Interjurisdictional Determinants of Property Assessment Regressivity Justin M. Ross ABSTRACT. The previous literature on vertical equity in property assessment has focused on parcellevel data within a single area and has produced mixed conclusions on whether the process is progressive or regressive. This paper advances the discussion to identifying what differences between jurisdictions might account for the mix of findings. Using data from Virginia cities and counties between 2001 and 2007, evidence is presented that indicates having tax maps available online, appointed assessors, and senior citizens all influence the level of regressivity observed between jurisdictions. Overall, the results support the hypothesis that interjurisdictional differences are determinants of vertical inequity. (JEL H71, H73) I. INTRODUCTION Like all forms of taxation, the equity consequences of real property taxation have been a subject of interest to policy makers and researchers for decades (see Black 1977; Ihlanfeldt 1982). Nevertheless, the theory and empirics of real property tax incidence are arguably one of the most contentious subjects in public finance. 1 Unlike most taxes, property taxation undergoes an assessment process that adds an additional source of vertical equity concerns. Most taxes, like income or sales, are viewed as progressive or regressive on the basis of tax shifting and the wealth level of the groups on the receiving end of these shifts. While this may still be true for property, the incidence can be further augmented within a jurisdiction by an assessor determining the taxable value of the property owned by different households. As a result, researchers have long tried to determine the nature of vertical inequities in the property assessment process (Paglin and Fogarty 1972; Black 1977; Bell 1984; Clapp 1990). 2 Vertical equity issues arise in this process when the assessed value in proportion to true value, usually fair market value, systematically differs by price strata. For instance, if the assessment-to-sale price ratio (sales ratio) decreases as the sale price of the properties rises, the assessment process would be considered regressive. A progressive assessment process would be one in which the sales ratio increased with the sale price of the properties. In a recent and extensive literature review on the topic, Sirmans, Gatzlaff, and Macpherson (2008) found that studies of the assessment process routinely discover the existence of vertical inequities in residential properties, but that these studies are nearly evenly split between finding the process to be regressive or progressive, with a slight majority detecting a more regressive process. There are various technical reasons why vertical inequity might exist, such as time lags in assessments (Mikesell 1980; Clapp 1990; Ihlanfeldt 2004), lack of comparable sales across price strata (McMillen and Weber 2008), and the assessment appeals process (Weber and McMillen 2010). Some of this is due to a difference in econometric method- 1 Theoretically, the benefit view of property taxation has no place for incidence as the tax approximates a user fee for local public services (Fischel 2001). The capital tax view is more similar to the traditional public finance view of other forms of taxation, where there is excess burden and shifting (Zodrow 2001). Land Economics February 2012 88 (1): 28 42 ISSN 0023-7639; E-ISSN 1543-8325 2012 by the Board of Regents of the University of Wisconsin System 2 Black (1977) found, for instance, that effective property taxes in 1960 Boston were more regressive than were previously thought after it was taken into account that assessors tended to understate the value of property at the higher end of the market. The author is assistant professor, School of Public and Environmental Affairs, and affiliated faculty, Workshop in Political Theory and Policy Analysis, Indiana University, Bloomington.
88(1) Ross: Interjurisdictional Property Assessment 29 ology, as documented by Sunderman et al. (1990), as well as by Sirmans, Diskin, and Friday (1995). Another possibility, however, is that the mix of progressive or regressive findings is due to the literature s reliance on parcel-level data within a city, county, or metropolitan area. If assessment progressivity is partly an institutional outcome of local characteristics like tax structure or socioeconomic characteristics, then different conclusions on the progressivity or regressivity of the area would arise based on the researcher s sample choice. While differences across assessing jurisdictions have been studied for their role in creating horizontal inequity, the same is not true for vertical inequity. To my knowledge, the only previous paper to examine this possibility was by Smith, Sunderman, and Birch (2003). These authors examine the social and economic determinants of a vertical inequality index estimated for Indiana counties in 1999. 3 While the ability to produce this index from a wide sample of parcel-level data across counties allowed the analysis to take place, it also proved to be a limitation of the work. Only 39 of Indiana s 92 counties are represented in the sample, and these 39 county-level index scores are assigned to each of the 28,816 individual parcels that were used in constructing the index. Since the 21 explanatory variables were also county-level aggregates assigned to the individual parcels, the parameter estimates would have been weighted according to county representation in the overall sample of parcels, and the standard errors would have been artificially low. It is also important to remember that the assessor s office is not exempt from the political process. 4 In many cases, assessors are directly elected by the population or are otherwise appointed by elected officials. In a theoretical model of assessor behavior, John- 3 The vertical inequality index calculated and employed in Smith, Sunderman, and Birch (2003) was one developed by Birch, Sunderman, and Hamilton (1992). 4 There is a large literature distinguishing the consequences of having an elected or appointed policy maker in office. See Hoover (2008) and Whalley (2010) for recent literature reviews and empirical work. son (1989) assumed that elected assessors maximize their personal wealth by providing tax relief to voting taxpayers via lower assessments. In contrast, appointed assessors were assumed to appease elected officials who would like higher revenues while setting lower tax rates. Empirically, the consequences of having an appointed or elected assessor has been studied for its effect on both the sales ratio (Lowery 1982; Bowman and Mikesell 1989; Ross 2010) and measures of horizontal equity (Lowery 1984; Bowman and Mikesell 1989). 5 This paper aims to contribute evidence regarding possible institutional determinants of vertical equity in property assessment by drawing on data from Virginia cities and counties during the 2001 to 2007 time period. The Virginia Department of Taxation conducts a sales ratio study on an annual basis for each jurisdiction and computes a price-related differential index score that reflects the area s level and nature of vertical inequity. 6 The results indicate that assessment practices that differ between jurisdictions are important determinants of vertical inequity. II. BACKGROUND ON VIRGINIA Virginia has several attractive attributes for studying the institutional determinants of vertical inequity. A sales ratio study is conducted on an annual basis by the Virginia Department of Taxation (VDOT) at the jurisdictional level in which the assessment takes place: counties and cities. 7 These studies randomly sample fair market transactions that took place during the year across the state and compare them to the assessed value of the properties on the 5 Bowman and Mikesell (1989) present only the results of determinants of horizontal inequity using the coefficientof-dispersion as the dependent variable, but in footnote 9 of their paper they mention an attempt to estimate their econometric model using the sales ratio as the dependent variable. 6 The VDOT actually refers to the price related differential as a regression index, but this paper follows the terminology consistent with the academic literature on the subject. 7 In Virginia, a handful of cities have the same legal status as counties. As such, this paper adopts the term jurisdiction in referring to cities and counties.
