a building permit).

ED

[Table 2 about here]

To control for other disorder on the blockface that may be correlated with changes in crime

PT

and building activity, the models also include a measure of the total number of building code

CE

violations reported on the blockface in the past year. For this and all subsequent regressions, the data covers the period from 2005 through 2011 for the dependent variable. We add the

AC

appropriate lags for crime and violations to complete the dataset. To motivate our approach, we first present an unadjusted OLS model estimating the raw

correlation between crime counts at the blockface level and subsequent permit activity. The basic model is strengthened in the following specifications. First, we add year fixed effects to control for citywide changes in crime and building activity during the study period. Next, we add

11

ACCEPTED MANUSCRIPT

blockface fixed effects to control for time-invariant characteristics of each blockface. To capture policy and practice changes at the precinct level that may be related to crime and perceptions of crime, the next specification interacts police precincts fixed effects with year fixed effects. These models essentially compare changes in permit activity on a blockface over time to changes in

CR IP T

blockfaces within the same police precinct. Finally, to capture time-varying characteristics of neighborhoods, we include time trends at the census tract level in our preferred model. The

census tract*year fixed effects are expected to control for broader neighborhood level changes, such as revitalization or policing, that might affect both investment decisions and crime levels.

AN US

Our analysis is bolstered by multiple robustness tests of the preferred model, including

estimation using log transformations of the dependent variable, negative binomial models, omitting blockfaces with extreme numbers of crimes or permits, and trimming the sample based

M

on blockface length or dropping city centers.

While total crime declined in both cities over the study period, there is significant variation in

ED

crime trends at the neighborhood level. Figures 7 and 8 show the distribution of census tracts in each city that experienced increases in crime over the study period. Clearly, neighborhoods that

PT

experience crime increases in the face of citywide crime declines are likely to be different than

CE

neighborhoods that mirror the citywide trend. Therefore, we first explore variation in the impact estimates by investigating whether the relationship between neighborhood crime and permits

AC

differs for neighborhoods experiencing overall increases in crime during the period compared to those seeing overall decreases in crime during the period. We then extend these models to explore additional sources of variation by estimating threshold models using dummy variables instead of crime counts, stratified models by crime rates in 2005, and models exploring variation

12

ACCEPTED MANUSCRIPT

by crime type (violent, property, public order) and permit type (renovation, new construction, demolition). [Figure 7 about here]

CR IP T

[Figure 8 about here]

4. Results

The progression of models in our baseline specifications in Table 3 from the raw relationship to the most controlled specification show parallel stories in both cities. The inclusion of

AN US

blockface fixed effects – strong controls for time invariant blockface characteristics – causes the sign of the coefficients to flip and substantially improves the explanatory power of the models (column 3). This suggests that without the blockface fixed effects, the models are incorrectly

M

attributing the effect of changes in permits to changes in crime.4 Omitted blockface-level differences (in construction type, use, length, or other factors) appear to be a source of bias. Once

ED

these terms are added, crime and permits move in opposite directions in the year following a change in crime. The most controlled specification (column 5) includes blockface fixed effects,

PT

census tract*year fixed effects, and a control for the number of building code violations in the

CE

past year. The models yield slightly larger estimated effects of changes in crime on total permits in Los Angeles (-0.003) than in Chicago (-0.001). Given the different mean number of permits in

AC

the two cities – Los Angeles (0.72 permits), Chicago (0.12 permits) – an additional crime on a blockface in the last year is associated with 0.42 percent and 1.08 percent decreases from the 4

We also ran regressions at the census tract level (with census tract fixed effects), and the results largely mirrored those of the raw blockface-level analysis that lacked any such controls. This suggests that controlling for variation across blockfaces, as opposed to across the larger census tracts, is essential for identifying and understanding crimerelated relationships. It is possible that spillover from changes on surrounding blockfaces have an impact on a given block building permit activity. Further research is needed to systematically explore these potential interaction effects potentially by aggregating blockfaces or implementing spatial lag models.

13

ACCEPTED MANUSCRIPT

citywide mean number of permits over the time period, respectively. The patterns in both cities are similar despite differences in the underlying levels of crime and development during the study period.5 Additional contextual data, particularly on dimensions that vary over time, such as land use information at the parcel level as used in Lens and Meltzer (2016), would allow to

research. [Table 3 about here]

CR IP T

further explore the role of changing neighborhood environment and should be explored in further

In all of the following models, the coefficient on the lagged measure of building code

AN US

violations is positive and significant, suggesting that holding all else constant, an increase in building code violations is associated with in an increase in permit activity in the following year. While not the focus of this paper, the consistent positive relationship between increases in

M

building code violations filed with the city and subsequent investment activity suggests that controlling for a measure of disorder on the blockface, such as violations, is critical to

ED

understanding the dynamic relationship between crime and investment at the neighborhood level. A possible theory is that, in areas that are experiencing neighborhood change, a higher level of

PT

building code violations reported is a sign of increased building activity and collective efficacy.

CE

But additional research is needed to explore this relationship further.6 Neighborhood change is a dynamic process. Therefore identification is a primary concern

AC

when estimating the relationship between changes in crime and permit activity. There are several

5

Estimates produced without eliminating the outliers discussed in footnote 2 produce qualitatively similar results. Available from authors by request. 6 We also ran models without controlling for these variables and the point estimates on crime in year t-1 and t-2 remain similar (Appendix Table A). At the suggestion of a reviewer, we further looked at the relationship between crime and building code violations. We find a positive and significant relationship in Chicago and a positive but not significant relationship in Los Angeles. This offers support for a neighborhood change theory in which private investors respond to declines in crime by investing in and improving the buildings, resulting in lower number of violations. This issue merits further attention from researchers.

14

ACCEPTED MANUSCRIPT

types of identification challenges that must be considered. A first challenge is spatial in nature. In order to address this, we use blockface fixed effects and census tract*year fixed effects in our baseline regressions to estimate the impact of the change in crime on a given blockface on building permit activity. These introduce controls for the characteristics and trends of the larger

CR IP T

census tract and represent an explicit acknowledgment that broader spatial developments are relevant.

Another challenge arises when one considers when the investment and crime occur. This temporal issue has been noted in other contexts. As one example, land use has been shown to

AN US

affect patterns of crime, complicating the estimation of the relationship between neighborhood crime and land use. For instance, several studies have found significant increases in crime following mortgage foreclosures (Cui and Walsh, 2015; Ellen, Lacoe, & Sharygin, 2013; Lacoe

M

and Ellen, 2015), and another finds that commercial land uses lead to more street crime (Twinam, 2017). We test the extent to which the estimated relationship flows in both directions –

ED

that changes in crime cause permits to change, and the changes in permit activity predict changes in future crime – by estimating Granger tests (a similar approach was taken in Ellen, Lacoe, and

PT

Sharygin, 2013). These models add measures of contemporaneous (t) and future (t+1) crime to

CE

the baseline models. In Table 4 we find that in both cities, the main effects (columns 1 and 3) are unchanged by the inclusion of measures of contemporaneous and future crime, and the

AC

coefficients on these crime terms are insignificant (columns 2 and 4).7 These results provide

7

We conducted alternative specifications for the Granger tests suggested by reviewers. In the first, we regressed crime in period t on building permits in period t-1 along with crime in period t-1 and t-2, and violations in period t-1. In that specification, the coefficient on building permits in period t-1 is insignificant.

