Persistence in urban form: The long-run durability of employment centers in metropolitan areas

Regional Science and Urban Economics 39 (2009) 224–232

Contents lists available at ScienceDirect

Regional Science and Urban Economics j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / r e g e c

Persistence in urban form: The long-run durability of employment centers in metropolitan areas☆ Christian L. Redfearn School of Policy, Planning, and Development, University of Southern California, United States

a r t i c l e

i n f o

Article history: Received 5 December 2007 Received in revised form 20 September 2008 Accepted 20 September 2008 Available online 14 October 2008 JEL classification: R11 R12 R13 R30 R33 R00

a b s t r a c t This paper documents a marked persistence in the spatial distribution of employment in the Los Angeles Metropolitan Area. Over a medium-term of twenty years — a period of pronounced growth and change in the region's employment and population — lagged employment density dominates access variables in explaining levels and ranks of current employment density. Similarly, the probability that a tract is located within a current employment center is largely a function of past membership rather than proximity to highways or the central business district. Moreover, longer-term persistence is also readily apparent: concentrations of employment a century ago explain the current distribution of employment as well as access to the modern highway system. This stability in the location of employment and employment concentrations over mediumand longer-terms suggests important roles for agglomeration, adjustment costs, and the durability of fixed investment in modeling the evolution of metropolitan areas. © 2008 Elsevier B.V. All rights reserved.

Keywords: Persistence Urban form Spatial distribution of employment Fixed-investment

1. Introduction Recently, much has been made of rising gas prices and the evolution of urban areas toward greater density. Of course, this follows years of discussions about “sprawl” and longer-term location trends associated with lower transportation and communication costs. One element under-emphasized in both discussions is the role of fixed investment in urban areas — durable capital in the form of roads, structures, and institutions that acts as an impediment to changes in urban form resulting from marginal changes in the fundamental variables of firm and household location choice. One implication of the durability of this capital for urban areas is that past location decisions — even those made in the distant past — may persist for far longer than is generally acknowledged in urban modeling. This persistence is readily apparent in the layout of streets in the cores of older metropolitan areas, which remain as they were 200 years ago. This paper documents a similar persistence well outside the historic core of a major metropolitan area.

☆ Thanks to the participants of USC Lusk Center's 2006 Rena Sivitanidou Annual Research Symposium as well as those at the University of British Columbia's Sauder School of Business seminar, and to Dan McMillen, David Frame, Nate Baum-Snow, Greg Hise, Tsur Sommerville, Jennifer Gold, and two anonymous referees for helpful comments. This research has benefited from the assistance of Cara Mullio. E-mail address: [email protected]. 0166-0462/$ – see front matter © 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.regsciurbeco.2008.09.002

The existence of such persistence would be significant for modeling urban phenomena. Many models of urban form are ahistorical, with little focus on fixed investment or adjustment costs; Brueckner (2000) summarizes it thusly in the context of modeling housing markets: The element of time is deemphasized in the static model, and this is permissible because of the implicit assumption that housing capital is malleable. This assumption allows the city to be rebuilt every period as underlying conditions change, which means that the city's spatial structure can be predicted at any point in time without regard to its past history. Because the model has been successful at explaining the broad features of existing cities, this suppression of history is justified. However, the model fails to predict some aspects of urban form on a more detailed level, and this failure is connected to its ahistorical character. (p263–4). The “broad feature” explained in traditional urban models is monocentricity. However, with the CBDs of major metropolitan areas containing on the order of a tenth of metropolitan area employment, it seems relevant to explore firm location patterns beyond the CBD.1 In 2000, for example, only 2.5% of metropolitan area employment in Los

1 Source: U.S. Census Bureau. Among the top 50 metropolitan areas in 2000, the median share of metropolitan employment in the CBD is 9.95%; for 1990 it was 8.29%.