30 Land Economics February 2012 FIGURE 1 Comparing Progressive, Regressive, and Uniform Assessments (Based on Simulated Data of 100 Properties, PRD 100 Is Regressive; PRD 100 Is Progressive) books at the time of the sales. 8 In addition to traditional measures of assessment performance, like the sales ratio and coefficient of dispersion, a common measure of vertical inequity known as the price-related differential (PRD) is also calculated and reported in the study. The PRD is defined as the mean sales ratio, multiplied by 100, and divided by a sales-weighted ratio. 9 The sales-weighted ratio is the total of the assessed values divided by the total of the selling prices of all sales in the classification, which allows transfers with a larger selling price to have a greater impact on the ratio than those with smaller selling prices. A value of 100.00 indicates a uniform relationship between assessed values and sell- 8 Fair market transactions are voluntary arm s length transactions of property, so as to exclude exchanges between relatives or properties seized in foreclosure and auctioned. 9 The measures reported by the VDOT do not multiply the results by 100, but that is done in this paper to create more readable regression coefficients in the results. ing prices of properties with different prices. An index above 100.00 indicates that less expensive property has a higher assessment/ sales ratio than more expensive property, and is therefore indicative of a regressive process. To help illustrate the significance of changes in the PRD that could be inferred for the individual households, Figure 1 demonstrates simulated market values and assessments under three regimes: uniform, progressive, and regressive. The simulated data has 100 homes, ranging from $100,000 to $300,000 in true market value, with $200,000 being the mean. Though the data is simulated, the nonuniform regimes are approximately based on a standard deviation of the sample data (4.4) above and below a uniform PRD score for regressive and progressive comparisons, respectively. The graph demonstrates, for instance, that a house with a market value of $250,000 would have a $318,000 assessment in a progressive regime, and a $200,000
88(1) Ross: Interjurisdictional Property Assessment 31 assessment in a regressive regime. By comparison, a progressive regime would give a $99,600 assessment to a property with $120,000 in fair market value, while a regressive regime would assess it at $135,000. Unlike the sales ratio and coefficient of dispersion, Virginia has no codified incentives to maintain a particular range of PRD scores. While a jurisdiction whose sales ratio is too low may lose its share of revenues from alcohol sales, it is only advised that maintaining a range between 95 and 105 constitutes a reasonable PRD score (Virginia Department of Taxation 2009). By contrast, the International Association for Assessing Officers says that a range of 98 to 102 suggests good practice. The levels of observation used in this paper s regression analysis are the jurisdictions in the years they conduct a reassessment of properties. Only 55 of the 134 districts conduct annual reassessments, and therefore they are the only districts to have a possibility of appearing in the dataset on an annual basis. Virginia s tax code stipulates that reassessments must be conducted at least every four years, with exceptions for very small districts, which may extend that up to six years. One of the caveats of the design of the sales ratio studies is that properties that undergo reassessment early in the year might appear undervalued in the sales ratio study if they are sold later in the same year. This may lead to misleadingly low sales ratios in areas with greater housing price appreciation, such those near Washington, D.C., in northern Virginia. If housing price growth rates vary by price strata, it could have an impact on measures of assessment progressivity. Since housing price data are not available during this period for all counties, to control for such growth effects this paper follows Bowman and Mikesell (1989) by including the annual population growth rate as a proxy variable. The sales ratio study is based on a random sample of all properties sold in the state; it is not stratified by property class or county. As a result, some districts have no reported PRD score for some property classifications. The sales ratio study reports residential property PRD scores separately for urban and suburban classifications, but most jurisdictions in the ratio studies only have residential property extensively sampled from the suburban classification. 10 Therefore, the PRD score employed in this paper will be that for suburban residential properties, as most counties do not have enough urban property sales to be picked up in the ratio study sample in the years they undertake a reassessment. In the 2001 to 2007 period of this data, 97 of the 134 jurisdictions have suburban residential PRD scores reported at least once, with a total sample size of 243. Among those 97, there are 13 that appear every year, and 35 that appear only once. 11 On average, Virginia s jurisdictions are slightly regressive, with a mean PRD score of 103.1 in the years they conduct reassessments. In addition to the availability of vertical inequity data across assessing jurisdictions, Virginia has other attributes that make it favorable for interjurisdictional comparisons. A strong Dillon s rule in the state reduces the differences in the local public finance systems. 12 Furthermore, Virginia s constitution allows for a jurisdiction to use appointed assessors, or to leave the maintenance of the tax rolls, land books, and providing the final assessment with a locally elected official, typically the commissioner of revenue. Property owners wishing to appeal their assessment are encouraged or required to speak with these officials first. As a result, 48 of the 134 jurisdictions have appointed officials dedicated to property assessment, though it is very com- 10 PRD scores are also computed for urban residential, multifamily residential, commercial, and agricultural property classifications, but most counties have only a few observations to serve as the basis of their PRD scores, making their validity in a study like this highly questionable, and they are therefore excluded. The urban and suburban distinction between residential property classes is based on the density of the area, not a proximal location to a city center. 11 The breakdown of the number of observations per group for a total sample size of 243 over seven years (number of groups number of appearances): 13 7, 1 6, 2 5, 2 4, 6 3, 38 2, and 35 1. 12 The Dillon rule is the doctrine of limited authority for local governments and is used in legal settings to determine what powers are held by local governments. In states with a strong Dillon rule, like Virginia, if there is reasonable doubt as to whether the local government has a particular power, then that power is not conferred to it. The state must specifically authorize or delegate particular powers to the local government(s).
