An empirical analysis of the determinants of intrajurisdictional property tax payment inequities

JOURNAL

OF URBAN

ECONOMICS

An Empirical lntrajurisdictional

17, 90-102

(1985)

Analysis of the Determinants of Property Tax Payment Inequities

DONG HOON CHUN AND PETER LINNEMAN* The

Wharton

School,

of Pennsylvania,

University

Received

Philudelphia,

June I, 1983; revised

August

Pennsylvania

I91 04

29, 1983

I. INTRODUCTION The differential burden of local property tax payments has long been the focus of both popular and academic discussions. Previous empirical studies (Black [2], Cheng [3-51, Edelstein [6], Engle [7], Gloudemans [8], Kochin and Parks [12], Oldman and Aaron [14], Paglin and Fogarty [15], and Reinmuth [16]) have primarily concentrated on measuring the existence of dispersion in local property tax burdens. This study both measures the existence of inequitable intrajurisdictional property tax burdens and uses micro data to parameterize its determinants. It is found that significant systematic tax inequities exist (in 1978 Philadelphia). However, much of the variation in local tax burdens is random. Section II develops a simple empirical methodology for evaluating the determinants of local tax assessment errors. The third section uses this methodology to evaluate the sources of Philadelphia’s property tax inequities in 1978. The empirical findings are compared to several models of the assessment process in the fourth section of the paper. The paper concludes with a brief summary. II. MEASUREMENT

METHODOLOGY

The property tax payment for the i th property (q) is the product of the legally established tax rate (t) and the property’s assessed value (Ai)

T, = a;.

0)

If the assessed value is expressed as a proportion value of the property (F)

(a,) of the true market

A; = a,y, *The authors respectively.

are at the World

Bank

and the Wharton 90

0094-1190/S Copyright All rights

$3.00

d’ 1985 by Academic Press. Inc of reproduction m any form reserved.

(2) School,

University

of Pennsylvania,

PROPERTY TAX PAYMENT

INEQUITIES

91

then the effective tax rate (R,), the ratio of tax payments to true market value, is Ri G -$ = ta,. I

Thus, if (as is generally the case) the legal tax rate is invariant within a taxing jurisdiction, intrajurisdictional variations in the effective tax rate are caused by deviations of the assessment factor from the average assessment factor (a). Nonuniform assessment factors may be the result of either systematic misassessment or random assessment errors. Since the systematic bias in the assessment process may be associated with the characteristics of either the resident ( Ci) or the property (Hi), the assessment factor can be expressed as the average assessment factor, plus the bias component, plus a zero mean random error ( ei) a, = ii + f(C,, H,) + ei. Substituting

(4)

(4) into (3) yields Rj=

t[ii+f(Ci,Hi)+e,]

=tii+

tf(Cj,Hi)+ui,

(5)

where ui is a zero mean random term. In order to focus on disproportionate effective tax rates, one can define the effective tax gap (GAP) as the difference between the effective tax rate and the jurisdictions’s average effective tax rate, GAP, E R; - tZ = tf(C;y If,) + Ui = g(Ci, Hi) + ui.

(6)

If GAP, is positive the property is taxed disproportionately high, while GAP less than zero indicates a lower than average effective tax rate. Thus, the presence of intrajurisdictional inequities in effective property tax rates can be parameterized for a taxing entity by using microdata to regress the tax gap on measures of the characteristics of the residents and their residences. If measures of personal and housing characteristics prove to be insignificant regressors, one cannot reject the hypothesis that effective assessment variations are random. Alternatively, significant regressors measuring resident or residence characteristics indicate biased assessments. This methodology is not subject to the estimation bias described by Kochin and Parks [12] as it follows their procedure of using the assessment factor as a regressor. The primary distinction of the approach used here is that the assessment factor is explicitly modelled as a function of personal and residential traits via (4).

92

CHUN

AND

LINNEMAN

This specification allows a richer examination of the assessment process than simply using the assessment factor as the sole regressor. The economic loss for the ith property resulting from its being relatively overassessed is simply the assessment gap times the true property value, LOSS, = GAP; . v,.