C.L. Redfearn / Regional Science and Urban Economics 39 (2009) 224–232

Angeles was located in the central business district. The remaining 97.5% is spread irregularly throughout the region, with many concentrations of employment outside the central business district (Redfearn, 2007; Giuliano and Redfearn, 2007). Though polycentricity plays an obvious and central role in this research, it should be noted that polycentricity is not the main finding of this paper. The multi-nodal nature of American cities has long been recognized, popularly by Garreau (1991) and by many academics (McMillen and Smith, 2003; McMillen, 2001; Craig and Ng, 2001; Anderson and Bogart, 2001; Bogart and Ferry, 1999; McMillen and McDonald, 1998; Cervero and Wu, 1997; Gordon and Richardson, 1996; McDonald and Prather, 1994; Giuliano and Small, 1991). Despite the growing list of papers on polycentricity, there is little understanding of the empirical characteristics of employment centers within a metropolitan area. This paper focuses on one apparent regularity: persistence in the location of employment over medium- and longer-terms. In fact, polycentric organization of employment in Los Angeles today has its origins in the region's proto-centers, concentrations of employment that existed well before the region could be thought of as metropolitan at all. Indeed, it is possible to look at maps from the years around 1900 and see the same set of place names that one can find in today's commercial brokerage maps. It is in these places that economic activity continues to exist at far greater densities than other places in the region, just as it did a hundred years ago. Clearly, development and redevelopment have taken place, but in a manner that broadly preserves the relative ordering of locations with regard to employment density. This paper does not address the specific forces that led to the initial formation of the proto-centers or to their growth and decline over time. These are important issues, but ones left for future research. Rather, the focus here is the extent of the persistence in the spatial distribution of employment during a period in which all of the basic variables used in urban modeling varied widely (Glaeser and Kohlhase, 2003). This is neither the first paper on persistence in an urban setting nor the first on the role of long-lived fixed investment in influencing urban outcomes. A handful of papers have examined the role of durable capital in urban form (see Wheaton, 1982) and, in particular, durable housing (Glaeser and Gyourko, 2005; Harrison and Kain, 1974; Anas, 1978; Arnott, 1980). These papers, however, appeal to durability to explain short-run dynamics, not long-term persistence. One paper on path dependence in the urban context is Arthur (1988), but its focus is on path dependence arising from the initial location choice of influential firms, resulting in industrial concentration like that found in Silicon Valley. Cronon (1991) writes explicitly about fixed investment (the accumulation of a “second nature”) and its lasting impact on the development of Chicago into a major metropolitan area. In the case of persistence in the employment centers of Los Angeles, agglomeration, fixed investment, and an ossification of the rights of way available for improvements for transit infrastructure all play important roles. There are, of course, many papers on agglomeration, and an excellent summary can be found in Anas et al. (1998). This article addresses the forces of concentration and deconcentration in both monoand polycentric contexts. Papers on redevelopment and where it occurs include Dye and McMillen (2007), Munneke (1996), Rosenthal and Helsley (1994), and Brueckner (1980) among others. Braid (1995) addresses the persistence of transit systems as housing crowds rightsof-way, making expansion or reorganization difficult. It is left to future research to tie these various threads together to shed light on the precise mechanism that leads to persistence. What this paper demonstrates is that over medium- and longerterms, the spatial organization of economic activity in metropolitan Los Angeles is remarkably stable, despite a backdrop of pronounced growth in population and employment and pronounced change in the cross-sectional nature of both. Over the 1980–2000 period, the metropolitan area added a net four million residents and a millionand-a-half jobs. But these figures understate the dynamics of the

225

region: the population evolved from a white majority to a Hispanic plurality; the employment base experienced major shocks to manufacturing and to defense industries at the end of the Cold War; and, transportation and communications costs continued their long downward trends. This dynamic context makes the finding of stability more surprising than it might have been if the study area were static along these dimensions. In the medium term, a panel of employment data based on Census tracts is used to demonstrate that the location of employment lagged twenty years dominates access to the CBD and access to the current network highways in explaining current employment location. Both in levels and ranks, approximately three-quarters of the spatial variation in employment density is explained by lagged employment density. Regressed alone, the coefficient on distance to the central business district is highly significant and negative; regressed with distance to the nearest highway, the two negative and significant coefficients offer support for the basic urban model in which access is the fundamental variable. But, when employment density — lagged up to twenty years — is included, not only is the explanatory power vastly improved, the coefficient on the distance to the CBD is either insignificant or significantly positive and the coefficient on highway access falls by a factor of ten. In other words, holding constant the spatial distribution of employment from twenty years prior largely undermines the explanatory power of access to the CBD and to highways. Over this medium term, it appears that decisions made at least twenty years earlier continue to govern current urban form. Interestingly, old highway networks are more relevant in explaining the current spatial distribution of employment than is the modern highway system. Unlike the modern highway system, this antecedent system evolved in order to connect early centers of employment. And, it is these early centers that provide the evidence for longer-term persistence. While the medium-term persistence is somewhat surprising, it is a longer-term stability that is most noteworthy. Using historic maps, data on early municipal incorporations, and other historical data, this paper shows that the location of economic activity as of the turn of the last century is as good a predictor of current economic activity as the modern highway system. Access to the modern highway system — largely completed by 1960 — appears to be a necessary condition for locations to support concentrations of employment, but not sufficient. Lack of proximity to highways is consistent with a lack of employment density, but there are hundreds of miles of highway without proximal agglomerations of employment. There may be any number of contributing factors to the persistence in the spatial distribution of employment over time, and further research is necessary to sort through them. They are likely to include fixedinvestment in buildings and infrastructure, agglomeration, coordination problems, regulation and others. Whatever they may be, the existence of persistence of such magnitude and length should give pause when considering an application of a static model to questions about changes in urban form. In the case of Los Angeles, there appears to be much that resists change in the location of employment over time. The paper is organized as follows. Section 2 introduces the data and documents the dynamic context of employment and population change within the Los Angeles metropolitan area, against which the stability of employment location should be contrasted. Several measures of persistence are developed and reported in Section 3. These include tests of persistence in the medium- and longer-terms. This section also offers several conjectures that relate the apparent persistence to large fixed investments made a century ago. Conclusions and several possible extensions are presented in Section 4. 2. The dynamic context of Los Angeles The area analyzed in this research is that covered by a “compact set” of tracts in the urbanized portions of greater Los Angeles, including portions of the five-county Los Angeles CMSA. The compact

226

C.L. Redfearn / Regional Science and Urban Economics 39 (2009) 224–232

Fig. 1. Tract inventory: in and out of centers — 2000.