32 Land Economics February 2012 mon among both types of assessors to contract out the actual assessments to private firms. The mean PRD score was 101.8 and 104.4 for appointed and elected assessors, respectively. Interestingly, Virginia has truth in taxation laws that require the constituency to be told that the assessors are not responsible for their tax burden, but just the assessment of their property. Presumably, these laws exist because of constituents pressuring their local assessors during or following the assessment process. Though locally elected officials all serve four-year renewable terms, election years are staggered among the cities and counties so that local elections across the state take place every two years. For all of these reasons, Virginia has been the frequent target of researchers studying interjurisdictional differences in assessments (e.g., see Bowman and Mikesell 1978, 1989; Bowman and Butcher 1986; Ross 2010). III. REGRESSION MODEL The dependent variable will be the PRD score, which is defined by the VDOT as the mean sales ratio, multiplied by 100, and divided by a sales-weighted ratio. The salesweighted ratio is the total of the assessed values divided by the total of the selling prices of all sales in the classification. It allows transfers with a larger selling price to have a greater impact on the ratio than those with lower selling prices. A value of 100.00 indicates a uniform relationship between assessed values and selling prices of properties with different prices. An index above 100.00 indicates regressivity in assessments by attributing less expensive property with a higher sales ratio than more expensive property. Letting PRD it represent the PRD score for jurisdiction i undergoing reassessment in year t, the specification of the model will try to explain variation in the PRD score using voter population characteristics (income, racial homogeneity, share of population over age 65), district characteristics (election year, population growth rate, commercial property base, property tax rate, fiscal stress), and assessment institutions (frequency of reassessments, appointed assessor dummy, availability of online tax assessment maps, contracted assessments, physical inspection dummy): PRD b SocioEcon b District it 1 it 2 it b Assess e. [1] 3 it it The definitions and sources for these variables are listed in Table 1, while their descriptive statistics are displayed in Table 2. In equation [1], a positive coefficient would indicate that the variable is directly correlated with more regressivity, while a negative sign would indicate greater progressivity. The rest of this section will describe the motivations behind these variables that might warrant their inclusion. 13 The selection of control variables is motivated primarily by the literature focusing on both horizontal and vertical equity. While vertical inequity implies horizontal inequity, the reverse is not necessarily true. Nevertheless, there is support in the literature for the notion that factors that create horizontal inequity will have a tendency to also make the process more regressive. For instance, Weber and McMillen (2010) present evidence from Cook County, Illinois, that the self-selection in the assessment appeals process results in more regressivity, and that this self-selection is due to thin markets in higher-income and racially homogenous neighborhoods that influence both the decision to appeal an assessment and successfully win that appeal. Therefore, in estimating equation [1], the expected signs for income and racial homogeneity are both positive. Similarly, if the county is p art of a Metropolitan Statistical Area (MSA), then comparable sales in nearby economically integrated jurisdictions might help offset this regressivity by providing useful information about their market value. Much of the previous literature focused on variables associated with information quality that improves accuracy, as well as variables that incentivizes an informed citizenry (e.g., 13 Unfortunately, one of the consequences of unbalanced panel data is that it undermines attempts at testing for spatial dependence, since jurisdictions are often conducting their assessments at different points in time from their neighbors. This could be an opportunity for future research in a state without this limitation.