15

ACCEPTED MANUSCRIPT

support for the estimated effects flowing from changes in neighborhood crime to building permit decisions, and not in the reverse direction.8 [Table 4 about here] Remaining challenges to causal inference would come from systematic measurement

CR IP T

errors or unobserved characteristics that operate at the block level and are systematically

correlated with permit and crime activity. This possibility, which exists for nearly all research of this nature, remains a potential factor. 4.1 Variation in estimated effects

AN US

Aside from the direct effect of changes in crime on subsequent building permit activity, we are interested in the contours of that relationship. Is the relationship between crime and investment constant? When we consider how permit activity responds to increases or decreases

M

in crime, we find similar patterns by city (Table 5). In both cities, the effect of changes in crime on building permit activity is present only for neighborhoods where crime increased overall

ED

during the study period, with no significant effects observed among neighborhoods where crime decreased. This finding is notable – blockfaces in neighborhoods with increasing crime

PT

experience decreased investment, but those in neighborhoods that are improving do not see a

CE

symmetrical increase in building investment. Neighborhood crime in Los Angeles and Chicago has an asymmetric effect on building activity, which is consistent with previous research at the

AC

city level that finds that, though increases in city-level crime “pushes” people out of cities and

8

It is nonetheless possible that in some cases the investment decision predates the building permit issuance by several months or even years. In these cases, crime level at t-1 might actually be contemporaneous of the investment decision as pointed out by a reviewer.

16

ACCEPTED MANUSCRIPT

into suburbs, decreases in crime do not have the effect of attracting people back into cities (Ellen and O’Regan, 2010).9 [Table 5 about here] Another dimension to consider is whether there are threshold effects, such that reductions in

CR IP T

crime are only meaningful if levels of crime exceed some minimum level.10 We test for threshold effects by creating dummy variables indicating if a blockface had crime in the previous year that exceeded a certain level and substituting these in the regression for the variable indicating the count of the number of crimes. We look systematically at the effect of each additional crime on

AN US

the blockface. The results are shown in Table 6, stratified by the tract crime trend over the study period. We see that there is weak evidence of threshold effects, though the results vary between Los Angeles and Chicago. We find no evidence of larger effects on building permits in Los

M

Angeles as the number of crimes on a blockface increases. However, in Chicago neighborhoods that experienced crime declines, the number of building permits generally declines as the number

ED

of crimes on the blockface grows, though the estimated effects are small. We do not see much consistent evidence of threshold effects in Chicago neighborhoods that experienced crime

PT

increases, but the neighborhoods with higher crime counts (6, 9, or 10 or more crimes) in a year

CE

had larger, significant declines in permit activity. [Table 6 about here]

AC

Another way to consider the dynamics of this relationship is to first group blockfaces according to their level of crime at the beginning of the analytical period and then re-estimate the

9

In an alternative specification we use the whole sample and interact the blockface crime measures with the census tract crime trends. The results show that declining crime tracts, an increase in crime on the blockface is associated with an increase in building permits. These results are available from the authors on request. 10 Or conversely, such that increases in crime are only meaningful if they cause the level of crime to exceed some minimum level.

17

ACCEPTED MANUSCRIPT

base equation. Are neighborhoods with lower or higher initial crime rates affected differently by changes in crime? To explore effects across blockfaces with low baseline crime rates compared to those with higher initial crime rates, we stratify the sample by quartiles of the census tract crime rate in 2005. High crime neighborhoods have the potential for larger changes in crime

CR IP T

rates than low crime neighborhoods. For instance, a decrease in crime may be more noticeable in a high crime area. Alternatively, resident perceptions of safety in neighborhoods with low crime rates may shift more drastically in response to minor increases in crime.

The results, presented in Table 7, show varied effects of crime changes in neighborhoods that

AN US

experience overall increases or decreases in crime. For neighborhoods experiencing crime

declines, we see either no or a weakly significant relationship between changes in crime and subsequent building permit activity on blockfaces in tracts that were high crime at the beginning

M

of the period. However for those blockfaces that were in low crime tracts at the beginning of the period (first quartile), we see some evidence of a negative effect of changes in crime on permits.

ED

In both cities, the largest negative effects on permit activity are in tracts where crime increases over the study period. These effects are spread somewhat evenly across tracts by initial crime

PT

level, suggesting that the overall crime trend is a more important factor in the relationship

CE

between crime and building activity than the original crime level at the beginning of the period. In many of these neighborhoods, the crime effects persist beyond a single year.

AC

[Table 7 about here]

If the primary mechanism through which neighborhood crime affects investment decisions is

through perceptions of the neighborhood as an attractive location for development (due to low or decreasing crime) or an unattractive one (due to high or increasing crime), one may expect effects to vary by permit type. Table 8 shows different patterns of effects by neighborhood crime

18

ACCEPTED MANUSCRIPT

trend in Los Angeles. In decreasing crime neighborhoods, the sole effect on permits for new construction is observed for the two-year lagged measure of crime (Table 8, Panel A, column 1). We observe the opposite relationship for demolition permits in Los Angeles. When crime decreases, demolition permits also decrease (Table 8, Panel A, column 3). This is perhaps

CR IP T

intuitive. In Los Angeles’ increasing crime neighborhoods, there are shorter-term effects of

crime on permits for new construction and renovations: one year following a crime increase, new construction and renovation permits decrease (Table 8, Panel A, columns 6 and 7). In these neighborhoods we find no effect on demolition permits.

AN US

[Table 8 about here]

When we consider the land use type (residential or commercial) in Los Angeles, we find that that commercial permits decrease in response to crime increases in both increasing and

M

decreasing crime neighborhoods, while residential permits only decrease in increasing crime neighborhoods (Table 8, Panel A, columns 5, 9, 10). Residential permit activity appears to be

ED

more sensitive to shorter term changes in crime, while commercial permit activity is affected by changes in crime two years prior. The model in column 5, explains the majority of the variation

PT

in commercial permits (r square 0.624).

CE

In Chicago, we find that changes in crime in the previous two years affect permits for new construction in both types increasing and decreasing crime neighborhoods (Table 8, Panel B,

AC

columns 1 and 6). Patterns for renovation permits vary by neighborhood crime trend: in decreasing crime neighborhoods, a decrease in crime decreases renovation permit activity, while in increasing crime neighborhoods, an increase in crime decreases renovation activity (Table 8, Panel B, columns 2 and 7). We see no relationship between crime and demolition permits in Chicago.

