McGill University Working Paper

Does Spatial Variation in Heterogeneity Matter? Assessing the Adoption Patterns of Business Improvement Districts

Leah Brooks Department of Economics McGill University

April 2006

I am grateful to Tara Syed for research assistance, and to the advice and support of Janet Currie, Jean-Laurent Rosenthal, Naomi Lamoreaux, Paul Ong and Sandy Black at UCLA, to new colleagues Maxim Sinitsyn, Mary MacKinnon, Jenny Hunt, Dee Sutthiphisal, Daniel Parent, the Public Economics group at the CeMent workshop and participants at the February 2006 DeVoe Moore Center Critical Issues Symposium. I am also deeply indebted to the many municipal officials who helped me assemble the dataset and without whose cooperation this project would have been impossible. This work was supported in part by a NBER Non-Profit Dissertation Fellowship, and by a grant from the Lincoln Institute for Land Policy. c

2006 by Leah Brooks.

1

Does Spatial Variation in Heterogeneity Matter? Assessing the Adoption Patterns of Business Improvement Districts

Because they supplement the municipal provision of local public goods, Business Improvement Districts (BIDs) provide an opportunity to examine the space, scope, and determinants of the provision of local public goods. A BID is formed when a group of merchants or commercial property owners in a neighborhood vote in favor of package of self-assessments and local public goods to be funded with those assessments. These districts solve a collective action problem in the provision of public goods because once a majority have voted in favor, participation is compulsory for all merchants or commercial property owners in the neighborhood. I use a unique dataset on adoption patterns of BIDs in California to test two main claims suggested by the theoretical literature: first, that businesses respond to individual heterogeneity that determines the quality of local public goods, and second, that the type of heterogeneity – overall or spatial – matters. In contrast to the literature on residents, this study finds at best a weak correlation between a city’s adoption of a BID and heterogeneity. In addition, despite the theoretical preference for spatial over overall heterogeneity, BIDs are not more likely to be adopted by spatially heterogeneous cities.

Leah Brooks Department of Economics McGill University 855 Sherbrooke St. West Leacock Hall, Room 439 Montreal, QC h3a 2t7 CANADA [email protected] http://people.mcgill.ca/leah.brooks/

2

Because public goods are critical determinants of economic growth, determining how their level is set is critical to understanding overall economic well-being. In particular, research by this author (Brooks, 2005) and others (Calanog, 2004; Hoyt, 2005) has shown that a specific institution for providing local public goods, the Business Improvement District (BID), is effective in reducing crime. Because security and the other local public goods that BIDs provide are so important to economic well-being, it is of interest to know why BIDs are adopted some place and not others. Specifically, this paper asks whether a city’s adoption of a BID can be explained by the prevailing hypothesis about public goods provision in the literature, which argues that supplementary provision should be driven by heterogeneity of demand. In California, a Business Improvement District (BID) is formed when a group of business or commercial property owners in a neighborhood vote in favor of a package of taxes and expenditures to provide local public goods. Once a majority of owners vote in favor of establishing the district, BID taxes are binding upon all members, thus resolving the problem of collective action such a neighborhood would otherwise face in providing local public goods. BID revenues are generally spent on cleaning, marketing and security. Though total BID expenditures are not large, they are locally sizeable; for example, though total BID expenditures are less than one percent of the total city budget in Los Angeles, in some neighborhoods BID expenditures are more than double city expenditures. The preeminent strain of research on the level of local public goods argues that heterogeneity in demand for public goods lowers the level of their provision. Building on theoretical work by Alesina and Spolaore (1997)1 , Alesina et al. (1999) show that more racially heterogeneous jurisdictions spend less, as a percentage of their budget, on productive public goods2 1

Epple and Romano (1996) discuss the related issue of the impact of different schemes of provision for publicly provided private goods. Though the issue is not highlighted in their paper, a close reading shows that the provision scheme matters only in heterogeneous cities. 2 Which the authors define in contrast to redistributive public goods, such as welfare.

3

such as education, roads and sewers. Using a similar division of public goods, Chaudhary (2005) shows that caste-heterogeneity is associated with lower levels of education spending across rural districts in colonial India. Additionally, Vigdor (2004) shows that people in racially, generationally, and socioeconomically heterogeneous communities are less likely to “undertake actions generating public benefits,” as measured by response to census forms.3 By studying the adoption patterns of Business Improvement Districts (BIDs), I can address three gaps in this literature. First, the literature has in general focused on the impact of preference heterogeneity on the quantity of local public goods provided – usually measured by spending – rather than on quality.4 Though expenditures on public goods should be correlated with quality of those public goods, the strength of the correlation should vary both by place and type of expenditure. Here I use an outcome measure of revealed preference – the adoption of an institution to supplement local public goods provision – which I relate to heterogeneity. By choosing to form a BID that adds to city services, a BID neighborhood signals that the quality of municipally-provided services is insufficient for their needs given their budget. Second, the literature to date has used the index of fragmentation (or Hirfendahl index) almost exclusively to measure heterogeneity, as it captures the overall disparity of groups (e.g., racial groups) across categories. However, where the provision of public goods has an important local – local in space – component, a measure of heterogeneity that includes this spatial distribution should explain the level of public goods better than overall heterogeneity. Though theory argues that the dissimilarity index should explain supplemental local public goods provision better than the index of fragmentation, and that the interaction of the two 3

