Public school reform, expectations, and capitalization: what signals quality to homebuyers?
Zahirovic-Herbert, Velma ; Turnbull, Geoffrey K.
1. Introduction
Education is one of the most important services provided by local
governments. Low test scores, high dropout rates, high teacher turnover
rates, and other problems indicate that large urban school districts in
the United States serve their students inadequately. These deficiencies
provide incentives for affluent families to leave cities for the suburbs
or to move their children to private schools. When these families move,
urban tax bases and economic activity are reduced. When good students
move to private schools, the average academic quality of the remaining
public school students declines, which can reduce the quality of the
education received in the public schools through influence on peer group
effects and declining parental involvement and political support. For
these reasons, education reform remains a key concern in urban areas.
This paper draws on local education reforms that occurred in East
Baton Rouge Parish, Louisiana, in 1996 and 2001 to examine several
questions left unresolved in the capitalization literature, especially
the literature concerned with how housing markets respond to different
types of government-supplied information about school quality. Two
policy events in the school district provide a rare opportunity in this
regard; changes ordered by the federal court supervising the school
district present natural experiments for observing the extent to which
school characteristics are reflected in the housing market. More
importantly for our purposes, the setting and the two policy events
themselves highlight the roles of home owners' expectations on
measured capitalization.
As a legacy of a 40-year desegregation lawsuit, the local school
system was under direct federal court control for more than 15 years. In
the years preceding our first policy event, students were randomly
assigned to individual schools in an effort to equalize racial
composition, a method that eliminated school quality as a
location-specific house attribute (e.g., it was not unusual for
different children in the same family to attend different schools). (1)
In a surprise move in the summer of 1996, the presiding judge ordered
the elimination of random school assignment in favor of stable
attendance zones, thereby creating a direct tie between house location
and school quality. This change from random to zone school assignment
represents a natural experiment and an opportunity to observe how the
local housing market values the policy change. Further, in light of the
history of the 40-year federal lawsuit and 15 years of direct court
control of the local school system, the public had little confidence in
the federal court's ability to improve education quality. The
overwhelming defeat of a major school tax referendum following the
court's creation of school attendance zones is evidence of this
lack of confidence. This is precisely the type of environment in which
we expect to find little or no systematic capitalization of measured
school quality differences.
There is evidence that public confidence in the school system
improved during the next few years. Enrollments ceased their inexorable
declines and stabilized within a few years and, more importantly, voters
passed a major school tax referendum at that time as well. Both
observations signal an important reversal in public opinion regarding
the perceived viability of school reform. The second policy event in our
study, major changes in school attendance zones in 2001, occurs after
these observations. Thus, we argue that the second policy event, which
changed the previously established attendance zones, occurs in an
environment of fundamentally different expectations than the first
policy event, which established those zones.
In the second policy event, the redistricting affected the housing
market both directly and indirectly. Many houses were assigned to new
schools, changing their locational attributes directly. At the same
time, even for those houses experiencing no change in school assignment,
changes in attendance zone boundaries in other neighborhoods induced
large changes in the characteristics of the students assigned to their
schools, introducing indirect changes in school characteristics. This
second event, therefore, also allows us to measure the extent to which
families perceive different benefits from improving their
children's current school versus sending their children to a
different but better school.
Finally, we also pursue an alternative direct test of the
expectations-capitalization nexus. We examine Fischel's (2001)
homevoter hypothesis--the notion that home owners vote for local
policies that increase their property values and oppose those that do
not. We draw on precinct-level voting outcomes on the second major
school tax referendum to measure capitalization differences across
precincts.
The unique setting from which the data are drawn allows us to
sidestep common difficulties encountered in other school quality
capitalization studies. In particular, the single school district is
coterminous with the unified city-parish government jurisdiction
boundaries, a unique feature that minimizes spatial variation in local
property tax rates and school spending, as well as other public
services. (2)
There is a decade and a half trend toward increasing school
accountability as a part of education reform in the United States. There
is, however, no consensus about how to make schools accountable in terms
most useful to residents. The economic literature on how residents
interpret school letter-grade or qualitative performance rankings yields
mixed results (Figlio and Lucas 2002; Kane, Staiger, and Samms 2003).