19

ACCEPTED MANUSCRIPT

These results may reflect neighborhood change dynamics in which rising crime dissuades new higher income individuals from moving in to the neighborhood or renovating properties, but may decrease or have no effect on demolition decisions. This is consistent with results from Ellen, Horn and Reed (2016) who find that higher income college educated households are

CR IP T

particularly sensitive to crime decline, contributing to increased investments in these areas.

An additional question is whether investment is sensitive to the type of crime that occurs, as individuals considering investing in a neighborhood could be more invariant to changes in

violent crime than changes in property crime. We test for this by decomposing the crime variable

AN US

and including independent variables for violent, property and public order crimes.11 Following our baseline specification, we apply the two-year lag structure. The results, shown in table 9, indicate that the building permit response to changes in crime differs by type of crime. We do not

M

observe sensitivity to type of crime in decreasing crime neighborhoods in either city. In increasing crime neighborhoods in Los Angeles, we see a marginally significant effect of

ED

changes in property crime in the previous two years on permit activity (Table 9, column 2). However, in Chicago’s increasing crime neighborhoods, we find that increases in all types of

PT

crime decrease permit activity, with the largest effects driven by increases in violent and

CE

property crime (Table 9, column 4). The effect of a change in public order crime on permit activity is smaller in Chicago, suggesting that investors are more agnostic (or potentially less

AC

aware) of changes in this less serious type of crime. [Table 9 about here]

11

The measures of violent, property, and public order crime do not sum to the total number of crimes on a blockface. Violent crimes are UCR part I category crimes (murder/homicide, robbery, felony assault, rape), property crimes are UCR part I category crimes (burglary, larceny/theft, motor vehicle theft, arson), and public order crimes are lower level crimes including graffiti, loitering, prostitution, drugs, and weapons offenses.

20

ACCEPTED MANUSCRIPT

In summary, in both cities we find a negative effect of changes in crime on total permits, with an additional crime on a blockface in the previous year being associated with a 0.4 percent decrease in the number of permits in Los Angeles and a 1.1 percent decrease in the number of permits in Chicago. In both cities, the relationship is driven by decrease in building permit

CR IP T

activity in neighborhoods experiencing an overall increase in crime (rather than by an increase in neighborhoods experiencing a decline in crime), no matter the original level of crime at the

beginning of the study period. These asymmetric results are consistent with previous research that found that increases in crime “push” people out but decreases do not immediately have the

AN US

opposite effect in the short run (Ellen and O’Reagan, 2010).

When looking at the impact of different types of permits, in both cities we find a stronger effect on construction and renovation permits than on demolition permits. In addition, in Los

M

Angeles, commercial permits seem more affected by crime increases than residential permits. In both cities, we don’t measure a significant response to changes in property and violent crimes in

ED

neighborhoods with decreasing crime, but we do find an effect of changes in violent, property and public order crimes in increasing crime neighborhoods in Chicago.

PT

4.2 Robustness tests

CE

We conduct several tests of the robustness of the results to variation in how the dependent variable is constructed. First, we model log transformed counts of permits to account for the

AC

skewed distribution of the dependent variable toward zero (Appendix table B). The models with log transformed counts of permits show results consistent with those generated using count variables. In both Los Angeles and Chicago, the estimated effects of a change in crime two years ago on permits is larger than the effects of a change in crime in the prior year. An alternative approach to modeling the skewed distribution of permits across blockfaces is to estimate

21

ACCEPTED MANUSCRIPT

negative binomial and Poisson models. The results of the negative binomial and Poisson models are largely consistent with the OLS models. The estimated relationship between crime and building permits is negative across both cities, though for the negative binomial model the estimates in Chicago do not reach statistical significance (Appendix table C; Recall that for

CR IP T

negative binomial estimates, a coefficient less than 1 represents a negative relationship).

Finally, we test the sensitivity of the results to sample composition by excluding extremely long blockfaces (which are likely different in both criminogenic risk and investment

characteristics) from the samples in both cities (Appendix table D), and excluding the downtown

AN US

business districts from the samples (Appendix table E). The main results are robust to both types of sample restrictions. 5. Conclusion

M

Given the long-standing interest in community development and the perception that crime is an important barrier to progress, understanding the relationship between changes in crime and

ED

changes in private investment in neighborhoods is important. Despite the considerable attention that has been given to crime’s effect on neighborhoods, the precise relationship has not been

PT

demonstrated empirically. This research explores how changing patterns of micro-level crime

CE

affect private investment activity across a diverse set of neighborhoods in two major cities. The results can inform how we think about the influence that variation in crime has on how agents

AC

(property owners) allocate their private investment spatially across neighborhoods. Using detailed data from Los Angeles and Chicago, the analysis examines the relationship

between changes in crime on blockfaces and subsequent changes in the number of building permits pulled on that blockface. We find that changes in crime are negatively related to changes in building permit activity. The overall effects are small – the change in permits resulting from

22

ACCEPTED MANUSCRIPT

an additional crime on a blockface represents a 0.4 percent decline from the sample mean in Los Angeles, and a 1.1 percent decline from the sample mean in Chicago. The result is robust to multiple definitions of crime and the elimination of outliers and the main commercial district. Our analysis suggests that reactions to changing neighborhood crime levels are not

CR IP T

symmetric and investors do behave differently in increasing versus decreasing crime

environments. Given the overall context of declining crime in Los Angeles and Chicago during this time, the experiences of investors in declining crime neighborhoods are of particular interest. We find no overall effect of changes in crime in decreasing crime neighborhoods in either city.

AN US

As we delve more deeply into the specific context of these neighborhoods, we find that there is some effect of reductions in crime on permit activity on blockfaces experiencing larger changes in crimes in a year, and evidence that decreases in crime in these neighborhoods also triggers a

M

decrease in demolitions (in Los Angeles, 0.1 percent decline from the sample mean) and an increase in new construction (in both cities, 0.1 and 0.3 increase from the sample means). The

ED

magnitude of the effect of falling crime levels on building permits is much smaller than the effect of increases in crime. The bulk of the evidence generated here supports the overall conclusion –

PT

that declines in crime have little systematic effect on permit activity.

CE

The opposite story is true for those neighborhoods experiencing increases in crime, in the face of concurrent city-wide crime declines. In these places, increases in crime are found to

AC

reduce private investment. We find effects in Los Angeles that are twice as large as in Chicago (though the Chicago effect represents a large percent change from the sample mean), but little evidence of threshold effects in either city. Our analyses show that the crime trend in the neighborhood has a greater influence on permit activity than the initial crime level – suggesting that investors are more interested in the trajectory of crime in these neighborhoods than point-in-

23

ACCEPTED MANUSCRIPT

time measures. Further, increases in crime decrease new construction and renovations in both cities (0.2 and 1.8 percent declines from the sample means), but have little effect on demolitions. In Los Angeles, there are effects of similar magnitudes on permit activity for residential and commercial properties (both about a 0.6 percent decline from the sample mean), suggesting that

CR IP T

investors in Los Angeles consider similar factors in making these decisions. In Chicago

neighborhoods, the effects of increases in violent or property crime on permit activity (both approximately 7 percent declines from the sample mean) are larger than the effects of increases in public order crime (approximately 3 percent decline), suggesting that investors there are more

AN US

sensitive to serious crimes.