In addition, research on education spending by Poterba (1997) shows that greater heterogeneity in the age distribution decreases education spending, and Temple (1996) finds that more heterogeneous small cities in Illinois are more likely to keep state-imposed tax and expenditure limits. In a related vein Alesina et al. (2004) show that jurisdiction size decreases in the heterogeneity of population. 4 The notable exception here is Vigdor (2004), which uses demand for public goods measured by census form returns as the outcome.

4

indices should provide the most explanatory power of BID adoption, I do not find this to be the case. Finally, the literature to date has focused on the response of municipal citizens to heterogeneity. In addition to citizens, cities also have firms, and firms are important actors in determining city policy. Here I ask whether firms are affected by residential heterogeneity. A commercial property owner wants to maximize the stream of rent from his property, so his interest is in finding tenants to pay maximum rent. These business tenants maximize profits which come from revenue of nearby residents. Thus, the aims of commercial property are the same as those of local residents. However, if each neighborhood is as heterogeneous as the city as a whole, then it is unlikely that any neighborhood will wish to supplement the municipally provided level of public goods. However, if neighborhoods are more homogeneous than the city as a whole – that is, if heterogeneity varies across space – neighborhoods with demand exceeding the municipally provided level of public goods should indeed desire to supplement.5 This paper tests two main claims: first, that businesses respond to individual heterogeneity that determines the quality of local public goods, and second, that the type of heterogeneity – overall or spatial – matters. To this end, I combine neighborhood and city level data from the decennial census, economic data from the economic census, government information from the census of governments, and data on crime from the Federal Bureau of Investigation. In addition, because there is no centralized repository of BID information, I conducted a survey to determine cities’ BID status. I contacted city officials in the 114 cities in California that of more than 25,000 people in 1980, and all cities (123 additional) in four 5

This line of argument also suggests the hypothesis that BID adoption at the neighborhood level should be explained by a neighborhood’s heterogeneity relative to the city as a whole. Unfortunately I am unable to address this interesting question in this paper because I have been unable to collect that on other important elements of demand, such as measures of commercial extent and industry type at a sufficiently disaggregated level.

5

Southern California counties6 . This unique dataset describes the adoption of one important class of special assessment districts providing local public goods. Despite the theoretical backing, residential heterogeneity is at best weakly associated with BID adoption in either the Southern California or larger cities sample. Furthermore, spatial heterogeneity does not explain BID adoption better than overall heterogeneity. The conclusion offers three reasons for this disagreement with the literature: that all cities face funding constraints, that heterogeneity is poorly measured, and that the key unit of measurement may be the neighborhood.

1

Theoretical Framework

Suppose a law is passed that allows commercial neighborhoods to overcome the collective action problem in the provision of local public goods. For example, commercial tenants might jointly like their neighborhood to be safer, but the individual cost of affecting the neighborhood’s crime rate is prohibitively high, and tenants do not trust one another to commit to providing security. Does the Alesina et al. (1999) model have any predictions as to which types of cities might have neighborhoods adopting such districts? With a few slight modifications, the answer is yes. First, associate citizens in Alesina’s world not just with a city, but also with a neighborhood. In addition, I add two not terribly restrictive assumptions. First, assume that the firms of interest are retail businesses. This conforms very closely with the actual composition of BIDs. Second, assume that firms serve local residents and local public goods such as crime, parking and cleanliness are an important part of the retail shopping experience. Specifically, assume that firms’ profits are decreasing in the difference between the municipally provided level of the public good and the locally desired level of the public good.7 Formally, this is 6 7

PJ j

min(0, g ∗ − gj∗ ), where g ∗ is the public

The counties of Los Angeles, Riverside, Orange and San Diego. In some cities, where small business owners have historically lived over their shops, there could be some