88(1) Ross: Interjurisdictional Property Assessment 33 TABLE 1 Variable Definitions and Sources Variable Price-related differential a Number of sales a Election year Racial homogeneity c Share age 65 up c Income d Population growth c Share commercial a Property tax rate a Fiscal stress d Unemployment rate c Appointed b Reassessment frequency f Online tax maps b Contracted assessments f Physical inspection f Metropolitan statistical area Definition The mean sales ratio divided by a sales-weighted ratio, multiplied by 100, where the salesweighted ratio is the total of the assessed values divided by the total of the selling prices of all sales in the classification Number of sales in the sales ratio study with the same classification Indicates that the year of observation is an election year for that jurisdiction Author s calculation: [(%White) 2 (%Black) 2 (%Asian) 2 (%Indian) 2 (%Miscellaneous) 2 ](1/10,000) Residents over the age of 65 as a proportion of the total population The median adjusted gross income on all state tax returns in thousands of dollars for year of regression Average annual population growth from previous year The number of property sales classified as commercial divided by total number of sales of all property types Nominal property tax rate per $100 of assessed value levied by the district The residual from the regression of fiscal stress index on income and property tax rate. The fiscal stress index is intended to measure the budgetary pressure facing the district s government. Regression results available upon request. The percentage unemployment rate in the jurisdiction Indicator variable where a jurisdiction with an appointed assessor takes a value of one, else takes a value of zero The number of years between reassessments authorized by the jurisdiction. For example, a jurisdiction with annual reassessments would have a value of one. Dummy variable indicating that a database of property assessments can be found online Dummy variable takes a value of one if the assessment process is outsourced to an outside firm, and zero if it is conducted by in-house staff Dummy variable takes a value of one if the reassessment involved a physical visit to the house, and zero if it relied on computer-assisted mass appraisals. Jurisdiction is part of a metropolitan statistical area, as defined by the U.S. Office of Management and Budget. a Virginia Department of Taxation (2009). b Author s research with aid of Virginia Association of Assessing Officers jurisdiction directory (www.vaao.org). c U.S. Census Bureau. d Virginia Department of Housing and Community Development 2010. e U.S. Bureau of Labor Statistics f Knapp et al. 2010. Bowman and Mikesell 1978). As an example, senior citizens are more likely to have paid off their home and find the property tax bill to be their primary rent expense. This might increase the salience of the tax for seniors by taking it out of escrow. Senior citizens are also less likely to be earning incomes that would cause them to itemize their taxes, leaving them without the ability to deduct their property taxes from their federal taxable income. Not itemizing would give seniors a greater incentive to appeal the amount of their assessment at the margin. Finally, since the conventional political wisdom is that seniors are an important segment of voters, it is expected that their lifetime of accumulated wealth will cause them to be correlated with higher levels of regressivity. 14 14 Almost every jurisdiction in Virginia offers some kind of limited property tax exemption to senior citizens, but the way these exemptions are structured differs considerably in the necessary eligibility requirements. Though there is not data available for the years of the sample period, in 2008, just a total of 76,394 beneficiaries claimed this exemption across the state (Knapp, Shobe, and Kulp 2010), so presumably these requirements are collectively binding. Still, it should be the case that the more senior citizens a jurisdiction has, the greater the likelihood that they will be favorable targets for the assessor for underassessment. It is also worth noting that the sales ratio studies this paper s data are drawn from are used for judging assessor accuracy, so statutory exemptions are not subtracted off the assessed value prior
34 Land Economics February 2012 TABLE 2 Summary Statistics, N 243 Variable Mean Std. Dev. Min. Max. Price-related differential 103.08 4.43 94 121 Number of sales 704.21 993.89 71 6,404.00 Election year 0.25 0.43 0 1 Racial homogeneity 0.68 0.15 0.44 0.98 Share age 65 up 0.13 0.04 0.04 0.28 Income 33.01 9.50 17.49 67.28 Population growth 0.01 0.02 0.02 0.09 Share commercial 0.02 0.02 0 0.08 Property tax rate 0.80 0.27 0.29 1.43 Fiscal stress 0.42 6.58 37.72 17.49 Unemployment rate 3.71 1.21 2.10 8.60 Appointed assessor 0.53 0.50 0 1 Reassessment frequency 2.47 1.73 1 6 Online tax maps 0.82 0.39 0 1 Contracted assessments 0.39 0.49 0 1 Physical inspection 0.45 0.50 0 1 Metropolitan statistical area 0.74 0.44 0 1 As discussed in Section II, the inclusion of a population growth rate variable is motivated by the notion that high-growth areas might vary by housing price strata (e.g., Jud and Seaks 1994), in which case lags in assessment would result in more regressive inequities. This also motivates the inclusion of reassessment frequency, which was found by Bowman and Butcher (1986) to improve uniformity. Lowery (1984) found that uniformity erodes in jurisdictions under fiscal stress, and though there is no clear expectation as to how that uniformity erosion might affect regressivity, it is a reasonable explanatory variable of interest. 15 The Commission of Local Government, under the state s Department of Housing and Community Development, constructs a fiscal stress index intended to measure the to calculating the sales ratio. I thank an anonymous referee for clarifying these two points. 15 The Virginia fiscal stress measure, used in some state revenue sharing formulas, is constructed partially on the basis of income and the property tax rate. Since this makes the reported index collinear with those control variables, this paper uses the residuals from a regression of the fiscal stress index on income and property tax rate, with the interpretation that they represent fiscal stress unexplained by those two factors. fiscal condition of each jurisdiction. 16 The formula used in deriving the index is rather complex in several of its underlying components but can be generally described as a function of the jurisdiction s revenue capacity per capita, its level of effort in generating revenue from the tax base, and the median household income. Revenue capacity is derived from a series of estimates in the taxable base that exists in the jurisdiction, as well as the existing permissions granted for charging various service and license fees. The revenue effort component is intended to capture the degree to which the jurisdiction actually exercises extraction of revenue from these potential sources. In addition to including the above fiscal stress index, the unemployment rate is also included, with the intuition that areas with higher unemployment would likely be under greater fiscal stress in ways not necessarily captured by the fiscal stress index. The presence of commercial property could arguably have consequences for vertical equity in the assessment of residential pros- 16 See Appendix B in Report on the Comparative Revenue Capacity, Revenue Effort, and Fiscal Stress of Virginia s Counties and Cities, 2007/2008 (available at www. dhcd.virginia.gov/commissiononlocalgovernment/pdfs/ stress08f.pdf) for more information.