Overall, this research reflects an initial exploration into the question of how crime affects decisions of private parties. The results raise interesting questions regarding how the level and

M

changes in the level of crime contribute to a neighborhood’s trajectory and the efficacy of crimereduction strategies for promoting equitable economic development. In particular, the findings

ED

highlight several areas ripe for future research. For example, the analysis suggests that investors and others may distinguish between the presence of disorder (as reflected by building violations)

PT

and the presence of crime when making decisions about whether to invest in, move into, or move

CE

out of a neighborhood. Our analysis shows that the effects of crime persist over time, but a more detailed analysis could reveal more about the circumstances where effects might linger longer or

AC

phase out more rapidly. Variation could be associated with the type of crime, macroeconomic conditions, or demographic changes. It might also depend on broader trends affecting the block’s environment. Finally, our analysis is effectively a case study focusing on two cities; an interesting question is whether similar patterns are observed in other places (Ellen, Horn, and Reed, 2016).

24

ACCEPTED MANUSCRIPT

6. Acknowledgements

AC

CE

PT

ED

M

AN US

CR IP T

The authors thank Benjamin Robinson for outstanding research assistance and the Bedrosian Center on Governance and the Public Enterprise for research support. The authors are grateful for feedback from the participants of the 2014 Rena Sivitanidou Annual Research Symposium at the Lusk Center for Real Estate at the University of Southern California and the 2014 Israel Real Estate and Urban Economics Symposium. All errors and omissions are the authors’.

25

ACCEPTED MANUSCRIPT

7. References Abadie, A. and S. Dermisi (2008), “Is terrorism eroding agglomeration economies in central business districts? Lessons from the office real estate market in downtown Chicago,” Journal of Urban Economics, 64, 451-463. Aschauer, D. (1989). Is public infrastructure productive? Journal of Monetary Economics, 23, 177-200.

CR IP T

Becker. G. (1968), “Crime and punishment: an economic approach,” Journal of Political Economy, 76, 169-217.

Boggess, L. N. and J. R. Hipp (2010), “Violent crime, residential instability, and mobility: Does the relationship differ in minority neighborhoods?” Journal of Quantitative Criminology, 26, 351-370.

AN US

Bollinger, C. and R. Ihlanfeldt (2003), “The intraurban spatial distribution of employment: which government interventions make a difference?” Journal of Urban Economics, 53, 396412. Bostic, R. W. (2014), “Resilient economic development: Challenges and opportunities,” in Metropolitan Resilience in a Time of Economic Turmoil, M. A. Pagano (Ed.), University of Illinois at Chicago, University of Illinois Press, 58-80. Bostic, R. W. and A. C. Prohofsky (2006), “Enterprise zones and individual welfare: A case study of California,” Journal of Regional Science, 46(2), 175-203.

M

Burchfield, K. B. and E. Silver (2013), “Collective efficacy and crime in Los Angeles neighborhoods: Implications for the Latino paradox,” Sociological Inquiry, 83 (1), 154176.

ED

Corman , H. and N. Mocan (2005), Carrots, sticks, and broken windows. Journal of Law and Economics, 48(1), 235-266.

PT

Cullen, J. B. and S. D. Levitt (1999), “Crime, urban flight, and the consequences for cities,” The Review of Economics and Statistics, 81(2), 159-169.

CE

Chetty, R., N. Hendren, and L. F. Katz. (2016), "The effects of exposure to better neighborhoods on children: New evidence from the Moving to Opportunity experiment." The American Economic Review 106(4), 855-902.

AC

Ellen, I. G., Lacoe, J., & Sharygin, C. A. (2013). Do foreclosures cause crime?. Journal of Urban Economics, 74, 59-70. Ellen, I. G., & O’Regan, K. (2010). Crime and urban flight revisited: The effect of the 1990s drop in crime on cities. Journal of Urban Economics, 68 (3), 247-259. Ellen, I. G., Mertens Horn, K., & Reed, D. (2016). Has Falling Crime Invited Gentrification? New York, NY: Furman Center for Real Estate and Urban Policy. Gautier, P., A. Siegmann, and A.Van Vuuren (2009), “Terrorism and attitudes towards minorities: the effect of the Theo van Gogh murder on house prices in Amsterdam,” Journal of Urban Economics, 65, 113-126.

26

ACCEPTED MANUSCRIPT

Glaeser, E. L., J. Gyourko and R. Saks (2005), “Why is Manhattan so expensive? Regulation and the rise in housing prices,” Journal of Law and Economics, 48(2), 331-369. Groff, E. R., Weisburd, D., & Yang, S. M. (2010). Is it important to examine crime trends at a local “micro” level? A longitudinal analysis of street to street variability in crime trajectories. Journal of Quantitative Criminology, 26(1), 7-32. Harcourt, B.E. and J. Ludwig (2006), Broken windows: New evidence from New York City and a five-city social experiment, The University of Chicago Law Review, 73 (1), 271-320.

CR IP T

Katzman, M. T. (1980), “The contribution of crime to urban decline,” Urban Studies, 17(3), 277286. Kelling, G. L. and J. Q. Wilson (1982), “Broken windows: The police and neighborhood safety,” The Atlantic Monthly, March. Knaap, G. (1985), The price effects of urban growth boundaries in metropolitan Portland, Oregon, Land Economics, 64(1), 26-35.

AN US

Kolko, J. and D. Neumark (2010), Do some enterprise zones create jobs? Journal of Policy Analysis and Management, 29(1), 5-38.

Krivo, L. J. and R. D. Peterson (1996), “Extremely disadvantaged neighborhoods and urban crime,” Social Forces, 75 (2), 619-648. Lacoe, J., & Ellen, I. G. (2015), Mortgage foreclosures and the changing mix of crime in microneighborhoods. Journal of Research in Crime and Delinquency, 52 (5), 717-746

M

Lens, M. C. and R. Meltzer (2016), “Is crime bad for business? Crime and commercial property values in New York City,” Journal of Regional Science, 56 (3), 442-470.

ED

Levitt, S. (2004), “Understanding why crime fell in the 1990s: four factors that explain the decline and six that do not,” Journal of Economic Perspectives, 18, 163-190.

PT

R. W. and R. W. Bostic (2003), “Black home-owners as a gentrifying force? Neighbourhood dynamics in the context of minority home-ownership,” Urban Studies, 40(12), 24272449.