6

good provided by the city, and gj∗ is the level of the public good desired by person j of neighborhood J. Retailers’ profit increases as the difference between the municipally chosen level of public good, g ∗ , and each neighborhood resident’s desired level of public good, gj∗ shrinks.8 From the Alesina et al. (1999) model, we know that the amount of the public good provided by the city, g ∗ , declines in consumer heterogeneity.9 Because firms’ profits are increasing in g ∗ , firms’ behavior is also impacted by the heterogeneity in taste that lowers the provision of public goods. So what does this tell us about BID adoption? First, consumer tastes matter for firm decisions. Second, if consumers in a given neighborhood are equally as heterogeneous as consumers in the city as a whole, local businesses should be unlikely to supplement the level of public goods. However, if neighborhood consumers are more homogeneous than the city as a whole, business neighborhoods should supplement where consumers feel that the level of public goods provided is too low. In sum, entirely homogeneous cities should not have BIDs.10 Furthermore, heterogeneous cities that are equally heterogeneous in all neighborhoods are unlikely to have BIDs. BID adoption is most probable in heterogeneous cities, where that heterogeneity varies over space. In the discussion of the empirical test, I operationalize these two types of heterogeneity. As we usually do not measure preferences for public goods directly, I measure this preference in terms of correlates for this demand: income, education, race, and age. I expect heterogeneconfusion between and firm and resident. As an empirical matter, this is rarely the case in Los Angeles, so this model accords well with the data. 8 The minimum operator restricts “too much” of the public good from being harmful. 9 Formally, Alesina et al. (1999) call this heterogeneity the “median distance from the type most preferred by the median voter.” 10 Theory also suggests, as in Helsley and Strange (1999) and Epple and Romano (1996), that cities should respond to the existence of a BID by lowering the quantity of public services offered. This is an interesting theory which this paper does not aim to test. In the event that this theory is true, however, the overall theory of heterogeneity causing supplemental provision still holds; we would simply expect more supplemental provision than if the level of public goods provided by cities stayed constant.

7

ity of socio-economic factors to be more salient predictors of profit-driven businesses than heterogeneity of race. Such a result would accord with the finding of Alesina et al. (2004) showing that racial heterogeneity is a determinant of school district size, but not of special district (e.g. water districts, garbage districts) size.

2

Business Improvement Districts

Because BID legislation varies by state, this study focuses on one large state in order to look at adoption patterns within an otherwise constant institutional environment. California, the focus of this study, has had some variant of a BID law on the books since the 1950s, when the state allowed merchant districts to compel taxation to provide parking lots. Over the years, the state has broadened the mandate of these BIDs: in 1989 the state expanded the mandate of permitted services from mostly parking to include marketing and various neighborhood improvements; a major law change in 1995 allowed for the taxation11 of property owners to provide security and other larger structural improvements. Taxation of residents or residential properties is explicitly forbidden in all versions of the law. In order to establish a BID, property or business owners in a neighborhood decide upon a boundary, assessment schedule and budget for the district. They then attempt to convince their neighbors that they, too, should support the BID. The city administers formal voting, and votes are weighted by assessment. Properties or businesses in BIDs may be assessed in any way commensurate with the benefits that property receives. Usually the assessment of properties is some combination of building square footage, lot square footage, and front footage; for businesses it is frequently a percentage of the city’s business license tax. If a majority of assessment-weighted votes are cast in favor of the BID – which is the entire bundle of boundaries, assessments and expenditures for the 1 to 5 year life of the BID – it is 11

This new law escapes the stringencies of California’s Proposition 13 by calling this tax an assessment.

8

established and taxes are mandatory for all merchant owners within the district. The BID then functions as a not-for-profit corporation. BIDs are quite small. For the city of Los Angeles, I have complete information on each individual BID and present information on them as illustration. City of Los Angeles BIDs are usually much smaller than a square kilometer and account for less than two percent of the city’s land area; in 2002, they spent, in total, roughly 19 million dollars. A third of that expenditure went to security, and the remaining funds went to a mix of marketing, cleaning, special projects and administration. Compared to the hundreds of millions in federal monies spent on the Section 8 housing program or Community Development Block Grants, these numbers may seem small. However, when compared to city spending, BID expenditures are large local investments. The Hollywood Entertainment District BID covers roughly three-quarters of a square kilometer and its $1.4 million per square kilometer of security spending slightly exceeds LAPD expenditures of $1.3 million per square kilometer in the same area (Los Angeles Police Department, Information Technology Division, 2003; City of Los Angeles, 2003). The Chinatown BID, at 0.3 kilometers square, in addition to spending on security patrols, spends $280,000 annually on cleaning and maintenance. In comparison, the city of Los Angeles spends $55,000 per square kilometer (City of Los Angeles, 2003).12 Thus, though BID expenditures may be small in total, they are locally substantial, sometimes outpacing the city’s own expenditures. The very existence of a BID-like entity is strong evidence that neighborhoods have difficulty providing public goods without a coercive mechanism, as theorized by Olson (1971). Of the 253 cities sampled for this paper, slightly less than one-third had at least one BID in 2000. 12

BID information for the city of Los Angeles comes from city council files.