(7)

Since the property value is a reduced form function of resident demand characteristics, the dollar loss associated with misassessments is a function of both personal and housing characteristics LOSS, = L(C,, Hi) + u,,

(8)

where ui is a mean zero random error. Thus, microdata can also be used to parameterize the wealth transfers induced by inequitable assessments. III.

DATA

DESCRIPTION

AND

EMPIRICAL

RESULTS

The 1978 Philadelphia SMSA Annual Housing Survey was used to estimate linear approximations of the tax gap (6) and resource loss (8). The sample contained 2428 owner-occupied dwelling units with complete responses to all relevant questions in the city of Philadelphia. Table 1 displays the means for a variety of personal and housing characteristics which were employed in this study. Table 2 reports the sample statistics for: property tax payments (T,), self-reported property value (y/i>,’ the effective tax rate (R ;), and the loss from overassessment (LOSS;). Since the measure of the effective tax rate employed here is the actual property tax payment divided by self-reported property value, any systematic bias in the owner’s estimate of property value relative to true property value induces bias. Since previous studies of these questions fail to ‘Since property value is reported in bracket value form the observation’s bracket midpoint was used as the estimate of property value. The brackets are: each $2,500 for property values under $20,000; each $5,000 for property values between $20,000 and $39,999; each $10,000 for property values between $40,000 and $59,999; $60,000 to $74,999; each $25.000 between $75,000 and $149,999; and over $150,000. This process induces measurement error and reduces efficiency. Further, since it is unlikely that the true property values are uniformly distributed within any bracket, this midpoint procedure may cause inconsistent estimates. However. since the bracket sizes are quite small, these problems are likely to be of little empirical importance. In fact, several alternative procedures for dealing with bracket reporting were conducted and yielded the same substantive results as the bracket midpoint procedure. ‘Kish and Lansing [ll] find that the owners of single family units on the average overestimated the value of their homes by about 4%. Kain and Quigley [lo] indicate that owners on average underestimate the value of their homes by less than 2%, and that the difference is not statistically significant. Kain and Quigley also fail to find systematic variation in the discrepancy between market value and owner-estimated values with respect to the socio-economic characteristics of the owner-occupants.

PROPERTY

TAX

PAYMENT TABLE

1978 Philadelphia Resident

93

INEQUITIES

1 Sample

Means

traits

Age of head 1 if married 1 if male head Family size 1 if head white Family income Residence

54.3 .67 .75 3.0 .79 13,853

truits 6.4 1.4 .90 .02 .ll .06 37.2 .06 .23 .76 .83 .87 .18 .26 .83 .17 .33 .91

Number of rooms Number of bathrooms 1 if single attached unit 1 if must go through bedroom to get to bathroom 1 if roof leaks 1 if cracks in walls or ceiling Age of building 1 if regular exterminator service 1 if property improved in last 12 months 1 if local school is viewed adequate 1 if local police protection is viewed adequate 1 if garbage collection publicly provided 1 if neighborhood crime is viewed bad 1 if neighborhood traffic is viewed bad 1 if local public transit is viewed adequate 1 if local streets in need of repair 1 if local nonresidential activity viewed bad 1 if local hospitals are viewed adequate

demonstrate any such bias, self-reported property value is used here as an unbiased estimate of true property value.2 However, the measurement errors associated with self-reported property values reduce estimation efficiency. In order to calculate GAPj, the sample’s average effective tax rate (tZ = 0.01625) was subtracted from each observation’s effective tax rate

TABLE Summary

Statistics

2

for Tax Variables, Mean

Tax payments (7;) Property value ( y) Effective tax rate (R,) Overassessment loss (LOSS,)

$ 435 $29,438 0.01625 $ -43.8

Philadelphia

1978

Std. deviation 244 $16,355 0.00747 $ 192.7

94

CHUN

AND

LINNEMAN

(Ri). The distribution of GAP in Philadelphia was nonsymmetric and highly skewed to the right. Further, 59% of the sample was relatively underassessed. This nonsymmetric distribution is reflected in the finding that there is an average gain from the assessment process of $44. Alternatively stated, if all residents had been taxed at the average tax rate, about $44 more per household would have been collected in taxes. Table 3 reports the mean relative overassessment rate for various socioeconomic groups. A positive tax gap reflects relative overassessment while a negative GAP indicates relative underassessment. Perhaps the most notable finding with respect to the tax gap are the substantial differences associated with race and income (for whites). On average, white households are substantially underassessed (relative to nonwhites), with the degree of underassessment being the greatest among those with the highest incomes.

