«AmericAn neighborhoods: inclusion And exclusion Volume 16, Number 3 • 2014 U.S. Department of Housing and Urban Development | Office of Policy ...»
If a census tract does not have an HOA throughout the entire sample period (1970 to 2008), this observation is equivalent to a “censored observation” (never observed to have failed) in the duration analysis terminology.7 We note a caveat to our approach. The address of an officer in our dataset is self-reported, and two potential reasons may point to why the address may not be the actual residence of the officer. First, the officer may have put the HOA’s management office as his or her address. Second, the officer uses the HOA unit as a second or vacation home or rents it out. We have devised an algorithm to identify these suspect HOAs, and we are forced to drop them from our sample.8 We are confident that our assumptions are reasonable and, if anything, err on being conservative in terms of determining the scope of HOAs in the state.9 Census Data For the time-varying covariates, we supplement our HOA map with data on census tract economic and demographic characteristics from the Geolytics Neighborhood Change Database. This database contains census data and normalizes the census tract boundaries to 2000 geographic definitions so that the tracts can be analyzed as a panel across 1970, 1980, 1990, and 2000 census years. Tracts enter the analysis with census covariate values from 1970, and, as long as they remain without an HOA, their census covariates change with the decennial census.
HOAs are rarely, if ever, dissolved.
Few census tracts exist in which the first HOA was formed before 1970, the start of our sample period. For this analysis, we assume these tracts to have had the first tract formed in 1971 (that is, “failure” almost immediately).
We will not elaborate on the algorithm here, but a nonexhaustive list follows of reasons that would cause us to reject an address as being the actual location of an HOA: (1) the address reported is zoned commercial, (2) identical addresses are reported for more than one HOA (which is likely an office building), and (3) the address belongs to a different city from the other officers in the same HOA.
We test and verify the robustness of the HOA boundary assignment in a separate paper (Meltzer and Cheung, 2014).
Cityscape 79Cheung and Meltzer
On the demand side, we include in our main specification the following tract-level variables as covariates: percent Black; percent Hispanic; percent under 5 years old; percent 65 years old and older; percent with a bachelor’s degree or higher; average family income10; percent foreign born;
percent taking public transit to work; and percent living in the same house 5 years ago. On the
supply side, we include as covariates in our main specification the following tract-level variables:
(1) distance to the central business district,11 (2) vacancy rate, (3) owner-occupancy rate; and (4) percentage of houses that are 30 years old or older.12 Finally, to explore the importance of the institutional context, we include public finance variables on government revenues and expenditures from the U.S. Census of Governments. We rely on data from 1972, 1982, 1992, and 2002, the years closest to the decennial years for which a census of governments for all municipalities is conducted. Each tract is assigned the revenue or expenditures of its host municipality. Because some census tracts are not located in incorporated cities, the sample size is significantly smaller for the models with public finance variables. All variables are real, per capita values. On the revenue side, we include total own-source revenue.13 On the expenditure side, we include total general expenditures, as well as spending on four major categories that are presumed substitutable with HOA expenditures: (1) roads, (2) police, (3) solid waste collection, and (4) parks and recreation.
Description of the Sample Our data cover census tracts in 26 of the 67 counties in Florida. We dropped counties from the analysis because of incomplete data. First, areas designated as census tracts in 2000 and 2008 were not necessarily designated as tracts in 1970 and 1980, and we need to be able to follow the census tracts through the entire study period to estimate the hazard ratio. Note that areas that were not designated as tracts in 1970 tend to be rural and nonmetropolitan; these areas, even today, do not tend to have HOAs. We also drop counties if they were missing subdivision and GIS parcel files or because of lack of variation in HOA membership. Exhibit 1a shows that our data ultimately cover most urban areas in the state. By retaining the most populous counties that together account for 85 percent of the population of the state, our sampling method does not cause us much concern for the validity of our results.
All dollar values throughout this article have been expressed in 2000 dollars, based on the Consumer Price Index.
We used GIS to measure the straight-line distance between the centroid of a census tract and its central business district (CBD). The CBD is the point in the city designated by the Census Bureau as the center of the metropolitan statistical area.
In other specifications, we explore more covariates, such as percentage with a high school diploma or higher, unemployment rate, and poverty rate. Because these covariates do not add much to the main results, they are not included in the reported specifications.
We also run models with revenue from three major categories (property taxes, sales taxes, and charges/fees), but the results do not add anything substantively to the model with aggregate revenues. Therefore, it is omitted from the presented analysis.
Our entire working dataset consists of 2,176 census tracts, with a mean population of 3,127, in the demand- and supply-side models, and 1,270 census tracts, with a mean population of 3,493, in the public finance models. A list of all the variables in the analysis, along with their summary statistics pooling all four censuses together, is presented in exhibit 1b.
HOAs in Florida Like trends in the rest of the United States, HOAs in Florida have proliferated during the past 30 years and during the past decade in particular. Exhibit 2 provides evidence of this proliferation.
The first recorded HOA was established in 1959 and, since 1990, the number of HOAs in Florida has increased by nearly 140 percent. To put this growth in context, the number of new housing units in Florida has increased by 14 percent during the same period, and the number of units in HOAs nationwide has increased by about 50 percent (Community Associations Institute, 2008).
The maps in exhibit 3 also illustrate that the growth of HOAs has been unevenly distributed throughout the state. They have primarily emerged along the coasts and increasingly in the central peninsula and pockets of the northern panhandle. As expected, they are most prevalent in the central and suburban parts of the state, where developable land is abundant. The number of jurisdictions with HOAs has grown dramatically as well. In 1970, only 39 cities (out of 397) in our sample had an HOA. This number grew to 113 by 1980, 158 by 1990, and 178 by 2008. Within a jurisdiction, the number of HOAs varies considerably; as of 2008, some places had only one HOA while others had 300 or more.