set of tracts is a subset of the tracts included in the 2000 urbanized area of the Los Angeles CMSA, removing from the urbanized area those tracts too sparsely populated with jobs to support the statistical tools required to identify employment centers, specifically those developed in Redfearn (2007). For these tracts, employment and population data for 1980, 1990 and 2000, as well as shape files for each year, were provided by the Southern California Association of Governments (SCAG). The employment data are developed by SCAG from wage and compensation data reported to the State Economic Development Department (EDD) of the California Labor and Workforce Development Agency. For each of the three cross-sections, a consistent geography based on the 1990 Census tract definitions is used.2 The final data set includes 2337 tracts covering a total area of about 2.15 million acres (about 3350 square miles). Fig.1 plots the centroids of these tracts with an overlay of the region's system of highways. It also differentiates between those tracts in and out of employment centers.3 A close look at one of these centers — the Glendale/Burbank center — is shown in Fig. 2. Together, these two figures hint at one of the conjectures that will be examined below: the role of fixed investment in explaining the stability of the employment centers over time. The highways that run through Fig. 2 are Interstate 5 (diagonally from northwest to southeast) and state route 134 (the east–west artery); highways 2 and 170 form the eastern and western borders. These highways represent enormous fixed investments that are essentially permanent — if not a result of their cost and durability, then of the existence of rights-of-way that will heavily favor their reuse rather than creating new freeways. The commercial structures in Glendale and Burbank are certainly not

2

See Giuliano and Redfearn (2007) for more details. The definition of an employment center used in this research follows Redfearn (2007): a center is a contiguous set of census tracts that are significantly more dense (with regards to employment) than their surrounding tracts. Centers in this definition are therefore a function of relative density — not absolute. In this way, Riverside and Oxnard — clearly local centers of economic activity — can be identified despite their low absolute level of employment density. For reference, the center of downtown Los Angeles contains six contiguous Census tracts in excess of 100 jobs per acre with a total employment of almost 200,000. In contrast, Ventura County contains only seven tracts with greater than 10 jobs per acre. The total employment in these seven tracts is just over 30,000, but because of their spatial proximity and the lower levels of employment density surrounding them, they form a center. 3

permanent, but they are very long-lived. Moreover, they offer a flexibility of usage that implies that renovation will often dominate demolition and rebuilding as a response to economic change. This is especially true in the post-manufacturing economy, in which office buildings can be adjusted to accommodate a wide-range of productive activities. To the point of longer-term persistence in the location of centers, the Glendale/Burbank center, once established, offers agglomerative economies (and appropriate zoning) that favors additional real estate investment within the center over sites outside it. So, while buildings are routinely renovated and adjusted and new structures added to the skyline, the location of the center does not change. No small change in the fundamental variables used in urban modeling will result in wholesale change in a center's aggregate fixed investment. Land prices and rents may change, but because of high adjustment costs and a long horizon of uncertainty, land use is very slow to change, yielding a stable urban form. Fig. 1 suggests the same interaction on a broader scale. As this illustration suggests, it would not be an inappropriate generalization to state that centers exist as a function of access to the transportation network. This is both obvious and a direct prediction of basic urban models. What is interesting is that most of the current centers formed decades ago, on a freeway system that pre-dated WWII and the development of the modern highway system. Another way of viewing this spatial arrangement of economic concentrations is to look for where there are no centers on current highways (and, in particular, on the intersections of these highways). The region is full of “transit advantaged” locales that reveal no employment centers. In fact, many of these locales are on the newer freeway segments with no antecedent in the older freeway systems, offering no original employment concentration from which a current one might have evolved. These points are discussed more completely below. As discussed previously, a finding of stability in the location of an urban area's employment centers would be less compelling if the urban area itself were relatively static. The empirical results presented in this section are intended to suggest that Los Angeles and its surrounding cities form a highly dynamic metropolitan area — perhaps one of the most dynamic urban areas in the United States over the two decades from 1980 to 2000. It should be noted that dynamism and growth are not synonymous. Certainly the Los Angeles metropolitan area has grown, adding 1.4 million jobs and 3.7 million

C.L. Redfearn / Regional Science and Urban Economics 39 (2009) 224–232

227

Fig. 2. Employment center example: Glendale/Burbank — 2000 (numbers indicate jobs per acre within census tracts).

residents. But, as large as these numbers are, they understate the transformation of region's economy and demographics. Like the rest of the country, the region's economy has experienced a broad trend toward services and away from manufacturing employment. And though it retains more manufacturing than many other major metropolitan areas, the period from 1980 to 2000 represents an upheaval in the composition of manufacturing employment. Heavily concentrated in defense industries, Los Angeles experienced major dislocation within the manufacturing sector at the close of the Cold War and its associated cuts in defense spending. In fact, a significant portion of the changes in the system of centers may be attributable to the decline of employment in these particular industries. The “churning” of the population is no less remarkable than the change in the cross section of employment. While adding 3.7 million residents, the white population lost its status as the majority racial/ ethnic group, with Hispanics reaching a plurality. The Los Angeles region is now home to large communities of immigrants from around the globe. It has been the rapid growth of these groups that has fueled the overall growth in the region's population. Table 1 offers a glimpse into these dynamics. It reports that the 1980s were a period of robust growth in both employment and population for all the counties that make up the five-county CMSA. However, growth during these decades was asymmetric across counties, with the peripheral counties growing much faster than the core. Table 1 Percent change in employment and population by county Area Los Angeles Orange County Riverside San Bernardino Ventura Metro Area