(3) Student test scores provide an alternative measure of school
quality; Haurin and Brasington (1996) and Black (1999), for example,
find positive relationships between standardized test scores and house
prices. In a different vein, some education and labor economists suggest
that school achievement might not be the proper measure of school
quality. Instead, they suggest focusing on the growth in achievement
scores over time as a measure of schools' value added (Hanushek and
Taylor 1990; Hayes and Taylor 1996; Figlio 1999; Dowries and Zabel 2002;
Brasington and Haurin 2006). Simply put, there is no broad agreement
about the best way to measure school quality.
Yet in addition to test scores, parents care about peer effects and
the environment in which their children are learning, including the
socio-economic and demographic composition of the student body. While
Hoxby (2000) and Hanushek, Kain, and Rivkin (2002) examine peer effects
and their relation to school performance, few recent studies consider
direct student peer effects as measured by the socio-economic
characteristics of the students in the school and their impact on house
values. There is, however, evidence that these factors can matter.
Weimer and Wolkoff (2001), for example, show that ignoring the
percentage of an elementary school's student body that receives
reduced-price lunch results in substantially larger house price
capitalization estimates for elementary test scores.
Our results add to this body of evidence. The difference between
capitalization effects of the first policy event, which established
attendance zones, and the second policy event, which changed the zones
five years later, reveal the strength of community expectations
mediating capitalization of reported school quality differences. In
particular, once the population had experienced stable school attendance
zones, as well as a degree of local school board control over education
management during the years 1996-2001, the housing market responded to
the attendance zone changes in 2001. While test scores are not
systematically capitalized into house prices, variables that capture
broader improvement or value-added, such as an improvement in
categorical ranking of school performance, significantly increase house
prices. We also find a direct tie between expectations and measured
capitalization in the 1998 school tax referendum results, which is
consistent with Fischel's (2001) homevoter hypothesis model of
capitalization. We find that precincts supporting the tax referendum
exhibit even stronger capitalization of school improvement than the
pooled sample. At the same time, precincts opposing the referendum show
no statistically significant capitalization of improved school rankings,
whether the improved ranking is from better performance of a given
school or from reassignment to a better school.
The rest of the paper is organized as follows. Section 2 summarizes
background information on the education reform experience in the East
Baton Rouge Parish school system being studied. Section 3 describes the
data and the empirical model. Section 4 exploits the initial
establishment of attendance zones in 1996 to examine the extent to which
the housing market values student test scores and peer composition.
Section 5 draws from the changes in attendance zones in 2001 to
investigate whether the market values direct improvements in the quality
of a given school versus being assigned to a better school. Section 6
examines the expectations-capitalization nexus more closely, studying
differences in capitalization of school quality ratings and peer group
composition across precincts supporting and opposing an earlier school
tax referendum. The final section concludes the paper.
2. Historical Background
The East Baton Rouge Parish School System serves the greater Baton
Rouge (Louisiana) metropolitan area. It is the third largest district in
the state and among the top 75 nationally in terms of student
enrollment. It is composed of 88 schools and approximately 45,000
students. The school system has gone through many changes because of its
battle with a 40-year school desegregation law suit, Davis et al. v.
East Baton Rouge Parish School Board (1956). In 1981, the federal court
instituted a plan that closed schools and imposed widespread busing, a
scheme that evolved into random school assignments. (4) In a sudden
reversal of policy, the court ordered the establishment of
community-sensitive school attendance zones in late summer of 1996. The
residents resoundingly defeated a $2 billion tax proposal for new school
construction shortly thereafter in 1997. In 1998, however, they passed a
trimmed-down $280 million tax plan; it is likely that both the more
modest size of the proposal and a growing sense of confidence in the
school system contributed to the passing of the referendum. In 2001, the
federal court ordered attendance zone changes that affected more than
2000 students. The desegregation case was finally settled in 2003.
The federal court's desegregation order in 1981 provoked
strong public resistance and an immediate withdrawal of many white
students from the public school system; the system immediately lost
about 8000 students. The flight to private schools reflected a broad
lack of public confidence in the court-run public system. The black
percentage of the student body jumped from 41% in 1980 to 44% in 1981.
By 2000, almost 70% of the students in the public school system were
black. In contrast, the percentage of white students in private schools
went from 20% in 1980 to 25% in 1981 and 48% by 1998. (5)
These changes paralleled the pattern of decline in public school
achievement test scores. The difference in performance between black and
white students in the district was 47.4 points in 2003, with whites
having an average performance score of 109 points and blacks having an
average performance score of 61.6 points (out of 150). The parish also
had a considerable poverty achievement gap, slightly over 40 points,
measured by the difference in performance between students who pay for
their lunch and those who receive free or reduced-price lunches.