Exhibit 2 Number of Homeowners Associations in Florida Over Time Source: Meltzer, Rachel. 2013. “Do Homeowners Associations Affect Citywide Segregation? Evidence From Florida Municipalities,” Housing Policy Debate 23 (4): 688–714 Exhibit 3 Spread of Homeowners Associations Across Florida
Regression Results We fit a Cox proportional hazards model with time-varying covariates to predict the likelihood of HOA formation in a census tract. All standard errors are clustered by census tract.
Demand-Side Predictors We first describe the results for the models including demand-side predictors only (see exhibit 4).
Column (a) reports the coefficient estimates, while column (b) reports the hazard ratios (exponentiated coefficients). We see that race/ethnicity and income are more significant predictors than age or education.14 Neighborhoods with higher shares of Black and foreign-born residents are less likely to form HOAs. The likelihoods of forming HOAs specifically are reduced by 37 and 59 percent, respectively, when the share of Black or foreign-born residents in a tract goes up by 1 unit (that is, the share rises from 0 to 100 percent).15 Although the coefficient on the share Hispanic is not significant, it is also negative. Tracts with higher average family incomes are more likely to form HOAs— 14 percent more likely for a $10,000 increase. Because we know that HOA properties tend to sell at
Note that more parsimonious models without education produce essentially the same coefficient for income (it is slightly larger); therefore, multicollinearity is not a concern.
Hazard ratios are obtained by taking e to the power of the coefficient.
higher prices than other comparable houses (in addition to the required membership fee), new HOA residents are likely similar (in terms of affluence) to those already living in the area. That is, this finding is not consistent with the prediction that the HOA is creating an enclave for relatively affluent households in the context of less affluent neighborhoods. Along similar lines, the findings suggest that HOAs are more likely to emerge in predominantly nonminority neighborhoods—this coefficient could be picking up some income-related mechanism, but it may also reflect a different proclivity for exclusionary communities. We also find the neighborhoods with higher shares of newcomers and commuters using public transportation are less likely to form HOAs. These results suggest that HOAs tend to form in younger (or more transient) communities that are not transit oriented (that latter finding could, again, be picking up some differences in income as well).
Supply-Side Predictors Next we run models with only supply-side predictors; these results are displayed in exhibit 5. All the variables are significant. HOAs are more likely to form in neighborhoods that have higher vacancy and homeownership rates and, on average, newer housing. Therefore, HOAs are formed in the context of new housing developments (as predicted), they tend to govern homeowners (versus renters), and they tend to emerge in less constrained markets (as indicated by the reverse relationship with vacancy rates). Neighborhoods located farther from the central business district (CBD) (that is, closer to the municipal outskirts) are also more likely to form HOAs. A 1-mile increase in distance to the CBD increases the hazard ratio by 0.7 percent. This finding is consistent with the expectation that HOAs need larger swaths of land, which tend to be situated toward the city’s fringe.
We proceed by combining demand- and supply-side variables into a single model. These results are displayed in exhibit 6, columns (a) and (b). The general pattern of the coefficients is consistent;
however, the coefficients do tend to decrease in magnitude (this pattern is consistent with the fact that the demand- and supply-side variables inevitably pick up overlapping mechanisms). We note two important changes in the coefficients: (1) education is now significant (still positive) and (2) distance to the CBD assumes a slightly larger coefficient (it is still positive and significant).
Number of tracts (observations) 2,176 *** significant at the 1-percent level.
Note: Robust standard errors (in parentheses) clustered at the census tract.
Exhibit 7 presents a plot of the survival curve calculated at the mean values. The horizontal axis begins at 1970 (year 0). The survival falls steeply for roughly the first 15 years, representing the rapid adoption of HOAs in the 1970s and 1980s. It hits 0.5 around 1976. The survival curve flattens in the 1990s and 2000s. This slower rate of adoption suggests that HOAs have become more clustered, because fewer tracts are receiving their first HOA in later years.16 We also stratify the data by running separate hazards for different sizes of the first HOA to see if it explains their proliferation across different neighborhoods; we see no evidence to suggest that the overall formation patterns of HOAs are differentiated by the actual size of the HOA.
Exhibit 7 Survival Curve From Full Model Note: The estimated survival curve (at mean values) is plotted with 1970 as the beginning of the analysis time.
Finally, we also augment the model by stratifying by county. In this specification, we allow for the hazard baseline to vary by county to control for any unobserved heterogeneity in the broader geography that could be correlated with the likelihood of HOA formation. As the coefficients in the second vertical panel of exhibit 6 indicate, the results are substantively the same, except now distance to the CBD is insignificant (but still positive).
Municipal Institutional Predictors We add to the combined demand- and supply-side covariates measures of citywide fiscal conditions in exhibit 8. In all specifications, we stratify by county.17 Because of space constraints, we report only the coefficient estimates rather than the hazard ratios. Column (a) adds the total per capita general expenditures of the city, and the coefficient is significantly negative and large: a oneunit change in city expenditures (an increase of $1,000 per capita) will decrease the hazard ratio by 20 percent. This finding suggests that census tracts located in cities that have high public spending are less likely to form an HOA, all else being equal. This result provides additional evidence to Cheung (2008a, 2008b) that homeowners may regard public and private government spending The public finance results tend to be less stable with respect to the mix of covariates and whether we stratify by county.