Employment

Population

1980–1990

1990–2000

1980–1990

1990–2000

15.1% 41.6% 35.0% 57.4% 56.9% 23.1%

−4.5% 15.7% 43.2% 28.2% 19.1% 3.3%

16.3% 24.1% 49.1% 54.4% 28.1% 21.5%

7.4% 18.2% 19.6% 20.9% 14.0% 11.2%

The region's employment base grew from 5.2 million to 6.5 million during the 1980s. Growth moderated substantially over the 1990s, yielding a net addition of only 200,000 jobs over the full decade. These broad trends mask the continued rise of the suburban counties of Orange, Riverside, San Bernardino, and Ventura — all of which maintained a brisk rate of growth even after the region's economy slowed. Note that while Los Angeles County actually lost significant employment, it remained the dominant employer within the region throughout the 1990s, with almost twice the employment level of the four peripheral counties combined. To a great extent, population shifted in concert with the trends in employment, underscoring the transformation of the outer counties from bedroom communities of commuters to employment centers in their own right. Unlike the case of employment, Los Angeles County did not lose population after the job cuts in the defense industries after the early 1990s. Because centers are an area of focus and because the working definition of a center rests on the relative density of employment, it may be useful to revisit the employment figures in terms of job density. Table 2 reports the employment densities for the Los Angeles metropolitan area and its five component counties.4 The table makes comparisons between the counties easier by removing variation in their geographic size. Los Angeles County is the most dense, but Orange County shows a marked densification over the twenty-year sample period. The other three counties have also become relatively more dense. Table 2 also reports the growth of employment outside the boundaries of the 1980 centers. That is, employment rose faster outside the centers in the twenty years for which there is data. This does not suggest stability. Instead, it demonstrates consistent migratory trends away from the center. Indeed, each of the outer four counties account for a growing share of the metropolitan area's jobs and residents at the expense of the region's traditional center, Los Angeles County. Moreover, the relative density of tracts in and out of the 1980 centers declines. This 4 Again, keep in mind that for all five counties the sparse rural regions have been removed from the data. Our focus is on intra-urban area employment.

228

C.L. Redfearn / Regional Science and Urban Economics 39 (2009) 224–232

might suggest a weakening of the relationship between employment concentrations in the distant past and today. Making sense of these trends while arguing for general stability in the spatial distribution of employment is the subject of the next section. 3. Documenting persistence Over a short period of time, persistence in the spatial distribution of employment is obvious, given the durability of the fixed investment in buildings and infrastructure. The empirical issue addressed in this section is the duration of the persistence beyond the short-term. This is addressed in two ways. First, a series of simple regressions are estimated that make use of consistent-geography employment and spatial data. These data permit a medium-term of twenty years to be examined, specifically the two decades bounded by three crosssections of Census data: 1980, 1990, and 2000. The longer-term is explored less formally, linking the organization of current employment around significant locations from various eras, stretching back a hundred years to the turn of the 20th century. In both frameworks, the central empirical approach is to examine the significance of historical data as it explains the current spatial distribution of employment. To the extent that urban models are ahistorical — with the spatial distribution of employment a function only of contemporaneous or recent levels of transportation and communication costs, wages, production functions, etc., data from previous epochs should have little to say about where employment is located today — especially given the pronounced growth and change in the cross sections of employment and population in the Los Angeles metropolitan area. Note that the variable used in this analysis is the location of employment, not rents. Given slow adjustment in the capital stock, it is likely that rents are more responsive to shocks. But the actual location of economic activity should change in response to changes in rents over the medium term. Of course, there are many mechanisms by which persistence may be attenuated or enhanced: zoning, the decentralization of control over economic development, local geography, etc. The paper does not test for their individual relevance. Rather, the goal is simply to examine whether or not the past is central to the present in a manner significant enough to warrant consideration. Future research will be needed to speak to these specific points. 3.1. Medium-term persistence To examine the stability of the location of employment over time, a series of simple regressions can be estimated that simply ask how past employment in a tract predicts subsequent employment. Clearly from day to day these regressions would be uninteresting. However, the dates of the Census cross-sections fit well with a pronounced business cycle in the metropolitan area. The 1980s represented a boom in the region, especially for defense manufacturing. With a rapid build up of armed forces toward the end of the Cold War, the region's traditionally large share of employment dedicated to defense industries (dating back to WWII) experienced significant growth. These were high-multiple jobs, with many other jobs created to support and serve them. The 1990s, by contrast, saw the end of the Cold War and a shock significant enough to reduce the labor force by almost a million jobs, dropping from 12.6 M at Table 2 Employment density and center shares by county Area Los Angeles Orange County Riverside San Bernardino Ventura Metro Area

Density (jobs/acre)

Density ratio in/out

1980

1990

2000

1980

1990

2000

3.7 2.2 0.7 0.8 0.7 2.5

4.3 3.1 0.9 1.2 1.0 3.0

4.1 3.5 1.3 1.6 1.2 3.1

6.5 7.3 11.1 9.3 4.3 7.6

5.4 5.4 6.2 5.5 4.1 6.0

5.5 5.1 5.5 3.3 3.3 5.7

Table 3 Persistence in the density of employment Variable

Model 1

Model 2

Model 3

Model 4

Model 5

Model 6

Model 7

df r2 Intercept

2334 0.156 2.381 (29.45) −0.456 (16.22) −0.233 (9.65)

2335 0.793 0.145 (9.32)

2335 0.649 0.625 (37.41)

2334 0.805 0.202 (12.79)

2333 0.652 0.528 (8.66) 0.031 (1.56) − 0.073 (4.62)

0.636 (65.72)

0.688 (43.29) 0.160 (12.21)

2333 0.793 0.085 (1.75) 0.020 (1.35) −0.025 (2.01) 0.850 (84.86)

2332 0.807 0.018 (0.39) 0.062 (4.16) −0.021 (1.75) 0.688 (43.20) 0.172 (12.79)

ln(dcbd) ln(dhwy00) ln(ed90) 80

ln(ed )

0.850 (94.56)