88(1) Ross: Interjurisdictional Property Assessment 35 perities, as well. Since commercial property tends to have nonresident owners, it could be that they would be targeted by assessors for tax exporting. Since Virginia assessors need to maintain a minimum median assessmentto-sale price ratio, they could complement the underassessment of residential property with the overassessment of commercial property. If they furthermore favored a particular price stratum of residential property, it would reduce vertical equity in the process. Lowery (1984) found that states with elected assessors had less uniformity in their assessment process than states with appointed assessors, but this does not lend itself to any immediate expected sign on regressivity. Rather, the sign would likely indicate whether political pressure is stronger from the lower or higher end of the price strata. Other characteristics of the jurisdiction s assessment process may be inherently prone to regressivity through error. For example, reliance on physical inspections of homes during reassessment, as opposed to computer-assisted mass appraisals, may have errors correlated by price strata. If poorer households undertake renovations that are difficult for computers to factor into the models, like remodeling of major kitchen appliances, the process may become more regressive. Similarly, the availability of an online assessment map happens to be more beneficial to particular groups in a way that could have consequences for vertical equity. One mechanism could be by encouraging owners to undertake the appeals process by aiding in the discovery that they have been overassessed relative to comparable properties. Furthermore, since the burden is on the property assessor to demonstrate that owners have been overassessed, the online tax maps might help them win their case. Ex-ante, however, there is no hypothesized reason as to why this would affect any particular group differently from another. 17 Finally, the property tax rate is included in the regression with the intuition that it represents the marginal cost of an additional dollar 17 It is also possible that these variables are actually capturing the effect of some other factor. For instance, an anonymous referee suggested that it might be capturing the effect of assessor professionalism. of assessment to the property owner. As this rate increases, it would be expected that citizens would be more active in terms of aggressively seeking a lower assessment. If this behavioral response is stronger among the owners of more expensive property, then the consequence of higher property tax rates will be greater regressivity. However, it should be noted that underassessing high-value property, as would be done in a regressive system, could result in a higher property tax rate to meet the levy requirement in the district. 18 Alternatively, overassessing a single expensive house by 1% would generate more tax revenue than overassessing a lower-value property and could allow for a lower property tax rate. These endogenous responses of the property tax rate to vertical inequity in the assessment process motivate the use of an instrument variable approach. IV. RESULTS Table 3 provides the main estimates of equation [1] using a random effects panel regression with robust standard errors clustered by jurisdiction. In all specifications, the property tax rate is instrumented with its one-year lag for reasons discussed at the end of the previous section. This instrument passed the Hausman test and carried an F-statistic between 10 and 20 in all cases. 19 All specifications omit eight observations where the PRD index was constructed using less than 70 observed sales. 20 When specifications A and B are tested against a fixed effects specification, the Hausman test cannot reject the hypothesis that the coefficients are the same at even the 18 I thank an anonymous referee for pointing out this possibility. 19 In the absence of the instrument, the coefficient on the property tax rate is positive instead of negative in all specifications, indicating that the instrument addressed a positive endogeneity bias. 20 Virginia requires a minimum sample size of six to construct sales ratio, coefficient of dispersion, and PRD scores for each jurisdiction s classification. The 70 cut-off was chosen because that is the point at which the model comparison test explained by Hausman (1978) and Cameron and Trivedi (2005) detects a statistically significant difference in the results, though the results are qualitatively the same. Results with the eight excluded observations are available upon request.
36 Land Economics February 2012 TABLE 3 Determinants of Price-Related Differential (PRD) for Virginia Cities and Counties, 2001 2007 Variable (Dependent Variable: PRD) A B C D Property tax rate a 0.134 (0.163) 0.243 (0.155) 0.082 (0.191) 0.103 (0.205) Number of sales 0.034 (0.093) 0.035 (0.092) 0.009 (0.091) 0.026 (0.090) Election year 0.082 (0.122) 0.201 (0.162) 0.071 (0.121) 0.216 (0.161) Racial homogeneity 0.041 (0.080) 0.037 (0.080) 0.099 (0.085) 0.094 (0.083) Share age 65 up 0.314** (0.128) 0.152 (0.132) 0.219* (0.118) 0.073 (0.124) Income 0.128 (0.138) 0.351** (0.159) 0.054 (0.144) 0.260* (0.158) Population growth 0.008 (0.092) 0.015 (0.093) 0.005 (0.093) 0.030 (0.094) Share commercial 0.029 (0.066) 0.023 (0.067) 0.021 (0.067) 0.005 (0.068) Fiscal stress 0.028 (0.074) 0.133 (0.098) 0.052 (0.072) 0.167* (0.096) Unemployment rate 0.047 (0.096) 0.045 (0.109) 0.029 (0.094) 0.058 (0.106) Appointed assessor 0.050 (0.277) 0.109 (0.273) Reassessment frequency 0.009 (0.094) 0.049 (0.094) Online tax maps 0.469** (0.210) 0.476** (0.205) Contracted assessments 0.136 (0.306) 0.128 (0.300) Physical inspection 0.182 (0.277) 0.135 (0.273) Metropolitan statistical area 0.317 (0.227) 0.343 (0.222) Year fixed effects included? No Yes No Yes Within-R 2 0.002 0.049 0.002 0.058 Between-R 2 0.266 0.307 0.361 0.410 Overall-R 2 0.237 0.285 0.297 0.350 Note: Sample size is 243, with 97 groups. Coefficients are standardized to represent the marginal effect of a standard deviation increase in the continuous independent variable (but a unit change for dummy variables) as a proportion of a standard deviation of the dependent variable. Robust standard errors clustered by area are in parentheses. a Endogenous variable instrumented by its lagged value. * Significant at the 10% level; ** significant at the 5% level; *** significant at the 1% level. 