TABLE Mean

Differential

Tax Burdens

3 for Socioeconomic

Groups GAP

- .oooo8

Male Head Female head White head Nonwhite head Married Unmarried Head age 20-29 Head age 30-39 Head age 40-49 Head age 50-64 Head age over 64 l-2 family size 3 -4 family size 5-6 family size Over 6 family size Nonwhite

White

income:

income:

.00026 - .00054 .00201 - .00019 .00040 - sJOO3o - .00022 - .00017 .00019 .oOOO7 .OOOOl - .OOOlB - .OOOll 00202 O-$9,999 1o,ooo-14,999 15,ooo-19,999 20,000-24,999 25,000-29,999 30,000 and over O-$9,999 10,000-14,999 15,000-19,999 20,000-24,999 25,000-29,999 30,000 and over

00173 .00230 .00303 .00128 .00149 .00258 -

.00018 00066 00072 .00078 .00069 .00109

PROPERTY TAX

PAYMENT

95

INEQUITIES

TABLE 4 Differential Tax Burdens by Property Value Categories Property values

GAP

o-9,999 lO,OOO-12,499 12,500-14,999 15,00&17,499 17,500-19,999 20,000-24,999 25,000-29,999 30,00&34,999 35,000-39,999 40,00&49,999 50,000-59,999 60,00&74,999 75,00&99,999 lOO,OO@-

.01084 .00542 .00285 .00133 .00019 .OOlll .00217 .00236 .00115 .00191 .00452 .00367 .00396 .00617

-

LOSS ($)

Percent of sub-group which is overassessed

95 61 39 22 4 -25 -60 -17 -43 -86 - 249 - 248 - 347 - 694

82 74 58 58 44 39 30 23 33 32 11 16 20 9

Table 4 displays the average GAP, LOSS, and the percent of the category which is overassessed for various property values. These results indicate that the proportion of properties which are overassessed declines with property value. Similarly, overassessment is most prevalent among those with the lowest property values, and least likely for those with high property values. Finally, the average loss associated with unequal assessment is positive up to property values of $20,000 and declines monotonically with property value. Another perspective of the nonneutrality of the assessment process is provided in Table 5 which reports the average socioeconomic characteristics of the overassessed and underassessed populations. Both of these subsamTABLE 5 Characteristics of Over- Versus Underassessed Populations

Mean age of head Percent married Percent male headed Mean family size Percent white Mean family income Mean property value Mean property taxes Mean LOSS Number of observations

Relatively overassessedpopulation

Relatively underassessedpopulation

54.5 66.1 74.2 3.0 71.3 13,198 23,412 493 113 988

54.2 68.2 75.3 3.0 84.1 14,301 33,574 395 - 151 1440

96

CHUN

AND

LINNEMAN

ples have the same average head age and family size. However, the overassessed population contains a disproportionately large number of: nonwhites, unmarried households, female-headed households, and families with relatively low incomes. It is also noteworthy that the average loss from overassessment was approximately 1% of income for the overassessed, while the underassessed population realized an average gain slightly in excess of 1% of their income. Taken together, the summary statistics on assessment practices in Philadelphia (in 1978) indicate that property tax assessments were not neutral with respect to socioeconomic characteristics. This view is reinforced by the regression estimates of Eqs. (6) and (8) which are displayed in Table 6. Specifically, both the tax gap and overassessment loss are systematically related to a vector of personal traits3 and a vector of residence traits. However, it is noteworthy that much of the variation in the tax gap is random.4 Married households are significantly underassessed (at standard confidence levels) and realize gains of approximately $25 annually from overassessment. Surprisingly, male-headed families are overassessed and lose $30 a year from these inequitable assessments. Since married households are generally categorized as male headed, these results indicate that the typical married household is basically fairly assessed.5 However, families headed by unmarried males fare significantly better than their female and married counterparts. No significant age of the head effect is found with respect to either GAP or LOSS. However, large families are found to be significantly overassessed. In fact, each additional child induces an overassessment loss of $9 annually. Perhaps the most dramatic finding in this study is that nonwhite households are significantly overassessed, particularly when compared to relatively high income whites. Specifically, white families gain a base amount of almost $18 a year from inequitable assessments (compared to nonwhites). This gain increases by over $2 annually per $1000 in family income.6 This interprets into a $48 (or 11% of annual tax payments) annual differential gain for whites when evaluated at the sample’s mean income. ‘A set of legal relief variables which reflected a Pennsylvania state law which provides state reimbursement for those families with heads over 64 and incomes under $9,000 was also included in the regression. Complete results are available upon request. ‘Measurement error in the property value estimator makes it impossible to interpret 1 - R2 as the % of assessment variation which is random. However, it seems likely that at least half of all assessment variations are random with respect to resident and residential characteristics. ‘The point estimates for the typical married household are - $.039 for GAP and $4.52 for LOSS. These estimates are not significantly different from zero as standard confidence levels. ‘This reflects the relatively imprecise $1.04 (per thousand) loss for nonwhites plus the $1.15 (per thousand) gain for whites.