0.632 (57.74)

the end of 1990 to 11.8 M just three years later. Much of the decade was spent restructuring and recovering from the shock to defense spending and its repercussions around the Los Angeles basin. Regressions across these two decidedly different periods should capture spatial reallocation of employment to the extent that there is any. Table 3 reports results for seven regressions that build off of what might be considered a basic urban model in which the log of tract-level employment density in 2000 is a linear function of the log distance to the CBD (dcbd) and the log distance to the nearest current highway (dhwy00). This base model, Model 1, is consistent with the basic urban model: coefficients on both distances are negative and highly significant. Models 2, 3, and 4 report comparisons to other highly simplified models of employment density, with tract-level employment density in 2000 a function of lagged employment density in 1990 and/or 1980 (ed90 and ed80). In all three models, the explanatory power is far greater than the base model, with 65 to 80% of the spatial variation explained by past employment density alone. When the two types of variables are included in the same regressions, three patterns emerge. First, the center loses its traditional role: the coefficient is now positive, though significant in only one of the three mixed models. Second, the importance of access to highways remains significant but is diminished by an order of magnitude. Lastly, the inclusion of the distance variables adds no explanatory power to the lagged employment density regressions.5 The polycentric nature of Los Angeles is well-noted, so it may be useful to examine the same type of variables in explaining the membership of Census tracts in employment centers. Table 4 reports logistic regressions of whether a tract is a member of an employment center or not on the same distance variables and on lagged center membership indicators, Ictr90 and Ictr80. Here the relationships differ somewhat when the dichotomous outcome is used as a dependent variable rather than the continuous variable used in the previous regressions. For example, distance to the CBD is not a significant predictor of tract membership in an employment center. Distance to the nearest highway is, however, and it is more robust than in the previous regressions. The key observation, however, remains: membership persists over this medium term. Past membership in an employment center is far more relevant than the measures of current access.6

5 In fact, Table 3 understates the extent of persistence. Where asymmetric growth in the spatial distribution of employment occurs, it reduces the explanatory power of the models of density level on earlier density levels, even where differential growth may not change the rank of employment density of one tract relative to others. For example, the central business district of Los Angeles has lagged in terms of growth over the entire twenty-year sample period, but remains by far the largest and most dense center of employment in the region. 6 With regard to the persistence of centers, there are several interesting statistics. There are 538 tracts that are members of 41 centers in 2000. This number is 555 in 1990 in 44 centers and 535 in 1980 in 41 centers. The majority of the change in tract membership comes from boundary changes, although there are several new centers in each decade as well as several that disappear. On the issue of stability of centers, 383 of the 538 tracts that were in centers in 2000 were in centers in 1990. This number is 351 when the comparison is 1980. The tracts common to centers in each of three crosssections comprise 85% of center employment in 2000.

C.L. Redfearn / Regional Science and Urban Economics 39 (2009) 224–232 Table 4 Persistence in employment center membership Variable

Model 1

Model 2

Model 3

Model 4

Model 5

Model 6

Model 7

df r2 Intercept

2334 0.021 −0.854 (8.20) 0.020 (0.55) −0.223 (7.22)

2335 0.310 − 0.432 (12.37)

2335 0.308 −0.369 (10.45)

2334 0.361 − 0.331 (8.89)

2333 0.312 −0.517 (4.35) 0.041 (1.00) −0.121 (3.37)

0.939 (26.64)

0.684 (16.57) 0.483 (11.49)

2333 0.319 −0.638 (5.32) 0.056 (1.35) −0.175 (4.8) 0.919 (26.12)

2332 0.370 −0.615 (4.89) 0.084 (1.94) −0.172 (4.58) 0.677 (16.33) 0.483 (11.44)

ln(dcbd) ln(dhwy00) Ictr90 I

0.928 (26.58)

ctr80

0.932 (26.28)

Table 5 Comparing the explanatory power of different highway systems Variable

Model 1

Model 2

Model 3

Model 4

Model 5

Model 6

df r2 Intercept

2334 0.156 2.381 (29.45) −0.456 (16.22) −0.233 (9.65)

2334 0.214 2.144 (26.83) −0.418 (15.39)

2333 0.221 2.097 (26.13) −0.403 (14.76) −0.112 (4.52) −0.361 (13.99)

2332 0.794 0.080 (1.65) 0.021 (1.39) −0.017 (1.32) −0.027 (1.97) 0.844 (80.47)

2332 0.655 0.511 (8.38) 0.031 (1.58) −0.051 (3.1) −0.075 (4.17)

2331 0.807 0.017 (0.36) 0.062 (4.16) −0.018 (1.43) −0.010 (0.77) 0.687 (42.84) 0.171 (12.65)

ln(dcbd) ln(dhwy00) ln(dhwy42) ln(ed90) ln(ed80)

−0.402 (16.55)

0.618 (54.16)

Are highways and the location of employment jointly determined? If so, the coefficients in these sets of regressions could be biased. During the sample period there was no significant change in the highway system — with the current regional highway system largely established by 1960. The difference in the highway systems as of 2000 and as of 1960 is essentially the addition of segments that run through areas of sparse employment. Along these routes there appear to be, at most, one center that might be attributable to the “new” route — all other centers explained by the 2000 network are also explained by the 1960 era network. Baum-Snow (2007) argues that the placement of the early national highway system was established in 1947, designed for national security and not future economic development. The stability of the highway system raises the issue, however, that the persistence in the spatial distribution of employment may have something to do with the persistence in the location of the transportation network. One further set of regressions is estimated to explore this notion. Table 5 reports regressions designed to make the comparison between the current highway system and that which existed earlier — the set of state highways that existed as of 1942. The table makes clear that this earlier system is more relevant to the current organization of employment in the region than the current system. The coefficient on distance to the older highway system (dhwy42) is both larger in magnitude and more precisely estimated than distance to the current system (dhwy00).7 The greater power of the earlier highway system to explain current employment location offers some evidence of a longer-term persistence than the twenty-year data reveal. While much of the current highway system follows rights-of-way established in the earlier system, the systems differ with regard to economic activity in two ways. 7 These results are echoed in logistic regressions of center membership on the same proximity measures. Again, the 1942 highways dominate the 2000 highways in terms of explanatory power.