25% confidence level, supporting the use of random effects. Interestingly, a fixed effect specification has no statistically significant variables and carries an R 2 near zero. Similarly, Table 3 reports both the within- and between-r 2, and the models appear to explain only between variation rather than within variation. The Breusch and Pagan Lagrange multiplier test rejected the null hypothesis of no random effects with a p-value of 0.06, suggesting that the random effects model is preferred over a pooled ordinary least squares approach. Both tables report the standardized coefficients, with their corresponding standard errors in parentheses. The standardized coefficient indicates the proportion of a standard deviation of the dependent variable that changes in response to either a standard deviation increase, in the case of a continuous control variable, or a unit increase, when the control is a dummy variable. To provide greater context to the magnitudes of these coefficients, recall that Figure 1 illustrates a standard deviation increase and decrease from a uniform process. Since the random effects estimator can include time-invariant variables, specifications C and D in Table 3 introduce the five assessment policy variables, as well as the MSA fixed effect. The model continues to explain only between variation, with R 2 values of 0.361 and 0.410. Though most variables are not statistically significant, two socioeconomic variables do stand out from the rest. When year fixed effects are included, as in specifications B and D, a standard deviation increase in income is correlated with less regressivity by 0.351 and 0.260 of a standard deviation. One possible explanation for this finding is that higher-income homeowners are more likely to be tax itemizers and therefore deduct their property taxes, whereas lower-income groups pay the standard deduction. The deduction might provide less incentive for higher-income groups to appeal assessments or pressure assessors at the margin. Specifications A and C, which exclude year fixed effects, indicate that the share of the population over age 65 is correlated with greater regressivity, as a standard deviation in-
88(1) Ross: Interjurisdictional Property Assessment 37 TABLE 4 Random Effect Estimates when Interacting with Assessment Policy Variables Variable (Dependent Variable: PRD) Interaction Variable Appointed Contract Online Physical Property tax rate a 0.089 (0.254) 0.091 (0.260) 0.166 (0.230) 0.043 (0.241) Number of sales 0.075 (0.223) 0.005 (0.099) 0.835 (0.563) 0.002 (0.102) Election year 0.203 (0.251) 0.031 (0.137) 0.039 (0.328) 0.024 (0.140) Racial homogeneity 0.072 (0.104) 0.011 (0.186) 0.162 (0.152) 0.007 (0.207) Share age 65 up 0.458*** (0.154) 0.078 (0.222) 0.693* (0.395) 0.176 (0.194) Income 0.178 (0.226) 0.264 (0.212) 0.745* (0.423) 0.271 (0.211) Population growth 0.164 (0.163) 0.075 (0.119) 0.011 (0.383) 0.072 (0.121) Share commercial 0.000 (0.089) 0.037 (0.096) 0.113 (0.158) 0.123 (0.104) Fiscal stress 0.026 (0.092) 0.021 (0.117) 0.294* (0.159) 0.027 (0.128) Unemployment rate 0.084 (0.132) 0.090 (0.134) 0.484** (0.236) 0.033 (0.139) Appointed assessor 3.021 (1.839) 0.027 (0.299) 0.018 (0.283) 0.255 (0.302) Reassessment frequency 0.019 (0.107) 0.058 (0.114) 0.002 (0.099) 0.041 (0.107) Online tax maps 0.458** (0.228) 0.538** (0.239) 4.949* (2.614) 0.497** (0.224) Contracted assessments 0.016 (0.331) 2.119 (1.910) 0.220 (0.309) 0.045 (0.326) Physical inspection 0.455 (0.309) 0.191 (0.305) 0.292 (0.278) 2.643 (1.835) Metropolitan statistical area 0.379 (0.260) 0.225 (0.270) 0.487** (0.238) 0.387 (0.266) Inter Num sales 0.083 (0.244) 0.332 (0.505) 0.809 (0.568) 0.099 (0.317) Inter Elect year 0.218 (0.290) 0.055 (0.310) 0.111 (0.353) 0.226 (0.288) Inter Racial homog 0.051 (0.224) 0.057 (0.205) 0.062 (0.174) 0.089 (0.229) Inter Shr 65up 0.543** (0.261) 0.527** (0.267) 0.445 (0.410) 0.695*** (0.235) Inter Income 0.447 (0.294) 0.423 (0.306) 0.812* (0.437) 0.419 (0.290) Inter Pop grow 0.260 (0.201) 0.258 (0.210) 0.044 (0.392) 0.212 (0.197) Inter Share comm 0.091 (0.142) 0.028 (0.138) 0.126 (0.171) 0.163 (0.141) Inter Fiscal stress 0.007 (0.154) 0.017 (0.147) 0.369** (0.173) 0.054 (0.151) Inter Unemp 0.129 (0.203) 0.082 (0.202) 0.584** (0.262) 0.022 (0.197) Within-R 2 0.111 0.0677 0.057 0.109 Between-R 2 0.345 0.3653 0.423 0.378 Overall-R 2 0.316 0.3259 0.356 0.344 Note: Sample size is 243, with 97 groups. Coefficients are standardized to represent the marginal effect of a standard deviation increase in the continuous independent variable (but a unit change for dummy variables) as a proportion of a standard deviation of the dependent variable. Robust standard errors clustered by area reported in parentheses. PRD, price-related differential. a Endogenous variable instrumented by its lagged value. * Significant at the 10% level; ** significant at the 5% level; *** significant at the 1% level. crease is associated with a increase in PRD of 0.314 and 0.219 of a standard deviation, respectively. Statistically significant or not, these two variables maintain the same sign throughout all specifications in Table 3. Similarly, in specification D with both assessment policy variables and year fixed effects, fiscal stress is associated with less regressivity at the 10% level of statistical significance. Among the policy variables, making tax maps available online reduces regressivity by nearly half a standard deviation. Since both the socioeconomic and policy assessment variables primarily explain the variation between jurisdictions, it is quite possible that these variables interact through each other. For instance, appointed assessors may respond differently across demographic groups than their elected counterparts. Similarly, physical inspections might be of greater consequence in higher unemployment areas if property improvements in these areas are less likely to file for building permits. Tables 4 and 5 explore these possibilities by interacting the assessment policy variables and the socioeconomic variables, respectively. 21 21 There are two notable exceptions. Since the property tax is treated as endogenous, it cannot be interacted with other variables without an instrument for every interaction term. Also, when interacted with the other policy variables, the election year variable uniquely identifies many observations that appear once in the data. This causes either those interactions to be dropped, or the sample size to be limited by dropping those observations.