PROPERTY

TAX

PAYMENT

97

INEQUITIES

A number of the residence characteristics are significant determinants of both the relative assessment rate and the assessment loss. An analysis of variance indicated that the vector of residence traits accounts for approximately two-thirds of the “explained” variances of both the overassessment rate and dollar loss. The source of these significant results for the residence characteristics is explored more thoroughly in the next section.

TABLE Regression

Results

6

for Philadelphia GAP

Age of head

1.3 x 10-s (0.96)

1 if married

-0.137 (2.35)

1978 LOSS 0.15 (0.42) - 25.42 (1.70)

1 if male head

0.098 (1.65)

29.99 (1.95)

Family

0.029 (2.55)

8.72 (3.01)

- 0.090 (1.52)

- 17.66 (1.16)

size

1 if white

head

Family

income

for blacks

3.7 x 10-6 (1.19)

1.04 x 10-s (1.31)

Family

income

for whites

-3.8 x 1O-6 (2.37)

-1.15 x 10-s (2.69)

Number

of rooms

0.028 (2.05)

1.41 (0.41)

Number

of bathrooms

0.029 (0.98)

- 1.45 (0.19)

0.113 (2.11)

65.47 (4.76)

- 0.226 (2.03)

- 28.85 (1.01)

Age of building

- 0.003 (2.74)

- 1.75 (5.81)

1 if roof leaks

0.092 (1.89)

7.47 (0.60)

1 if cracks

0.194 (2.91)

40.20 (2.34)

0.129 (1.90)

19.57 (1.12)

1 if single attached 1 if through

1 if regular

unit

bedroom

to bathroom

in walls or ceiling exterminator

service

98

CHUN AND LINNEMAN TABLE 6-( confirmed) GAP 1 if property

improved

in last 12 months

- 0.047 (1.31)

- 27.98 (3.01)

- 0.141 (3.69)

- 23.07 (2.35)

- 0.031 (0.73)

- 6.63 (0.61)

- 0.032 (0.65)

5.92 (0.47)

is bad

- 0.070 (1.99)

- 16.25 (1.80)

is adequate

0.082 (1.95)

27.11 (2.50)

1 if local school is adequate

1 if local police protection

1 if garbage

collection

1 if neighborhood

1 if local public

is adequate

publicly

traffic

transit

LOSS

provided

1 if local streets need repair

1 if local nonresidential

activity

is bad

-0.114 (2.79)

- 15.45 (1.47)

- 0.107 (3.29)

- 35.91 (4.2X)

1 if neighborhood

crime is bad

0.077 (1.92)

17.48 (1.70)

1 if local hospitals

are adequate

- 0.174 (3.18)

-41.18 (2.92)

0.058 (0.37)

- 7.54 (0.19)

constant

R2

0.071 Note.

Absolute

r-values

0.070

in parentheses.

In sum, Philadelphia’s assessment practices in 1978 appear to have created considerable variation in effective tax rates. While much of this variation was found to be random, characteristics of both the resident and their residence were found to be significantly related to effective tax rates. Finally, the assessment outcome disproportionately burdened nonwhites, large families, and low income families. In the next section these results are crudely evaluated in the context of a variety of simple models of assessment practices.