229

First, the sections of the current highway system that do not follow earlier highways have more to do with facilitating inter-regional traffic rather than providing new ways to move among existing centers. Second, significant portions of the earlier network have not been recycled into more intense uses — staying boulevards rather than becoming freeways — but retain significant concentrations of employment. Indeed, the second densest center in the region is found along old Route 66 and not astride a current highway. The large fixed investment in buildings and the large associated agglomerative benefits of the employment concentration on Santa Monica Boulevard in west Los Angeles persist despite losing the feature that played a significant role in its development.8 Because the earlier network was developed in order to link places of economic activity, it could be a better explanatory variable for current employment location if the current location of employment reflected a high degree of persistence. Without longer-term persistence in the location of employment, there would be no explanatory power in the older transportation system. Taken together, the tract-level and center-level statistics suggest a marked persistence in the spatial distribution of employment lasting at least twenty years. Without appealing to any contemporaneous variables, the explanatory power of historic (lagged-employment) models is approximately three-quarters of total variation. The last set of regressions suggests a process by which the persistence could arise. Consider the following iterative process: First, distinct economic places (i.e. not contiguous) become connected via simple transit technology, enhancing the productivity of all centers which become connected. This further favors these locales over undeveloped sites as optimal choices for incoming or expanding firms. As growth occurs, residents — who locate on the periphery of the distinct economic places — fill in the areas between them and effectively preclude truly new transit networks and force employment densification. This results, not in ongoing re-optimization in an unconditional sense, but rather in adjustments to employment within existing centers and to transit networks along established rights-of-way. The result is an ossification of employment center locations and the pathways the transit networks follow (even if modes along them or their intensity of usage may change over time). Following this logic, new highways should — as they do in Table 5 — have little predictive power relative to older ones. Furthermore, the first economic concentrations tied together by the early transit networks should retain some influence on the current location of employment, which they do. 3.2. Longer-term persistence The goal of this section is to explore longer-term persistence by examining the current spatial organization of employment around “significant places” — both current and historic. Having defined sets of “significant places,” it will be possible to see which set of points best explains the current distribution of employment. This is done by calculating the cumulative share of employment within concentric rings around the various sets of points. Again, to the extent that current and/or recent conditions explain location choice, significant places from the distant past should not be useful in explaining the current urban form. To disentangle the role of transit networks from original distribution of economic activity, two types of points were considered. The first are highway intersections, because it is possible that each point on the network is not as relevant as others. For example, in a national context, one might estimate regressions analogous to those in Table 3 and find

8 There a handful of significant exceptions to the rule that centers are located on highways. They are not glaring in the sense that none of them are truly far from the current highway network, but they do not appear to be organized with the current system in mind. These include the Wilshire/Santa Monica Boulevards center as well as the Hollywood center adjacent to it. In the San Fernando Valley, there is a center that runs diagonally and is unattached to any freeway; the same is true of a center directly south of the CBD; these two centers are manufacturing centers that abut a rail line.

230

C.L. Redfearn / Regional Science and Urban Economics 39 (2009) 224–232

Fig. 3. Cumulative distribution of employment around ‘significant places’.

the interstate system of freeways as equally irrelevant due to the vast expanses of low-density farm land that lies between Seattle and Chicago. So, the more relevant points on the transit network may be the intersections which favor firms over places not on intersections, e.g. Chicago and Seattle, but not the length of Route 94 that connects them. Therefore two sets of highway intersections are used as origins around which the share of total current (2000) employment is calculated; these are intersections of the highway systems as of 2000 and 1942. In addition to the highway intersections, a different type of location is also used as the basis for analogous calculations. Here, places that were either incorporated as cities or contained proven significant economic activity as of a particular year are included. The goal is to find economically significant and distinct places that existed earlier than any highway system. If the crude process outlined above holds, it is these places that should be tied together in the early years by rudimentary transportation networks which will in turn establish the rights-of-way that are still used today as transit corridors. These “significant places” are defined as such if one of two criteria were met. The first is incorporation. If a place was legally incorporated as a city by the cutoff year, the centroid of the current city was used as a significant place. Incorporation meant several things in the early 1900s and is therefore an imperfect measure of economic distinctiveness. That is, incorporation was often the first step taken by business leaders engaged in economic development — i.e., seeking fixed investment. That said, incorporation was also used defensively by residents to ward off annexation by neighboring cities. This defensive incorporation appeared to be just getting underway between 1905 and 1910 and so both years are included in the analysis, though it appears that the two purposes of incorporation will result in some noise in this measure. The second manner of being included as a significant place is if, in fact, there is evidence of a material concentration of economic activity. This definition was necessitated by the large number of cities which — while economically relevant for decades by the early 1900s — did not incorporate until much later. Thousand Oaks, for example, was long the site of the stage coach stop on the route connecting Los Angeles and Santa Barbara. It had a major inn and supporting services well before many of the incorporations in the late 1880s, but did not itself incorporate until 1956. Other places, such as Hollywood, incorporated relatively early — in 1903 —