CE

Morrison, C. and A. Schwartz (1996), Public infrastructure, private input demand, and economic performance in New England manufacturing, Journal of Business and Economic Statistics, 14(1), 91-101.

AC

Munnell, A. (1990), How does public infrastructure affect regional economic performance, New England Economic Review, September/October, 11-32. Papke, L. E. (1993), Tax policy and urban development: Evidence from the Indiana enterprise zone program. Journal of Public Economics, 54, 37-49. Pope, J. (2008), “Fear of crime and housing prices: households reactions to sex offender registries,” Journal of Urban Economics, 64, 601-614. Taylor, R. B. (1995), “The impact of crime on communities,” The Annals of the American Academy of Political and Social Science, 539 (1), 28-45.

27

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

M

AN US

CR IP T

Taylor, R. B., Steve D. Gottfredson, and Sidney Brower. (1985). “Attachment to place: Discriminant validity and impacts of disorder and diversity," American Journal of Community Psychology, 13, 525-42. Twinam, T. (2017), Danger zone: Land use and the geography of neighborhood crime. Journal of Urban Economics, 100, 104-119.

28

ACCEPTED MANUSCRIPT

8. Figures

Figure 1. Crime trends for Los Angeles and Chicago, 2006-2011 8,00,000 Total Crime (LA)

6,00,000

CR IP T

Violent Crime (LA)

5,00,000

Property Crime (LA)

4,00,000 3,00,000

Total Crime (CHI)

2,00,000

Violent Crime (CHI)

1,00,000

Property Crime (CHI)

2006

2007

2008

AN US

Number of reported crimes

7,00,000

2009

2010

2011

Notes. Total crime includes all reported crimes, violent crime includes murder/homicide, aggravated assault, rape, and robbery, and property crime includes burglary, theft, arson, and motor vehicle theft.

Figure 2. Permit trends for Los Angeles and Chicago, 2006-2011

M

80,000 70,000

ED

60,000 50,000

PT

40,000 30,000

10,000

2006

AC

-

CE

20,000

2007

2008 Los Angeles

2009

2010

2011

Chicago

29

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

M

AN US

CR IP T

Figure 3. Average Number of Crimes per Blockface, Los Angeles

30

AC

CE

PT

ED

M

AN US

Figure 4. Average Number of Permits per Blockface, Los Angeles

CR IP T

ACCEPTED MANUSCRIPT

31

CR IP T

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

M

AN US

Figure 5. Average Number of Crimes per Blockface, Chicago

32

AC

CE

PT

ED

M

AN US

CR IP T

ACCEPTED MANUSCRIPT

33

CR IP T

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

M

AN US

Figure 6. Average Number of Permits per Blockface, Chicago

34

AC

CE

PT

ED

M

AN US

CR IP T

ACCEPTED MANUSCRIPT

35

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

M

AN US

CR IP T

Figure 7. Distribution of increasing crime tracts, Los Angeles (2005-2010)

36

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

M

AN US

CR IP T

Figure 8. Distribution of increasing crime tracts, Chicago (2005-2010)

37

ACCEPTED MANUSCRIPT

9. Tables Table 1. Selected characteristics of Los Angeles and Chicago Characteristics Citywide characteristics (2010)

Chicago

Population

3,792,62

2,695,598

Density (persons per sq. mile)

8,092.3

11,841.8

Homeownership rate

38.0%

46.1%

Distribution of housing units by type Single family 2-4 units 5+ units Other (mobile home, RV, boat, etc.)

45.5% 8.9% 44.9% 0.7%

29.6% 31.7% 38.5% 0.2%

$49,745

$47,408

21.6%

22.5%

Median household income Percent population below poverty line

ED

Mean permits per blockface New building permits Renovation permits Demolition permits

9.2% 48.5% 28.7% 11.1% 2.6%

32.4% 28.9% 31.7% 5.4% 1.7%

6.80 0.82 2.91 0.80

3.91 0.22 1.02 1.30

0.72 0.04 0.51 0.02

0.12 0.02 0.09 0.01

M

Mean crimes per blockface Violent crime Property crime Public order crime

AN US

Percent population by race/ethnicity Black Hispanic White Asian Other Neighborhood characteristics (2006-2011)

CR IP T

Los Angeles

AC

CE

PT

Mean building code violations per blockface 0.43 1.33 Notes. Citywide characteristic from the 2010 Census and 2010 American Community Survey (5-year estimates). Neighborhood characteristics are averages across blockfaces over the study period (2006-2011) from Los Angeles Police Department, Chicago Police Department, Los Angeles Department of Building and Safety, and the Chicago Department of Buildings.

38

ACCEPTED MANUSCRIPT

Table 2. The relationship between building permits and crime, selected tests for the appropriate lag structure, crime annual measures VARIABLES

(1)

(2)

(3)

(4)

(5)

-0.00295**

-0.00301**

-0.00358***

-0.00350***

-0.00346**

(0.00137)

(0.00138)

(0.00135)

(0.00135)

(0.00136)

-0.00265

-0.00319*

-0.00313*

-0.00307*

(0.00163)

(0.00164)

(0.00163)

(0.00163)

-0.00307**

-0.00289**

-0.00276**

(0.00138)

(0.00136)

(0.00132)

0.00160

0.00153

(0.00134)

(0.00131)

Crime (t-1) Crime (t-2) Crime (t-3) Crime (t-4)

CR IP T

A. Los Angeles

0.00116

Crime (t-5)

(0.00161)

Observations R-squared

0.0100**

(0.00468)

(0.00469)

497,443

497,443

0.468

0.468

-0.00132**

-0.00130**

(0.000564)

(0.000558) -0.000164

(0.000414)

Crime (t-1)

ED

Crime (t-2) Crime (t-3)

CE

Crime (t-5) Violations (t-1)

AC

Observations

0.0100**

(0.00468)

(0.00468)

(0.00468)

497,443

497,443

497,443

0.468

0.468

0.468

-0.00128**

-0.00122**

-0.00118**

(0.000560)

(0.000559)

(0.000560)

-0.000257

-0.000213

-0.000175

(0.000392)

(0.000395)

(0.000396)

0.000334

0.000118

0.000180

(0.000341)

(0.000318)

(0.000321)

PT

Crime (t-4)

0.0100**

M

B. Chicago

0.0101**

AN US

Violations (t-1)

0.0100**

0.000629*

0.000427

(0.000346)

(0.000300) 0.000629 (0.000405)

0.00144***

0.00144***

0.00144***

0.00144***

0.00144***

(0.000481)

(0.000481)

(0.000481)

(0.000481)

(0.000481)

514,843

514,843

514,843

514,843

514,843

R-squared 0.557 0.557 0.557 0.557 0.557 Note. Robust standard errors in parentheses, clustered at the census tract level. All models include blockface fixed effects and year*census tract fixed effects. *** p<0.01, ** p<0.05, * p<0.1