38 Land Economics February 2012 Turning attention first to the assessment policy specifications in Table 4, there does appear to be some evidence that these institutional differences matter. 22 The results in the first column demonstrate that appointed assessors assess senior citizens differently from their elected counterparts by a 0.543 margin that is statistically significant at the 5% level. However, this is an offset to the direct effect senior citizens have in increasing regressivity. In other words, a standard deviation increase in the share of the population that is over the age of 65 increases regressivity by 0.458 of a standard deviation in the PRD score when the assessor is elected, but this same increase in senior citizens is correlated with a (0.458 0.543 ) 0.085 decrease when the assessor is appointed. Similarly, though the effects are small and not statistically significant, it is interesting that appointed assessors have off-setting signs from the direct effects of election years, racial homogeneity, income, fiscal stress, and unemployment. Collectively, these results may suggest that elected assessors have a behavioral response to these factors that is not shared by their appointed counterparts. Regarding the results in Table 4 where the policy variable being interacted is the indicator for assessments being outsourced by a private firm (Contract), senior citizens are again correlated with greater regressivity, but only in the direct effect when Contract 1. The same can be said of the specification where the interaction is a physical assessment. In both the Physical and Contract specifications, the direct effect of the senior citizen population is to reduce regressivity, but interactions with these two practices increase it. It is likely the case that these two variables are partially capturing the effect of having an elected assessor to an extent not captured by the Appointed control variable, as these practices are about 20% less common among the appointed assessors. If this is the case, then it would suggest that these practices are not sufficient to offset an inclination among elected assessors to appeal to senior citizens in a man- 22 Excluded among the time invariant variables being interacted in Table 5 are the MSA indicator and frequency of reassessment. ner that results in greater regressivity. This possibility will be explored in the next table. Perhaps the most interesting policy variable in Table 4 is the indicator of having tax maps available online (Online). Just as in Table 3, it is statistically significant in all specifications, but the specification that interacts it with the other variables demonstrates a much more nuanced story. In the absence of online tax maps, senior citizens, income, fiscal stress, and unemployment are all positively correlated with greater regressivity by a margin that is statistically significant. In each of these cases, however, their interaction with the availability of online tax maps has a comparable magnitude with the opposite sign that renders their joint effect not statistically significant. This would seem to suggest that the availability of online tax maps serves as a mechanism for citizens to monitor assessor performance and offset tendencies that would otherwise result in a more regressive assessment process. This finding should be qualified by the observation that many of the interaction effects themselves are not significant, and the magnitude of the direct effect of online tax maps increases to 4.949. Moving now to interactions of the socioeconomic interaction effects in Table 5, a couple common themes reemerge. Senior citizens continue to be associated with greater regressivity in all specifications by about 20% to 25% percent of a standard deviation. When the senior citizen variable is the interaction term of interest, its effect is almost entirely offset when the assessor is appointed, as well as when the actual assessment is contracted out to a private firm. This suggests that either mechanism gets the assessment process further from elected assessors and offsets the regressive tendencies associated with the presence of senior citizens. Physical assessments, however, result in more regressivity with senior citizens by a statistically significant margin. This might suggest that the physical inspections result in discovered improvements in below average market value homes among senior citizens that are otherwise not captured by computer-assisted mass appraisals. Elsewhere in Table 5, once again the presence of online tax maps appears to play a sig-
88(1) Ross: Interjurisdictional Property Assessment 39 TABLE 5 Random Effect Estimates When Interacting with Socioeconomic Variables Variable (Dependent Variable: PRD) Fiscal Stress Racial Homog. Share 65 up Income Share Commercial Unemp. Rate Property tax rate a 0.065 (0.221) 0.036 (0.226) 0.033 (0.230) 0.021 (0.233) 0.030 (0.222) 0.041 (0.223) Number of sales 0.049 (0.092) 0.028 (0.088) 0.029 (0.092) 0.019 (0.096) 0.021 (0.093) 0.035 (0.092) Election year 0.040 (0.121) 0.087 (0.119) 0.063 (0.121) 0.072 (0.121) 0.073 (0.123) 0.073 (0.121) Racial homogeneity 0.113 (0.085) 0.415 (0.377) 0.064 (0.090) 0.113 (0.089) 0.124 (0.086) 0.101 (0.084) Share age 65 up 0.258** (0.120) 0.257** (0.114) 0.218 (0.470) 0.228* (0.126) 0.248** (0.121) 0.239** (0.120) Income 0.029 (0.153) 0.010 (0.142) 0.129 (0.167) 0.064 (0.456) 0.006 (0.148) 0.014 (0.147) Population growth 0.045 (0.097) 0.018 (0.092) 0.001 (0.096) 0.012 (0.097) 0.007 (0.095) 0.034 (0.094) Share commercial 0.001 (0.068) 0.008 (0.067) 0.029 (0.070) 0.018 (0.069) 0.166 (0.283) 0.008 (0.068) Fiscal stress 0.327 (0.339) 0.058 (0.070) 0.002 (0.072) 0.038 (0.075) 0.032 (0.075) 0.045 (0.072) Unemployment rate 0.022 (0.094) 0.041 (0.094) 0.013 (0.093) 0.053 (0.100) 0.049 (0.095) 0.709** (0.319) Appointed assessor 0.058 (0.287) 1.383 (1.535) 0.451 (0.893) 0.185 (0.979) 0.103 (0.374) 0.617 (0.738) Reassessment frequency 0.017 (0.098) 1.276*** (0.337) 0.006 (0.424) 0.140 (0.352) 0.039 (0.125) 0.385 (0.278) Online tax maps 0.428** (0.212) 1.443* (0.815) 0.163 (1.299) 0.183 (0.877) 0.533 (0.333) 0.959 (0.711) Contracted assessments 0.130 (0.309) 2.350 (1.671) 0.849 (1.255) 0.082 (1.224) 0.333 (0.456) 0.662 (0.962) Physical inspection 0.226 (0.288) 2.839** (1.435) 1.837* (1.049) 0.461 (1.043) 0.716* (0.406) 0.493 (0.874) Metropolitan statistical area 0.386* (0.233) 0.487** (0.227) 0.271 (0.239) 0.371 (0.246) 0.439* (0.239) 0.409* (0.235) Inter Appointed 0.025 (0.282) 0.305 (0.308) 0.205 (0.273) 0.059 (0.292) 0.127 (0.501) 0.187 (0.209) Inter Frequency 0.034 (0.049) 0.286*** (0.070) 0.010 (0.110) 0.059 (0.112) 0.129 (0.139) 0.117 (0.081) Inter Online 0.406** (0.162) 0.219 (0.168) 0.080 (0.340) 0.224 (0.294) 0.189 (0.409) 0.419** (0.198) Inter Contracted 0.114 (0.280) 0.466 (0.327) 0.264 (0.386) 0.008 (0.370) 0.750 (0.632) 0.183 (0.294) Inter Physical 0.249 (0.274) 0.636** (0.314) 0.644** (0.311) 0.070 (0.312) 0.961* (0.575) 0.269 (0.269) Within-R 2 0.03 0.004 0.028 0.041 0.003 0.008 Between-R 2 0.390 0.456 0.400 0.348 0.392 0.402 Overall-R 2 0.326 0.372 0.342 0.291 0.320 0.327 Note: Sample size is 243, with 97 groups. Coefficients are standardized to represent the marginal effect of a standard deviation increase in the continuous independent variable (but a unit change for dummy variables) as a proportion of a standard deviation of the dependent variable. PRD, price-related differential. a Endogenous variable instrumented by its lagged value. Robust standard errors clustered by area reported in parentheses. * Significant at the 10% level; ** significant at the 5% level; *** significant at the 1% level. nificant role in offsetting vertical inequities. In each specification, the interaction effect with online tax maps takes a sign opposite that of the direct effect. For example, the direct effect of a one standard deviation increase in the unemployment rate is a 0.709 standard deviation increase in the PRD, but the interaction effect with tax maps decreases it by
40 Land Economics February 2012 0.419, with both being statistically significant at the 5% level. The net effect of unemployment and its interaction with online tax maps (0.709 0.419 0.29) is not statistically significant at the 10% level. The specification with fiscal stress offers a similar story, as its direct effect increases regressivity but the interaction with online tax maps reduces it. In the racial homogeneity specification, the interaction with physical inspections reduces regressivity, while the interaction with frequent assessments increases it, both by statistically significant margins. Though the latter effect is difficult to explain, the former effect may be a consequence of physical inspections resulting in more regressivity as racial diversity increases (i.e., racial homogeneity declines). If racially diverse neighborhoods have improvements in low-value properties that are better captured by physical inspection, it would increase the regressivity index. Finally, physical inspections in the presence of commercial property reduce regressivity, but this result is not particularly intuitive. There is a conventional wisdom among zoning lawyers that suggests that zoning restrictions are more stringent in areas with more commercial property, which might result in more undocumented work being done without building permits. If this is true, then the physical inspections might reveal these projects in a way that reduces regressivity. 23 V. CONCLUSIONS How policy makers view the property tax as a revenue raising instrument can be influenced by the public finance profession s understanding of its incidence. Any resolution of the property tax incidence debate will rely critically on assessment incidence. Since the literature on vertical equity has been divided between finding the property assessment process to be progressive or regressive, a great deal of attention has been focused on finding alternative measures or more appropriate functional forms for regression analysis. This paper argues that one of the reasons for the 23 Assessors are generally not required to discover building code violations, and I know of no cases in Virginia where assessors are responsible for reporting such violations. mixed findings lies in this literature s reliance on parcel-level data within a single area. Rather than determining whether the assessment process is regressive or progressive in a particular area, this paper tries to advance this debate to identifying jurisdictional features that might lend themselves to a more regressive or progressive process. Using Virginia cities and counties as the unit of observation over the 2001 to 2007 period, the PRD is regressed on a variety of socioeconomic, jurisdiction, and assessment policy characteristics. The results are supportive of the hypothesis that interjurisdictional differences matter. It appears that one of the most important policy tools for improving vertical equity is making property tax maps available over the Internet. The results indicate that this is an important tool for offsetting otherwise regressive tendencies. For instance, regressivity increased with income, share of the population over age 65, fiscal stress, and unemployment in areas without online tax maps, but not in areas with them. A similar story appears to emerge when an area has appointed assessors, who appear to be less likely to engage in regressive tendencies over these groups compared to their elected counterparts. This is particularly true for senior citizens, whose presence seems to increase regressivity in areas that do not have appointed assessors or contract out to a firm. Finally, physical assessments appear to increase regressivity in the presence of senior citizens, as well as in more racially diverse areas. It is important to note that the initial critique levied by this paper, that focusing on assessments within jurisdictions might account for the difference in the findings of the existing literature, to some extent still applies to the empirical work here. Since the data is drawn exclusively from Virginia, and during a period of housing price appreciation, the results may differ if drawn from other states or time periods. This is probably a bigger concern regarding the estimates of the individual coefficients than for the main thesis of the paper, but is a concern nonetheless. Future research could try to replicate these results in different states, as well as different time periods.