but then acceded to annexation to Los Angeles within a decade, leaving it then (as it now is) as a district of the city of Los Angeles. As a robustness check to these designations as “significant places,” two additional historic maps were used to provide alternative sets of significant places. The first is an Auto Club map from 1915. On it, there are almost 100 names places — with no indication of formal incorporation, just an approximation of their improved areas. These places were then scanned to obtain the latitude and longitude of the named places. A second map from 1918–21 displaying oil fields is also used to provide the coordinates of named places. Even at these early dates, it is clear that these sets of historic locales are related to the current spatial distribution of employment. A large number of the place names exist as submarkets in current commercial real estate markets. Fig. 3 displays the cumulative shares of employment within various concentric rings around the two sets of significant places. Note the line marked with the upright triangles — this is the cumulative share of employment organized around highway intersections in 2000. These are crude measures of access, but capture a set of advantaged locales that offer easy access to two or more highways. Half of all employment in 2000 in the sample area is located within 2.4 miles of these points. 90% of employment is located within 5.8 miles; 95% within 7.2 miles. By comparison, the set of 1942 highway intersections captures 50, 90, and 95% of total employment within 2.0, 5.4, and 8.4 miles. The older highway system explains more of the core employment, while the peripheral employment is better captured by the newer system. Across the set of historic locales, there is a natural progression of more employment being organized around a growing number of points as incorporation continues. The earliest set of points “explains” current employment least-well, but is a decidely more informative set of points than is a randomly drawn set of points. That this set does not do as well as others sets may be mechanical, as there are only 33 points as of 1895, but 80 in the 1942 highway intersection set. Not surprisingly, it does not explain the current distribution of employment as well as the other sets. That said, these are a set of points that represent economically distinct points at the end of the 19th century. That they have any explanatory power over current employment is somewhat remarkable. The other sets of historic points — those from 1905 and 1910 — produce virtually identical cumulative distribution

C.L. Redfearn / Regional Science and Urban Economics 39 (2009) 224–232

functions as the set of highway intersections in 2000. While crude, these measures suggest that the spatial distribution of employment as of 2000 is explained as well by the distant past as current conditions.9 The points provided by the two maps display the “best” explanation of current employment. These two sources are likely more accurate with regard to the economic significance of the early concentrations, given the noise in the date of incorporation as a proxy for economic significance, as well as the noise in the assigned date of proven economic activity. Current employment is clearly organized around these two sets of points to a much greater extent around the current highway intersections. The 1915 Auto Club map provides a set of points around which 50% of total employment is within 2.0 miles, 90% within 4.0 miles, and fully 95% within 4.6 miles — a third less than the 2000 highway intersections. The 1918 oil field map fits only slightly worse despite having only two-thirds the number of named places. There appears to be a marked persistence in the spatial distribution of employment. Be it the established set of locations by 1910 or the named places on the maps from 1915 or 1918, there is a consistent pattern of organization of current employment around the locations that were economically significant close to a century earlier. Contemporaneous access to transit is a frequent variable in urban models. This is because these models predict that travel time is a relevant variable in both firms' and households' location choice. Nothing reported in this paper disputes that hypothesis. Rather, the evidence collectively suggests that the minimization problems solved by households and firms in the distant past resulted in investment choices that appear to condition choices made many years later — in a way that is not frequently acknowledged in urban models. Here, an early set of locations remains a significant explanatory factor in the location of employment today — almost 100 years prior to the data being examined.10

231

periods, or where there is little change in employment or population, this is easily comprehended. That this appears to hold in such a dynamic region, and over such a long period of time, strongly suggests that urban form depends less on some of the common variables used to explain the shape of cities than on their investment histories — even when those investment decisions were made under dramatically different social and economic conditions. For example, over the last twenty years — the era in which the Internet has “destroyed space” — the spatial arrangement of employment has remained remarkably similar to that when e-mail was the domain of a nerdy few in the basements of computer science buildings. Moreover, the location of employment centers appears to have persisted though a long period of pronounced growth in income and ever-shrinking transportation costs (Glaeser and Kohlhase, 2003). The location of the region's employment is better explained by the initial conditions a century ago rather than changes in the set of variables usually thought of as urban fundamentals. The persistence exhibited in the Los Angeles region appears to be driven primarily by investment in place well before the beginning of the 1980–2000 sample period. The large majority of the employment centers in the region are located on portions of the highway system that existed as of 1960. Those centers that appear inconsistent with this general finding are likely to be older centers which formed around even older infrastructure — either the pre-WWII system of freeways or rail lines. In fact, using a simple accounting, the locations of concentrated economic activity in the region by 1905 explain the current spatial distribution of employment as well the current freeway system does. These are crude tests of employment location, but they strongly suggest that the duration of persistence in the spatial distribution of employment is long enough to warrant more serious consideration in urban models. That is, initial conditions, adjustment costs, and short-term dynamics may be more interesting than long-run equilibria.