PROPERTY TAX PAYMENT

IV. ALTERNATIVE

INEQUITIES

99

MODELS OF THE ASSESSMENT PROCESS

A variety of alternative descriptions of the assessment process have been offered in the past. In this section some simple empirical implications of the four most popular descriptions are evaluated in terms of the findings presented in the previous section. One popular version of the assessment process is that properties are reassessed only infrequently and therefore properties with growing property values tend to be relatively underassessed. A common extension of this description argues that properties with desirable housing traits are appreciating at relatively high rates and therefore the effective tax gap should be negatively related to desirable housing traits and positively related to undesirable housing traits. Table 6 reveals that 5 of the 8 desirable traits and 5 of the 10 undesirable traits exert negative tax gap effects.7 Since only 10 of the 18 housing traits display the predicted sign, the data seem to be inconsistent with this simple version of the infrequent assessment model.8 A second possible description of the local assessment process is that the assessors do the best job they can to fairly assess the true market value of each property. However, it is impossible for them to fully evaluate all aspects of the property in arriving at their assessment, Therefore, the assessors make “honest errors” in their assessments that cause properties with difficult to observe but relatively desirable (undesirable) housing traits to be relatively underassessed (overassessed). This “honest error” hypothesis implies that all personal characteristics and easy to observe housing characteristics are insignificant explanatory variables in the tax gap regression. As was noted in the last section the vector of personal characteristics was significant in the tax gap regression. Further, many easy to observe housing traits (e.g., the number of rooms) were also found to be significant regressors in the tax gap equation. However, four of the five hard-to-observe housing traits exhibit tax gap effects which are consistent with the “honest error” hypothesis.’ Taken together these results suggest that some, but not all, of the systematic variation in effective tax rates is associated with honest assessment errors. A third simple alternative description of the source of inequities in the assessment process is that the system is biased not because of assessment ‘The desirable housing traits, based on hedonic price studies (see, e.g., Linneman [13]) are: rooms, bathrooms, property improvements, adequate schools, police protection, transit, local hospitals, and public garbage collection. ‘An alternative version of this model would suggest that property values regress towards the mean and, therefore, bad housing traits imply property value growth. This version can also be rejected. ‘The hard to observe housing traits are: roof leakage, exterminator service, property improvement, privacy of bedroom, and cracks.

100

CHUN

AND

LINNEh4AN

bias, but rather because*;of selection bias in the assessment appellate process. A simple example vividly illustrates this argument. Imagine that the assessor’s office assessesall properties in an unbiased but imperfect manner. This process would result in totally random variations in tax gaps after initial assessment. However, these assessments may be appealed through judicial review. Since only owners of relatively overassessed properties have an incentive to appeal in hopes of reducing their tax burden, nonrandom final assessments can result from judicial review. Whether overassessed owners successfully appeal their assessment depends upon: (1) the extent of initial overassessment and (2) the net benefit of appealing. The extent of initial overassessment should not induce biased final tax burdens because the original assessment error was random. However, the net benefit of appealing one’s overassessment rises with the value of the property. If all residents face the same costs of appealing their assessment, the tax saving from redressing a randomly high assessment is greatest for those with high property values. For example, the benefits of eliminating a 10% overassessment are greater for the owner of a $100,000 property than for one with a $10,000 property. Thus, successful appeals will be disproportionately made by owners of expensive properties. This selection bias in the legal redress procedure could induce systematic errors in the assessment process even though all properties were initially randomly assessed. This bias will be reflected through a variety of personal and residential traits due to the correlation of these traits and property values. This model suggests that traits which are positively (negatively) related to property values should be negatively (positively) related to the tax gap. An examination of Table 6 reveals that only 10 of the 18 residential characteristics display the expected signs. Further, among the personal characteristics, only marital status and income for whites display GAP effects which are consistent with the legal redress hypothesis. These results indicate that the legal redress model of the assessment process fails to adequately explain observed assessment patterns. A final model argues that local tax inequities reflect the differential political influence of various socioeconomic groups. For example, high income residents tend to be more politically influential then those with relatively low incomes because their campaign contributions make them extraordinarily important constituents. In order to repay these high income constituents for their political support, local politicians create a local tax system (the assessment process) which is biased in favor of high income residents. Similar political economy arguments can be made with respect to other socioeconomic characteristics in terms of assessment inequities. This political economy hypothesis of assessment errors implies that personal characteristics associated with major political influence should be negatively related to the dollar loss associated with assessment errors. An evaluation of