39

ACCEPTED MANUSCRIPT

Table 3. Baseline results: Relationship between crime and building permits VARIABLES

(1)

(2)

(3)

(4)

(5)

0.00820***

0.00823***

-0.00370***

-0.00322**

-0.00301**

(0.00135)

(0.00137)

(0.00135)

(0.00133)

(0.00138)

0.0100***

0.00981***

-0.00257

-0.00233

-0.00265

(0.00155)

(0.00156)

(0.00166)

(0.00164)

(0.00163)

0.160***

0.160***

0.0101**

0.00994**

0.0100**

(0.0368)

(0.0370)

(0.00471)

(0.00466)

(0.00469)

497,443

497,443

497,443

497,443

497,443

0.020

0.022

0.453

0.454

0.468

0.00731***

0.00734***

-0.00144**

-0.00144**

-0.00130**

(0.000653)

(0.000649)

Crime (t-2)

0.00785***

0.00763***

(0.000837)

(0.000852)

(0.000416)

(0.000418)

(0.000414)

Violations (t-1)

0.00933***

0.00976***

0.00167***

0.00158***

0.00144***

(0.00178)

(0.00182)

(0.000478)

(0.000478)

(0.000481)

514,843

514,843

514,843

514,843

514,843

0.052

0.053

0.548

0.549

0.557

Year FE

No

Yes

Yes

Yes

Yes

Blockface FE

No

No

Yes

Yes

Yes

Police District FE

No

No

No

Yes

No

Crime (t-1) Crime (t-2) Violations (t-1)

Observations R-squared

R-squared

M

Observations

ED

Crime (t-1)

AN US

B. Chicago

CR IP T

A. Los Angeles

(0.000564)

(0.000561)

(0.000558)

-0.000305

-0.000317

-0.000164

AC

CE

PT

Census Tract FE No No No No Yes Note. Robust standard errors in parentheses, clustered at the census tract level. *** p<0.01, ** p<0.05, * p<0.1.

40

ACCEPTED MANUSCRIPT

Table 4. Granger causality tests Los Angeles

Crime (t-1) Crime (t-2)

(1)

(2)

(3)

(4)

-0.00301**

-0.00283**

-0.00130**

-0.00129**

(0.00138)

(0.00142)

(0.000558)

(0.000564)

-0.00265

-0.00257

-0.000164

-0.000114

(0.00163)

(0.00168)

(0.000414)

(0.000413)

-0.00243

Crime (t)

0.000037

(0.00184)

(0.000023)

0.00310

Crime (t+1)

0.000604

(0.00212)

Observations

(0.000643)

0.0100**

0.0100**

0.00144***

(0.00469)

(0.00469)

(0.000481)

497,443

497,443

0.00146*** (0.000480)

AN US

Violations (t-1)

CR IP T

VARIABLES

Chicago

514,843

514,843

AC

CE

PT

ED

M

0.468 0.557 R-squared 0.468 0.557 Note. Robust standard errors in parentheses, clustered at the census tract level. All models include blockface fixed effects and year*census tract fixed effects. *** p<0.01, ** p<0.05, * p<0.1.

41

ACCEPTED MANUSCRIPT

Table 5. Relationship between crime and building permits, by neighborhood crime trend over study period (2006-2010)

Crime (t-1) Crime (t-2) Violations (t-1)

-0.000482

-0.00793***

(0.00151)

(0.00263)

0.000328 (0.000640)

-0.00432*** (0.00105)

-0.00183

-0.00449

0.000654

-0.00209**

(0.00196)

(0.00387)

(0.000454)

(0.000864)

0.00940

0.0101

0.00157**

0.00121**

(0.00575)

(0.00734)

(0.000619)

(0.000616)

355,689

141,754

378,458

136,385

AN US

Observations

Chicago (3) (4) Crime Crime Decrease Increase

CR IP T

VARIABLES

Los Angeles (1) (2) Crime Crime Decrease Increase

AC

CE

PT

ED

M

R-squared 0.479 0.478 0.585 0.513 Note. Robust standard errors in parentheses, clustered at the census tract level. All models include blockface fixed effects and year*census tract fixed effects. *** p<0.01, ** p<0.05, * p<0.1.

42

ACCEPTED MANUSCRIPT

Table 6. Estimates using threshold crime levels Los Angeles

Crime(j=4)(t-1) Crime(j=5)(t-1) Crime(j=6)(t-1) Crime(j=7)(t-1) Crime(j=8)(t-1) Crime(j=9)(t-1) Crime(j=10+)(t-1)

0.00331

-0.00870

-0.00577***

-0.000376

(0.0102)

(0.0217)

(0.00173)

(0.00374)

0.0188

-0.0161

-0.00493*

-0.00241

(0.0134)

(0.0211)

(0.00264)

(0.00518)

0.0179

-0.0220

-0.00679*

-0.00120

(0.0194)

(0.0267)

(0.00397)

(0.00697)

-0.0108

-0.00366

-0.00535

-0.00189

(0.0174)

(0.0299)

(0.00564)

(0.00956)

-0.0324

-0.0116*

-0.00363

7.10e-06 (0.0199)

(0.0339)

0.00292

0.0120

(0.0212)

(0.0456) 0.000402

0.00904

(0.00674)

(0.0109)

-0.00354

-0.0365***

(0.00822)

(0.0119)

-0.0168*

-0.0136

(0.00873)

(0.0159)

-0.0200*

-0.00757

(0.0109)

(0.0162)

(0.0223)

(0.0569)

0.0141

-0.0565

(0.0275)

(0.0570)

-0.0265

-0.0442

0.00459

-0.0328*

(0.0300)

(0.0460)

(0.0121)

(0.0193)

0.0111

-0.0214

-0.0127

-0.0416**

(0.0854)

(0.00922)

(0.0168)

(0.0255) Violations (t-1)

CR IP T

Crime(j=3)(t-1)

(4) Crime Increase

AN US

Crime(j=2)(t-1)

(3) Crime Decrease

0.00938

PT

(0.00574)

M

Crime(j=1)(t-1)

(2) Crime Increase

ED

VARIABLES

(1) Crime Decrease

Chicago

0.0101

(0.00726)

0.00160*** (0.000618)

0.00109* (0.000616)

378,458 136,385 0.478 0.585 R-squared 0.479 0.513 Note. J connotes the number of crimes represented by the dummy variable (i.e. j=2 is the dummy variable for blockfaces with 2 crimes). Robust standard errors in parentheses, clustered at the census tract level. All models include blockface fixed effects and year*census tract fixed effects. *** p<0.01, ** p<0.05, * p<0.1. 355,689

141,754

AC

CE

Observations

43

CR IP T

ACCEPTED MANUSCRIPT

Table 7. Relationship between crime and building permits, by initial crime level (2005 tract crime rate by 2000 tract population) (1) Lowest Crime