4. Conclusions and extension References Though it is common to model cities as moving in response to marginal changes in fundamental variables from one unconstrained maximum to another, in reality the problems facing agents are very much constrained — taking the investment choices made by generations of previous households, firms, developers and governments as given. This fact leads to a persistence within urban areas in which starting points determine the current location of economic activity. Even where the basic building blocks of urban models — transportation costs, communication costs, production technology, etc. — have changed dramatically, the spatial distribution of employment within a metropolitan area may persist for very long periods of time. Against a backdrop of expansion and change, the location of employment centers is generally stable. This is not to suggest stasis. Indeed, there is substantial growth outside the traditional urban core, although it may be better characterized as reconcentration rather than sprawl — employment in the peripheral areas has become more urbanized. For example, employment has grown faster in Orange County than in Los Angeles County. However, rather than spawning countless low-density centers, this asymmetric growth has resulted in employment densities within existing suburban centers that now rival the densities of the employment centers in Los Angeles County (Giuliano and Redfearn, 2007). This process of densification within a stable set of employment centers raises the issue of persistence: economic concentration exists today where economic concentration existed yesterday. Over short

9 Given the high degree of medium-term persistence, it is not surprising that these sets of historic points explain the 1990 and 1980 cross sections of employment as well. 10 It is possible, in fact, to move even further back to the original ranchos which divided the land in the regions. It is along the boundaries of these parcels that the great boulevards were established. In fact, it's possible to see the rough outlines of spatial organization of land use prior to statehood through careful selection of current major surface streets. This is another way to see the long persistence of infrastructure investments. This is left for future research.

Anas, A.,1978. Dynamics of urban residential growth. Journal of Urban Economics 5, 66–87. Anas, A., Arnott, R., Small, K.A., 1998. Urban spatial structure. Journal of Economic Literature 36 (3), 1426–1464. Arnott, R., 1980. A simple urban growth model with durable housing. Regional Science and Urban Economics 10 (1), 53–76. Arthur, W.B., 1988. Urban systems and historical path dependence. In: Ausubel, J.H., Herman, R. (Eds.), Cities and Their Vital Systems: Infrastructure Past, Present, and Future. National Academy Press, Washington, D.C., pp. 85–97. Anderson, N.B., Bogart, W.T., 2001. The structure of sprawl — identifying and characterizing employment centers in polycentric metropolitan areas. American Journal of Economics and Sociology 60 (1), 147–169. Baum-Snow, N., 2007. Did highways cause suburbanization? Quarterly Journal of Economics 122 (2), 775–805. Bogart, W.T., Ferry, W.C., 1999. Employment centres in greater Cleveland: evidence of evolution in a formerly monocentric city. Urban Studies 36 (12), 2099–2110. Braid, R.M., 1995. The use of land for roadways in a growing Mills-de Ferranti urban area. Journal of Urban Economics 37 (2), 131–160. Brueckner, J.K., 1980. A vintage model of urban growth. Journal of Urban Economics 8 (3), 389–402. Brueckner, J.K., 2000. Urban growth models with durable housing. In: Huriot, J.M., Thisse, J.F. (Eds.), Economics of Cities: Theoretical Perspectives. Cambridge University Press, Cambridge, pp. 263–289. Cervero, R., Wu, K.L., 1997. Polycentrism, commuting, and residential location in the San Francisco Bay Area. Environment and Planning A 29 (5), 865–886. Craig, S.G., Ng, P.T., 2001. Using quantile smoothing splines to identify employment subcenters in a multicentric urban area. Journal of Urban Economics 49 (1), 100–120. Cronon, W., 1991. Nature's Metropolis: Chicago and the Great West. W.W. Norton, New York. Dye, R.F., McMillen, D.P., 2007. Teardowns and land values in the Chicago Metropolitan Area. Journal of Urban Economics 61, 45–63. Garreau, J., 1991. Edge City: Life on the New Frontier. Doubleday, New York. Giuliano, G., Redfearn, C.L., 2007. Employment concentrations in Los Angeles, 1980– 2000. Environment and Planning A 39 (12), 2935–2957. Giuliano, G., Small, K.A., 1991. Subcenters in the Los Angeles Region. Regional Science and Urban Economics 21 (2), 163–182. Glaeser, E.L., Gyourko, J., 2005. Urban decline and durable housing. Journal of Political Economy 113 (2), 345–375. Glaeser, E.L., Kohlhase, J.E., 2003. Cities, regions, and the decline of transport costs. Papers in Regional Science 83 (1), 197–228. Gordon, P., Richardson, H.W., 1996. Beyond polycentricity: the dispersed Metropolis, Los Angeles 1970–1990. Journal of the American Planning Association 62 (3), 289–295. Harrison, D., Kain, J.F., 1974. Cumulative urban growth and urban density functions. Journal of Urban Economics 1 (1), 61–99.

232

C.L. Redfearn / Regional Science and Urban Economics 39 (2009) 224–232

McDonald, J.F., Prather, P.J., 1994. Suburban employment centres: the case of Chicago. Urban Studies 31 (2), 201–218. McMillen, D.P., 2001. Nonparametric employment subcenter identification. Journal of Urban Economics 50 (3), 448–473. McMillen, D.P., McDonald, J.F., 1998. Suburban subcenters and employment density in Metropolitan Chicago. Journal of Urban Economics 43 (2), 157–180. McMillen, D.P., Smith, S.C., 2003. The number of subcenters in large urban areas. Journal of Urban Economics 53 (3), 321–338. Munneke, H.J., 1996. Redevelopment decisions for commercial and industrial properties. Journal of Urban Economics 39 (2), 229–253.

Redfearn, C.L., 2007. The topography of metropolitan employment: identifying centers of employment in a polycentric urban area. Journal of Urban Economics 61 (3), 519–541. Rosenthal, S.S., Helsley, R.W., 1994. Redevelopment and the urban land price gradient. Journal of Urban Economics 35 (2), 182–200. Wheaton, W.C., 1982. Urban spatial development with durable but replaceable capital. Journal of Urban Economics 12 (1), 53–67.