9

3

Data & Estimation

In order to evaluate whether heterogeneity leads to BID adoption, I combine data from a number of different sources. In addition, I compare across three different measures to heterogeneity to differentiate between overall heterogeneity and spatial heterogeneity. I construct the outcome measure – whether or not a city has adopted a BID in a given year – from my survey of 253 cities.13 This includes all 114 cities with a population of over 25,000 people in 1980 and the universe of cities in San Diego, Riverside, Orange and Los Angeles counties, which account for an additional 123 cities. I restrict the analysis to the state of California in order to look at a large number of cities while keeping BID rules constant. The four major Southern California counties provide a large universe of cities in within a otherwise relatively constant institutional environment. For each city, I searched online to find whether that city had any BIDs and what the earliest year of BID adoption was. If that search did not yield any evidence of a BID, I searched the city’s own webpage to find information about BIDs. As a supplement, and if neither of these methods revealed any information, as it frequently did not, I called the city to ask an economic development official about BIDs in that city,14 and took that official’s word as the final say. In order to determine that it is heterogeneity that determines BID adoption – not other variables such as population or share poor that are correlated with heterogeneity – I employ a rich set of controls. From the 1980, 1990 and 2000 decennial censuses, I use data on population, income, race, and household characteristics. From the 1982, 1987, and 1997 economic census I use data on the amount of retail sales; from the 1977, 1987, and 1997 13

This sample does not include the eight cities (from the four Southern California counties, San Jacinto; from the larger cities sample, Martinez, Ontario, Upland, Milpitas, Santa Clara, Sunnyvale and West Sacramento) I was forced to drop that did not respond to repeated enquires. The survey has a 97 percent response rate. 14 In general, I enquired whether a city had any BIDs, so the data are most complete for BIDs as of this decade. In very few instances did a city say that it had a BID in the past and did not have one now.

10

census of governments I use city expenditure per capita.15 Data on crime by city comes from the FBI’s Uniform Crime Reports from 1980, 1990 and 2000; information on municipal institutions comes from the 1987 Census of Governments. I measure the impact of heterogeneity on BID adoption at the city level by estimating

Pr(BIDi,t = 1|heti,t , controlsi,t , yeart ) = Φ(α0 + α1 heti,t + α2 controlsi,t + α3 yeart ) .

(1)

Observations are at the city-year level and the binary choice of whether or not a city i has any BIDs at time t, BIDi,t , is determined by a measure of heterogeneity, heti,t , a set of controls, controlsi,t , and year dummies, yeart . I use the year dummies to control for area-wide shocks such as the decline of the aerospace industry, which could affect the regional provision of public goods. I re-scale all heterogeneity measures so that heterogeneity is increasing from 0 to 1. If heterogeneity is associated with BID adoption, α1 will be positive and different from zero. Note that the estimation does not include city fixed effects. Across the variables of interest, the variance between cities is usually roughly double the variance within cities. Also, conceptually, restricting the model with city-level fixed effects would require all identification of BID adoption to come from changes within a city, whereas the theory concentrates on a city’s long-term level of heterogeneity.16 In order to further address the problem of causality, I present results where I restrict the sample to 1980 and 1990 census data and limit the dependent variable to whether a city ever adopted a property BID. The property BID law, which expanded the mandate of BIDs to tax properties, was passed only 1995. Therefore, city heterogeneity before this law – 1980 and 1990 – is clearly not determined by property BID adoption, and is exogenously related 15

See the appendix for a complete list of control variables. I do cluster standard errors at the city level in order to correct for the repeated observations of multiple cities which would otherwise overstate the true degrees of freedom in the regression. 16

11

to property BID adoption. I measure the key variable of interest, heterogeneity, with three separate indices. To motivate the use of these indices, consider two cities A and B, each with blue, green and yellow people, as depicted in Table 2. Both cities have one-third of their population in the blue, green and yellow groups respectively. However, in city A, each of the three neighborhoods has an equal number of each group; in city B, each group is completely segmented by neighborhood. The first measure of heterogeneity is the one used almost exclusively in this literature, frequently called the index of fragmentation or Hirfendahl index, H.17 This index is calculated as H = 1 −

Pn

i=1

s2i , where n is the number of groups and si is the share of each group in the

population (the share of each color’s population in the example). It goes from zero to one, where a city with zero has all members in one group, and the index for a city approaches one when the population is split equally between a very large number of groups. Note that because city A and city B have the same overall distribution of people across colors, they have the same Hirfendahl index, H = 1 −

(1/3)2 = 2/3.

P

Note also that while this measure does a good job accounting for overall diversity, it obscures the fact that people in city B are much more spatially heterogeneous than people in city A. The dissimilarity index, which measures segregation, or the amount by which the population distribution in the average neighborhood differs from the population distribution in the city as a whole, captures this difference. Note that this is a city-level measure of segregation, composed from neighborhood-level information. The index for a target group (e.g. blue residents relative to all other residents) in a city is calculated as PE

D= 17

e=1 (te kpe

− P k) 2T P (1 − P )

This index is identical to the Hirfendahl index for measuring concentration within an industry.

12

, where e ∈ E indexes neighborhoods, te is the total population of neighborhood e, pe is the target group’s share of population in neighborhood e (e.g., share of poor people in neighborhood e), P is the overall share of the target group in the city, and T is the overall population of the city. This index measures how evenly one group is distributed across neighborhoods relative to all other groups. To calculate dissimilarity indices, I use information about groups (race, income, etc.) at the census tract level. In city A, where blue people are distributed equally across neighborhoods, the dissimilarity index is 0; in city B, where blue people are completely segregated in one neighborhood, the dissimilarity index is 1. Therefore, I expect the dissimilarity index to be a better predictor – larger and more significant than the overall heterogeneity index – for whether or not a city will adopt a BID. Finally, cities most likely to adopt BIDs should be those with both a heterogeneous population and a population distributed unequally across neighborhoods. Therefore, theory suggests that the best predictor of BID adoption should be the interaction of the H and D indices. This is the final measure of heterogeneity that I employ, controlling for the individual indices.