Observations R-squared B. Chicago Crime (t-1) Crime (t-2) Violations (t-1)

-0.00759

-0.00172

-0.00126

(0.00499)

(0.00251)

(0.00219)

(5) Lowest Crime -0.00516

Crime increase (6) (7) 2nd 3rd

-0.00233

AN US

Violations (t-1)

-0.00272 (0.00756)

(0.0123)

(0.00706)

(8) Highest Crime

-0.0146***

-0.00316

(0.00524)

(0.00383)

-0.00928*

-0.0117***

-0.00304

-0.00266

-0.0142

-0.0119**

-0.00784*

-0.00839**

(0.00536)

(0.00412)

(0.00255)

(0.00308)

(0.00975)

(0.00608)

(0.00402)

(0.00329)

0.0175

0.0119

0.0137*

-0.00217

0.00331

0.00521

0.00655

0.0161

(0.0108)

(0.0115)

(0.00772)

(0.0115)

(0.0147)

(0.00491)

(0.0106)

(0.0173)

135,326 0.487

54,774 0.549

86,146 0.468

-0.00108

-0.00291

-0.00111

(0.00149)

(0.00189)

(0.00103)

79,443 0.526

-0.00608**

-0.00743***

-0.00304**

(0.00289)

(0.00278)

(0.00130)

-0.00201*

-0.000707

0.000469

(0.000883)

(0.000561)

0.00257***

0.00335***

(0.000949)

29,758 0.738

-0.0114***

(0.00113) (0.000972)

43,968 0.491

(0.00380)

-0.00167 -0.00495

27,750 0.536

-0.000401

(0.00102) (0.00788)

40,278 0.500

(0.000731)

M

Crime (t-2)

(4) Highest Crime

ED

A. Los Angeles Crime (t-1)

Crime decrease (2) (3) 2nd 3rd

0.00203*** (0.000462)

-0.00711**

-0.00374

-0.00337

-0.00186*

(0.00281)

(0.00314)

(0.00229)

(0.00110)

0.00129

-0.000430

(0.00217)

(0.00133)

0.00302** (0.00139)

0.00150* (0.000789)

AC

CE

PT

Observations 139,164 70,977 60,400 107,917 39,649 23,773 25,588 47,375 R-squared 0.710 0.353 0.561 0.581 0.448 0.514 0.549 0.568 Note. The fourth quartile includes those block faces with the highest levels of crime. The census tract crime rate in 2005 is calculated using the 2000 tract population. Robust standard errors in parentheses, clustered at the census tract level. All models include blockface fixed effects and year*census tract fixed effects. *** p<0.01, ** p<0.05, * p<0.1.

44

Table 8. Relationship between crime and building permits, by permit type

Violations (t-1)

Observations R-squared B. Chicago Crime (t-1) Crime (t-2)

Renovation

0.000356

-0.00137

(0.000500)

(0.00115)

Observations

0.000583** (0.000242)

(5)

(6)

Residential

Commercial

New

7.03e-05

-0.000552

-0.00115*

(0.00100)

(0.00107)

(0.000682)

-0.000761***

-5.31e-05

3.73e-06

0.000724

(0.000295)

(0.00176)

(0.000185)

(0.00163)

0.00190*

0.00523

0.000786

0.00759

(0.00113)

(0.00514)

(0.000661)

(0.00555)

355,689

355,689

355,689

355,689

0.242

0.470

0.253

0.423

4.34e-05

0.000649

-0.000365

(0.000173)

(0.000532)

(0.000276)

-0.000364** (0.000168)

Violations (t-1)

Demolition

(4)

-0.00101***

0.000775** (0.000379) 0.00230***

(0.000365)

(0.000347)

378,458

378,458

0.000242 (0.000204) 0.000282*

(7)

Crime increase (8)

Renovation

-0.00453**

(0.00186)

AN US

Crime (t-2)

New

Crime decrease (3)

Demolition 0.000143

(0.000403)

(9)

(10)

Residential

Commercial

-0.00493***

-0.00301

(0.00186)

(0.00184)

-0.00255**

-0.000726

-0.00179

0.000352

-0.000157

-0.00433**

(0.000992)

(0.000605)

(0.00351)

(0.000368)

(0.00341)

(0.00174)

0.00182

0.000432

0.00865

-0.000380

0.00952

0.000610

(0.00243)

(0.000921)

(0.00591)

(0.000486)

(0.00681)

(0.00156)

355,689

141,754

141,754

141,754

141,754

141,754

0.642

0.289

0.463

0.282

0.448

0.604

M

A. Los Angeles Crime (t-1)

(2)

ED

VARIABLES

(1)

CR IP T

ACCEPTED MANUSCRIPT

-0.00221***

-0.00194**

-0.000174

(0.000458)

(0.000913)

(0.000217)

-0.00134***

-0.000710

-3.95e-05

(0.000400)

(0.000702)

(0.000182)

-0.000812***

(0.000144)

(0.000200)

378,458

136,385

0.00178***

0.000239

(0.000503)

(0.000163)

136,385

136,385

AC

CE

PT

R-squared 0.292 0.609 0.327 0.314 0.542 0.270 Note. Robust standard errors in parentheses, clustered at the census tract level. All models include blockface fixed effects and year*census tract fixed effects. *** p<0.01, ** p<0.05, * p<0.1.

45

ACCEPTED MANUSCRIPT

Table 9. Relationship between crime and building permits, by crime type Chicago

Los Angeles (1) Crime decrease A. Violent Crime Crime (t-1)

(2) Crime increase

(3) Crime decrease

(4) Crime increase

-0.00722

-1.63e-05

-0.00831**

(0.00241)

(0.00417)

Crime (t-2)

(0.00371) -0.00415

(0.00537) -0.00131

0.000447

-3.63e-06

(0.00524)

(0.00122)

(0.00202)

Violations (t-1)

(0.00391) 0.00942 (0.00575)

(0.00728)

R-squared B. Property crime Crime (t-1)

355,689

141,754

0.479

0.478

0.00335

Crime (t-2)

(0.00429) 0.00146

Violations (t-1)

(0.00441) 0.00940 (0.00574) 355,689

C. Public order crime Crime (t-1)

378,458

136,385

0.585

0.513

4.36e-05

(0.00581)

(0.00169)

(0.00240)

-0.00973*

0.000307

-0.00401*

(0.00501)

(0.00151)

(0.00219)

0.0101

(0.00733)

0.00159***

-0.00834***

0.00105*

(0.000617)

(0.000618)

378,458

136,385

0.478

0.585

0.513

-0.00385

-0.0117 (0.00785)

0.00109 (0.000883)

-0.00326* (0.00180)

0.00306

0.00112

-0.00245*

(0.00603) 0.00943

(0.0106)

(0.000697)

(0.00143)

(0.00575) 355,689

(0.00730)

0.0101 141,754

0.00161***

0.00105*

(0.000618)

(0.000618)

378,458

136,385

CE

Observations

PT

Violations (t-1)

(0.000623)

141,754

(0.00652) -0.00651

Crime (t-2)

(0.000623)

0.479

ED

R-squared

0.00102

-0.00821

M

Observations

0.00161***

AN US

Observations

0.0101

CR IP T

-0.00451

AC

0.479 0.478 0.585 0.513 R-squared Note. Robust standard errors in parentheses, clustered at the census tract level. All models include blockface fixed effects and year*census tract fixed effects *** p<0.01, ** p<0.05, * p<0.1.