PROPERTY

TAX

PAYMENT

INEQUITIES

101

Table 6 reveals that, consistent with the political economy hypothesis, high income white constituents realize the largest gains from the local assessment process. Similarly, married and modest households are able to translate their differentially high voting propensities into small gains from the assessment process. However, the finding that male headed households bear a disproportionately large tax burden does not appear to be consistent with standard descriptions of the political process. V. SUMMARY This paper developed an empirical framework for evaluating inequities in the local tax assessment process which utilizes microdata. The framework models variations in effective local tax rates into a random component and a systematic component which is a function of both resident and residential characteristics. The framework was empirically implemented using a sample of Philadelphia homeowners for 1978. The data clearly indicate that considerable differences in effective tax rates existed. Further, these effective tax rate inequities were not neutral with respect to the socioeconomic characteristics of the taxpayer. In particular, nonwhites were found to bear disproportionately large tax burdens while high income whites fair extremely well. The result that both socioeconomic and residential characteristics of the owner were significant determinants of the tax gap was evaluated in the context of several simple descriptions of the assessment process. It is noteworthy that a large amount of the variation in local tax burdens was purely random or the result of “honest errors” in the assessment process. Some evidence, particularly the effect of income for whites, was discussed which is consistent with the view that local tax burdens reflect the outcome of economically induced local political influence. REFERENCES 1. H. Aaron, “Who Pays the Property Tax: A New View,” Brookings, Washington D.C. (1975). 2. D. E. Black, The nature and extent of effective property tax rate variation within the city of Boston, Nur. Tax J., 25, 203-210 (1972). 3. P. L. Cheng, The common level of assessment in property taxation, Nut. Tux J., 23, No. 1, 50-65 (1970). 4. P. L. Cheng, Statistical control of assessment uniformity, Manage. Sci., 16, 638-655 (1970). 5. P. L. Cheng, Bias and error detection in property tax administration, Munuge. Sci., 22, No. 11, 1251-1257 (1976). 6. R. H. Edelstein, Regressivity and the inequity of the residential property tax: The Philadelphia story, in “Research in Urban Economics,” Vol. 1. pp. 219-247, Jai Press (1981). 7. R. F. Engle, De facto discrimination in residential assessments: Boston, Nat. Tax J., 48, 445-451 (1965).

CHUN AND LINNEMAN

102

8. R. S. Gloudemans, Nonparametric statistics and the measurement of assessment performance, in “Analysing Assessment Equity,” pp. 79-99, Chicago (1977). 9. R. P. Inman and D. L. Rubinfeld, The judicial pursuit of local fiscal equity, Hurvurd Law J., 1661-1750 (June 1979). 10. J. F. Kain and J. M. Quigley, Note on owner’s estimate of housing value, J. Amer. Sfotisr. Assoc., 67, 803-806 (December 1972). 11. L. Kish and J. B. Lansing, Response errors in estimating the value of homes, J. Amer. Statist. Assoc., 49, 520-538 (1954). 12. L. A. Kochin and R. W. Parks, Vertical equity in real estate assessment: A fair appraisal, University of Washington mimeo, (1980). 13. P. Linneman, Some empirical results on the nature of the hedonic price function for the urban housing market, J. Urban Econ., 8, 47-68 (1980). 14. 0. Oldman and H. Aaron, Assessment-sales ratios under the Boston property tax, Nut. Tux J., 18, 36-49

(1965).

15. M. Paghn and M. Fogarty, Equity and the property tax: A new conceptual focus, Nut. Tax J., 25, No. 4, 557-565 (1972). 16. J. E. Reinmuth, The measurement of vertical inequitites in assessment practice, in “Analysing Assesssment Equity,” pp. 47-60, Intl. Assess., Chicago (1977). 17. J. E. Reinmuth, Recent advances in sales-ratio analysis, Assessors’ J., 101-117 (July 1976). 18. C. M. Tiebout, A pure theory of local expenditures, J. PO/it. Econ., 64, 416-424 (1956).