4

Results

To summarize the variance in these measures, Panel B of Table 1 presents first mean Hirfendahl and then mean dissimilarity indices for the Southern California and larger California cities samples. The proxies for heterogeneity of demand are poverty, income, education, household type (whether or nor a household is married or has children), and age. As a robustness check, I also include the heterogeneity of race, which I do not expect to influence decisions by neighborhoods of firms. To calculate Hirfendahl indices by city, I take

13

the categories into which the Census has divided the population as given.18 In contrast, the dissimilarity index requires the researcher to choose one group to which the rest of the population is compared; in this case the theory offers no particular guide as to which point of the distribution should matter most. Empirically, it seems that dissimilarity indices for poorer and less educated groups are the most associated with BID adoption, and thus these are the ones I present here.19 The two pairs of columns in Panel B of Table 1 compare the mean Hirfendahl index in cities with BIDs to the mean Hirfendahl index in cities without BIDs for all Southern California cities; the third and fourth columns compare larger cities with and without BIDs. With the exception of the dissimilarity index for blacks, all differences in these column pairs are significant. The difference in racial heterogeneity between BID and non-BID cities stems from the difference in the distribution of Hispanics. Note that in general, for this panel and the one below, large cities are more heterogeneous. In sum, Table 1 present modest evidence that spatial heterogeneity differs more between BID- and non-BID cities than overall heterogeneity. Moving this comparison to a regression framework, Table 3 presents estimation results for Equation 1 where heti,t is heterogeneity of poverty; coefficients in this table are marginal effects from the probit estimates where BID adoption in city i at time t is the outcome variable, and standard errors are underneath the coefficient estimates. Each column in each panel of the table is a separate estimation; this allows comparison between the Hirfendahl index, the dissimilarity index, the two indices together, and their interaction and main effects. The top panel of the table presents results for all southern California cities, and 18

In a given year, all cities have the same number of possible categories for a variable (e.g., income $0$10,000, $10,000-$20,000, etc). Though the number of categories into which the data are divided increases over time, it does not vary over cities. The change by year should be absorbed by the year effects in the regression unless the impact of the division of categories varies substantially by city. 19 It is possible that this association is due to a non-linearity in demand as a function of income and education, and this may be a point worthy of future research.

14

the first three columns of the table control for population and its square, racial shares, household characteristics and income. Olsonian theory suggests that larger populations should negatively impact the ability to provide collective goods event absent any problems of heterogeneity, and I include both population and its square in all estimations in the likely event that population does not have a linear effect (Olson, 1971). In this first specification, increases in overall heterogeneity (H) and spatial heterogeneity (D) are both individually significantly associated with BID adoption, as is their interaction. As it will be for the rest of the paper, the dissimilarity index for poverty measures the percentage of people with income less that one and one-half times the poverty line who would have to move to make the within-city distribution of people with income one and onehalf times the poverty line equal across neighborhoods. Controlling for the main effects, in the fourth column, the interaction remains significant, but the main effects lose significance. The increase in the coefficients on the interaction suggests that the regression suffers from multi-collinearity. The second set of estimates in columns five to eight controls additionally for retail sales, total sales, and city government expenditure, all in per capita terms. Estimates in columns nine to twelve further control for institutional features of cities – form of government, year of incorporation, presence of at-large council members – and crime and crime clearance rates. With the addition of these controls, though overall and spatial heterogeneity remain positively associated with BID adoption, they are no longer significant. As the number of control variables increases20 , the problem of multicollinearity worsens. Panel B of Table 3 repeats this exercise for the sample of all larger California cities. Across the three specifications, an increase in overall heterogeneity remains associated with BID adoption, while the D spatial measure fails to explain a city’s propensity to adopt a BID. 20

For the Southern California sample, this restricts the analysis to larger cities, as not all variables are available for smaller cities.