46

ACCEPTED MANUSCRIPT

10. Appendix

(1)

(2)

(3)

(4)

(5)

0.00999***

0.00999***

-0.00367***

-0.00319**

-0.00299**

(0.00129)

(0.00130)

(0.00135)

(0.00133)

(0.00138)

0.0112***

0.0110***

-0.00256

-0.00233

-0.00265

(0.00151)

(0.00152)

(0.00166)

(0.00164)

(0.00163)

497,443

497,443

497,443

497,443

497,443

0.010

0.012

0.453

0.453

0.468

0.00884***

0.00885***

-0.00139**

-0.00140**

-0.00126**

(0.000535)

(0.000537)

(0.000563)

(0.000561)

(0.000558)

0.00898***

0.00893***

-0.000246

-0.000255

-0.000101

(0.000755)

(0.000757)

(0.000416)

(0.000419)

(0.000414)

514,843

514,843

514,843

514,843

514,843

0.047

0.048

0.548

0.549

0.557

Year FE

No

Yes

Yes

Yes

Yes

Blockface FE

No

M

Table A. Relationship between crime and building permits without building code violations VARIABLES

No

Yes

Yes

Yes

No

No

Yes

No

Crime (t-1) Crime (t-2)

Observations R-squared

Crime (t-2)

Observations R-squared

Police District FE

ED

Crime (t-1)

AN US

B. Chicago

No

CR IP T

A. Los Angeles

AC

CE

PT

Census Tract FE No No No No Yes Note. Robust standard errors in parentheses, clustered at the census tract level. *** p<0.01, ** p<0.05, * p<0.1. These specifications are identical to those reported in Table 3 except for the exclusion of building code violations.

47

ACCEPTED MANUSCRIPT

(1)

(2)

VARIABLES

Los Angeles

Chicago

Crime (t-1)

-0.00282

-0.00307**

(0.00240)

(0.00146)

-0.00732***

-0.00375***

(0.00206)

(0.00116)

Crime (t-2) Violations (t-1)

0.00508***

0.000967*** (0.000226)

497,443

514,843

0.480

0.564

Tract-Year FE

Yes

Yes

Blockface FE

Yes

Yes

Observations R-squared

AN US

(0.00161)

CR IP T

Table B. Relationship between crime and building permits, Robustness #1, Log Transformation

AC

CE

PT

ED

M

CT Clustered SE Yes Yes Note. Robust standard errors in parentheses, clustered at the census tract level. All variable are added 1 so that log (0+1)=0. All models include blockface fixed effects and year*census tract fixed effects. *** p<0.01, ** p<0.05, * p<0.1.

48

ACCEPTED MANUSCRIPT

Table CC. Relationship between crime and building permits, Robustness #2, Negative Binomial and Poisson Models Poisson

Negative Binomial (2)

(1)

Los Angeles

Chicago

Los Angeles

Crime (t-1)

0.996538***

0.998688

-0.00372***

-0.00502***

(0.000755)

(0.00104)

(0.0005605) Crime (t-2)

0.995286***

(0.00146) 0.998789

(0.0005412) Violations (t-1)

0.99955 (0.0011758)

(0.000226)

340,277

Groups

39,168

Log likelihood

-137,857.99

Tract-Year FE

No

Police precinct-Year FE

Yes

Blockface FE

Yes

27,624

-0.00417***

(0.000791)

-0.00119

(0.000960)

0.00432**

0.00501***

(0.00186)

(0.000567)

339,390

144,220

10,144

39,168

-18,984.86

-137,857.99

No

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

AN US

Observations

(0.00116) 1.000251

CR IP T

(1) VARIABLES

AC

CE

PT

ED

M

Yes CT Clustered SE Yes Yes Yes Note. Estimates presented for the negative binominal models are incidence rate ratios. Samples are limited to blockfaces that have at least 1 permit during the period. Robust standard errors in parentheses, clustered at the census tract level. Models include blockface fixed effects and year*police precinct or year*census tract fixed effects. *** p<0.01, ** p<0.05, * p<0.1

49

ACCEPTED MANUSCRIPT

Table DD. Relationship between crime and building permits, Robustness #3, Trimmed sample of extreme blockface length Los Angeles (1)

Chicago

(2) Drop top 5% by blockface length

(3) Standard

(4) Drop top 5% by blockface length -0.00131**

Standard

Crime (t-1)

-0.00301**

-0.00289**

-0.00130**

(0.00138)

(0.00132)

(0.000558)

Crime (t-2) Violations (t-1)

(0.000551)

-0.00265

-0.00293*

-0.000164

-0.000419

(0.00163)

(0.00166)

(0.000414)

(0.000422)

0.0100**

0.00944**

(0.00469) 497,443

0.00144***

0.00167***

(0.00476)

(0.000481)

(0.000501)

481,842

514,843

490,614

AN US

Observations

CR IP T

VARIABLES

AC

CE

PT

ED

M

R-squared 0.468 0.457 0.557 0.561 Note. Robust standard errors in parentheses, clustered at the census tract level. All models include blockface fixed effects and year*census tract fixed effects. *** p<0.01, ** p<0.05, * p<0.1.

50

ACCEPTED MANUSCRIPT

Table EE. Relationship between crime and building permits, Robustness #4, Trimmed sample of city centers Los Angeles (1) Los Angeles Standard

Crime (t-1) Crime (t-2) Violations (t-1)

-0.00301**

-0.00275**

-0.00130**

(0.00138)

(0.00139)

(0.000558)

(4) Loop Excluded -0.00149*** (0.000557)

-0.00265

-0.00242

-0.000164

-0.000248

(0.00163)

(0.00166)

(0.000414)

(0.000409)

0.0100**

0.0101**

0.00144***

0.00155***

(0.00469)

(0.00472)

(0.000481)

(0.000478)

497,443

491,146

514,843

511,066

AN US

Observations

(3) Chicago Standard

CR IP T

VARIABLES

Chicago

(2) Downtown Excluded

AC

CE

PT

ED

M

R-squared 0.468 0.460 0.557 0.496 Note. Robust standard errors in parentheses, clustered at the census tract level. All models include blockface fixed effects and year*census tract fixed effects. *** p<0.01, ** p<0.05, * p<0.1.

51