15

Again, all estimates using the main effect and the interaction suffer from multi-collinearity. Because the taste for public goods may be correlated with things other than the poverty level, Table 4 presents the most complete specification from Table 3 using education, household type, family income, age and race as inputs for the measures of heterogeneity in the preferred sample of larger cities. Because the specification with the interaction and the main effects continually suffers from multi-collinearity, I omit it here. For reference, Panel A of Table 4 repeats the coefficients on poverty from the most complete specification of the previous table. Again, each column in each panel reports the results from a regression. Except for overall heterogeneity of household type, neither overall nor spatial heterogeneity, individually or in interaction, is significantly associated with BID adoption. However, out of the 18 cases, 16 are in the positive direction posited by theory. Because BIDs are profit-driven enterprizes, racial heterogeneity should not impact their adoption, and this prediction is borne out Panel F of Table 4. Neither the Hirfendahl index of overall racial heterogeneity, not the dissimilarity indices for Hispanic or African-American populations are significant in any of the specifications. Thus far, BID adoption has been measured dichotomously, possibly obscuring useful information about the strength of BIDs within a city. To address this concern, Panel A of Table 5 presents results from a regression of the three measures of heterogeneity – overall, spatial and their interaction - on the number of BIDs per total retail sales. With one exception, whether the heterogeneity measured is that of poverty, household type, or family income, it is unassociated with the intensity of BID adoption. The theoretical conjecture that spatial heterogeneity (D) should be more associated with BID adoption is not supported. In addition to this measure of BID intensity, the data allow for a robustness check on the direction of causality. Because BIDs are small, it is difficult to conceive that their adoption drives heterogeneity in cities, and not vice-versa. However, it is very plausible that an array of BID-like policies could retard Tiebout sorting at the municipal level, and 16

that this array of adoption could be correlated with the timing of BID adoption. In order to address this concern, I use the 1994 change in law which allowed for the assessment of property owners for BIDs (in addition to merchants) to look at the adoption of property BIDs post-1995 as a function of heterogeneity characteristics before the passage of this law. Pre-law heterogeneity is clearly not affected by post-law BID adoption. Thus I restrict the sample to observations from 1980 and 1990, and present the results of separately overall heterogeneity, spatial heterogeneity and their interaction on property BID adoption only in the bottom panel of Table 5. As in the previous panel, each cell in Panel B reports the results of a regression of a measure of heterogeneity, regressed here on whether a city ever adopted a property BID. In almost all cases, for heterogeneity of poverty, household type and income, heterogeneity is positively associated with property BID adoption as the theory predicts. However, only the interactions terms are significantly associated with property BID adoption, and this effect is difficult to evaluate because including the main effects (not shown) leads to serious problems of multi-collinearity.

5

Conclusion

In sum, I find only weak evidence that heterogeneity impacts BID adoption through the channels suggested by other authors. Though the heterogeneity coefficient usually has the positive sign expected by theory, it also frequently fails to significantly explain the variance in BID adoption at the city level. The negative result is robust to spatial and non-spatial measures of heterogeneity, and holds true across the demographic measures usually associated with demand for public goods: household type, income, poverty, education and age. Why does heterogeneity fail to explain much of the variance in city-level BID adoption patterns? First, it may be that BID adoption is predominantly explained by other features,

17

such as the quality of city governance, or the age of downtown structures that frequently contributes to physical collective action problems in the provision of local public goods. In this sample, year of incorporation persistently significantly explains BID adoption (see Brooks (2006) for a more thorough exploration of this effect). Second, it may be that California’s Proposition 13, which limits revenue collected from property tax, has left cities in California so constrained that even cities which would otherwise provide these services in-house turn to the BID process. This story may explain why BID adoption is commonplace among larger cities, but it fails to explain why so few smaller cities have BIDs. Finally, it may be that while city-level factors affect BID decisions, they are overwhelmed by the importance of neighborhood factors. Unfortunately, the data collected here – and indeed any publicly available data – are not available at a small enough geographic level to carefully evaluate this possibility.

18

References Alesina, Alberto, Baquir, Reza, and Easterly, William, 1999. “Public Goods and Ethnic Divisions.” Quarterly Journal of Economics pages 1243–1284. Alesina, Alberto, Baquir, Reza, and Hoxby, Carolyn, 2004. “Political Jurisdictions in Heterogeneous Communities.” Journal of Political Economy 112(2): 348–396. Alesina, Alberto and Spolaore, Enrico, 1997. “On the Number and Size of Nations.” Quarterly Journal of Economics pages 1027–1056. Brooks, Leah, 2005. “Volunteering to Be Taxed: Business Improvement Districts and the Extra-Governmental Provision of Public Safety.” McGill University working paper. Brooks, Leah, 2006. “Unveiling Hidden Districts: Describing and Assessing the Adoption Patterns of Business Improvement Districts.” McGill University Working Paper. Calanog, Victor Franco M., 2004. “Business Improvement Districts: Crime Deterrence or Displacement?” Unpublished manuscript. Chaudhary, Latika, 2005. “Social Divisions and Public Goods Provision: Evidence from Colonial India.” Unpublished manuscript. Epple, Dennis and Romano, Richard, 1996. “Public Provision of Private Goods.” Journal of Political Economy 104: 57–84. Helsley, Robert W. and Strange, William C., 1999. “Gated Communities and the Economic Geography of Crime.” Journal of Urban Economics 46: 80–105. Hoyt, Lorlene, 2005. “Do Business Improvement District Organizations Make a Difference?” Journal of Planning Education and Research 25: 185–199. Olson, Mancur, 1971. The Logic of Collective Action: Goods and the Theory of Groups. Harvard University Press. Poterba, James M., 1997. “Demographic Structure and the Political Economy of Public Education.” Journal of Policy Analysis and Management 16(1): 48–66. Temple, Judy A., 1996. “Community Composition and Voter Support for Tax Limitations: Evidence from Home Rule Elections.” Southern Economic Journal 62(4): 1002–1016. Vigdor, Jacob L., 2004. “Community Composition and Collective Action: Analyzing Initial Mail Response to the 2000 Census.” Review of Economics and Statistics 86: 303–312.

19

6

Appendix: Control Variables

Racial Shares • share African American • share Hispanic • share Asian • Source: Census of Population and Housing, 1980, 1990 and 2000, accessed from UCLA and ICPSR Household Characteristics • mean household size • share of households with children • share of single mother-headed households • population share 65 or older • share with high school education • share with bachelor’s degree • Source: Census of Population and Housing, 1980, 1990 and 2000, accessed from UCLA and ICPSR Income • median household income • mean family income • Source: Census of Population and Housing, 1980, 1990 and 2000, accessed from UCLA and ICPSR Business Characteristics • retail sales per capita • total sales per capita • city government expenditure per capita • Source: City and County Data Books, 1988, 1994 and 2000, accessed via the University of Virginia; contain information from the 1982, 1987 and 1997 economic censuses 20

Crime • offenses per capita • clearance rate • Source: FBI Uniform Crime Reports, 1980, 1990, 2000, accessed via ICPSR Institutional Characteristics • year of incorporation • whether council has at large members • whether city operates under homerule • whether or not city uses mayor-council form of government • Source: 1987 Census of Governments, accessed from Census

21

Table 1: Comparing BID and Non-BID Cities All Southern California Cities w/o BIDs with BIDs A. Descriptive Statistics Unique Cities in 2000 Observations Share of Total Share with any property-based BIDs Population Mean family income Per capita city expenditures, $1000s N for city expenditures

132 408 0.83 54,318 (164,540) 56.4 (42.8) 0.58 (0.44) 158

27 57 0.17 0.22 292,962 (680,962) 60.5 (33.3) 0.92 (0.50) 49

0.43 (0.20) 0.90 (0.05) 0.79 (0.08) 0.62 (0.19) 0.32 (0.04) 0.33 (0.03) 399

0.50 (0.15) 0.92 (0.02) 0.83 (0.05) 0.54 (0.18) 0.33 (0.08) 0.33 (0.04) 57

0.18 (0.09) 0.19 (0.09) 0.15 (0.09) 0.09 (0.07) 0.14 (0.09)

0.26 (0.08) 0.27 (0.09) 0.25 (0.10) 0.14 (0.07) 0.19 (0.08)

0.33 (0.15) 0.20 (0.11) 357

0.37 (0.16) 0.29 (0.12) 57

B. Outcome Measures Hirfendahl Index Poverty Family Income Education Household Type Age Race N Dissimilarity Index Poverty personal income < 1.5 * poverty line Family Income income < half of the county mean Education high school education or less Age age less than 18 Household Type households with children Race Black Hispanic N

All Larger California Cities w/o BIDs with BIDs 58 220 0.51 *

*

* * * *

* * * * *

* *

102,836 (226,448) 46.4 (26.2) 0.60 (0.40) 220

56 122 0.49 0.27 220,698 (489,910) 58.0 (28.4) 0.91 (0.59) 122

0.44 (0.16) 0.91 (0.03) 0.79 (0.07) 0.57 (0.19) 0.30 (0.03) 0.32 (0.02) 220

0.48 (0.15) 0.92 (0.02) 0.83 (0.05) 0.56 (0.18) 0.32 (0.05) 0.32 (0.02) 122

0.23 (0.08) 0.23 (0.08) 0.19 (0.09) 0.10 (0.05) 0.15 (0.07)

0.27 (0.07) 0.27 (0.07) 0.24 (0.10) 0.13 (0.05) 0.19 (0.07)

0.36 (0.14) 0.25 (0.10) 220

0.37 (0.12) 0.29 (0.12) 122

* Significant difference at the 0.05% level. Notes: This table compares cities with and without Business Improvement Districts (BIDs). Figures reported are means, with standard deviations below means. Hirfendahl indices are calculated from the categories given by the census. Dissimilarity indices are calculated from tract-level data, and compare the group listed below the variable name to the rest of the population. Source: BID information from author's survey; see Appendix for description of other variables.

22

* * *

* * *

*

* * * * *

Table 2: Example of the Difference Between the Hirfendahl and Dissimilarity Indices City A neigh 1 neigh 2 neigh 3 share

blue 5 5 5 1/3

green 5 5 5 1/3

yellow 5 5 5 1/3

total 15 15 15 45

H = 1 − 3i=1 (1/3)2 = 2/3 Dblue vs everyone else = 0 P

City B blue green yellow neigh 1 15 0 0 0 15 0 neigh 2 neigh 3 0 0 15 share 1/3 1/3 1/3

total 15 15 15 45

H = 8/9 Dblue vs everyone else = 1

Notes: This example shows that two cities with the same overall distribution of population across categories may have very different distributions of people across neighborhoods. The overall heterogeneity across categories is measured by the H index, and the heterogeneity across neighborhoods is measured by D.

23

Table 3: Heterogeneity of Poverty Weak Explanation of BID Adoption by City A. All Cities in Southern California H

0.56 (0.250)*

D

456

B. All Larger California Cities H 2.18 (0.667)** D

431

controls population population squared racial shares household characteristics income business characteristics institutional features crime

2.23 (0.702)**

342

0.48 (0.64)

H*D N

1.86 (1.08) -0.65 (1.72) 1.31 (3.71) 342

0.53 (0.219)*

H*D N

2.11 (1.020)*

1.25 (0.425)** 431

0.17 (0.27) -0.27 (0.39) 1.64 (0.805)* 431

0.82 (0.83)

3.57 (1.751)* 174

-0.08 (1.22) -2.55 (2.36) 7.71 (4.84) 174

2.09 (0.716)**

342

2.36 (1.200)* 342

1.87 (1.20) -0.31 (1.78) 1.11 (3.89) 342

342

1.48 (0.99)

174

174

0.73 (0.67)

1.64 (1.44) 174

-0.96 (1.07) -2.78 (1.99) 7.17 (4.10) 174

342

1.89 (1.21) 342

1.97 (1.11) -0.30 (1.74) 0.55 (3.81) 342

0.51 (0.82)

174

174

0.43 (0.65)

342

342

2.00 (1.18) 342

x x x

x x x

x x x

x x x

x x x

x x x

x x x

x x x

x x x

x x x

x x x

x x x

x x

x x

x x

x x

x x

x x

x x

x x

x x

x x

x x

x x

x

x

x

x

x x x

x x x

x x x

x x x

* Significant at the 0.05% level. * Significant at the 0.01% level. Notes: Each column in each panel of this table reports the marginal effect from a probit regression of the heterogeneity of poverty regressed on BID adoption by city; standard errors are below the coefficients. D is the dissimilarity index for people with incomes less than 1.5 times the poverty line. The table shows that BID adoption is sometimes predicted by the Hirfendahl index. The dissimilarity index is not predictive of adoption patterns, and the interaction specification suffers from multicollinearity. Individual controls are as listed in the appendix. Source: BID information from author's survey; see Appendix for description of other variables.

24

Table 4: Across Heterogeneity Types, Neither Index Explains BID Adoption A. Poverty H 2.09 (0.716)** D

0.43 (0.65)

H*D N

342

C. Education H -0.834 (1.20) D

342

E. Age H

342

342

0.492 (0.69) 342

H*D 342

N

342

342

N

342

0.77 (0.79)

342

2.83 (2.45) 342

-0.099 (0.38)

D, hispanic N

0.57 (2.26) 342

0.216 (0.42)

D, black 0.869 (3.04) 342

342

H*D

F. Race H 0.246 (0.99)

0.258 (0.71)

H*D

D. Household Type H 2.78 (1.75) D

1.149 (3.28)

D

N

1.89 (1.21) 342

0.51 (0.59)

H*D N

B. Family Income H 0.208 (1.29) D

342

342

0.199 (0.48) 342

* Significant at the 0.05% level. * Significant at the 0.01% level. Notes: Each column in each panel of this table reports the marginal effect from a probit regression of the heterogeneity of the variable noted regressed on BID adoption by city; standard errors are below the coefficients. The table shows that only heterogeneity of household type is significantly related to BID adoption. Spatial heterogeneity (D) is not explain BID adoption better than overall heterogeneity (H). All regressions control for the maximal set of variables from the previous table. Source: BID information from author's survey; see Appendix for description of other variables.

25

Table 5: Heterogeneity Indices Robust to Alternate Outcome Measures A. Outcome is Property BID Adoption Poverty H 0.67 (-0.35) D 0.45 (-0.28) H*D 0.91 (0.434)* N 228 228 228

B. Outcome is BIDs/Retail Sales Poverty H -0.02 (-0.06) D -0.10 (-0.05) H*D -0.11 (-0.11) N 342 342 342

Household Type 0.22 (-0.47) -0.21 (-0.31)

228

228

Family Income 0.35 (-0.77) 0.53 (0.266)* -0.24 (-0.85) 228

Household Type 0.24 (-0.27) 0.15 (-0.11)

342

342

228

228

2.26 (1.015)* 228

Family Income 0.23 (-0.17) -0.09 (-0.06) 0.50 (-0.33) 342

342

342

-0.06 (-0.23) 342

* Significant at the 0.05% level. * Significant at the 0.01% level. Notes: Each cell in this table is a marginal effect either from a probit regression (top panel) or OLS regression (bottom panel) of heterogeneity regressed on BID adoption by city with standard errors below the coefficients. Controls are the maximal set described in the previous table. Source: BID information from author's survey; see Appendix for description of other variables.

26