Louisiana's education accountability system holds schools and
districts accountable for raising student achievement. As part of the
program, since 1999, each school has received an annual School
Performance Score (SPS). The SPS is a weighted composite index using
results from statewide testing programs: 60% weight for the LEAP 21
tests (a state standardized test), 30% weight for the Iowa Tests, and
10% weight for an attendance and dropout index. The state assigns School
Performance Labels based on this score; Table 1 lists the six
performance categories.
The distribution of the school performance labels clearly indicates
that the district was hampered by low-performing schools. In 2003, there
were no schools of "Academic Excellence," and only 1% were
"Schools of Academic Distinction." Fifty-seven percent of the
district elementary schools were classified as "Academically Below
the State Average." (6) The state also uses a two-year
accountability cycle, during which each school receives an SPS and a
Growth SPS, calculated at the end of a cycle and used to determine if a
school has achieved its growth target. According to the state, the
average SPS for elementary schools in East Baton Rouge Parish increased
five points from 1999 to 2004.
3. Data and Empirical Methodology
The hedonic price for detached single family houses is a function
of the vectors of physical characteristics of the house, H; neighborhood
characteristics, N; localized market conditions, M; school
characteristics, S; and fixed effects for geographic location and year,
month, and season of sale, F; or
ln Price = c + [alpha]H + [beta]N + [delta]M + [gamma]S + [phi]F +
[epsilon], (1)
where c is the regression constant and [epsilon] the error. The
house transactions data draw from the Multiple Listing Service (MLS)
sales reports for Baton Rouge, Louisiana, for transactions from 1994
through 2002. (7) Each house is geocoded to a specific elementary school
and census tract. There are 49 attendance zones in our sample. Their
configuration resembles patterns typically found in urban areas; the
attendance zones range from a radius of a few blocks in the older
inner-city neighborhoods to much larger areas in suburban neighborhoods
in the Parish. The boundaries are highly irregular as a rule, and the
variation in geographic size reflects both student population density
and school size.
Selling price is in real 1999 dollars (1999$). (8) The house
characteristics, H, include standard features, such as number of
bedrooms and bathrooms, age, living area, and net area. Age is measured
using a set of dummy variables reflecting the reported age range
categories; we fill missing observations using the difference between
the reported date of construction and the date of sale when both are
available. Living Area and Net Area are measured in thousands of square
feet (Net Area = Total area under the roof - Living Area). Neighborhood
characteristics, N, are the 89 census tract observations for median
income, proportion of population that is black, proportion of houses
that are owner occupied, and proportion of school-age children enrolled
in private schools.
Location is indicated by a set of dummy variables that control for
543 individual subdivisions. We also estimate the model using the set of
dummy variables controlling for 12 MLS areas. The number of transactions
varies across these areas, from 30% of sales (Area 43) to 1% (Area 60),
with more than half of the areas each accounting for 6-12% of sales.
Fixed effects for year, month, and season are obtained using
appropriately defined sets of dummy variables.
Neighborhood housing market conditions, M, are measured in part by
the number of competing houses that are for sale at the same time this
house is on the market. The rationale for including neighborhood market
condition variables in the hedonic model is very simple (Turnbull and
Dombrow 2006): The number of houses for sale in a small neighborhood
surrounding a particular house can have localized effects on the
distribution of prospective buyers and sellers, which is the rationale
typically used to justify spatial interdependencies in sales prices. A
greater number of houses for sale increases the competition among
sellers for buyers considering houses in the neighborhood the localized
competition effect. Similarly, a greater number of houses for sale may
draw more prospective buyers to the neighborhood, potentially increasing
the chance of matching a particular house with a buyer--the shopping
externality effect.
Following Turnbull and Dombrow (2006), neighborhood market
conditions are measured by the average number of competing listings in
the neighborhood each day the house is on the market, Listing Density.
This measure for each house i is calculated as follows:
Listing Density = [summation] [(1 - D(i, j)).sup.2]0(i, j) / s(i) -
l(i) + 1,
where the summation is taken over all houses 20% larger or smaller
(in terms of living area) that are within 1 mile of house i. Here, l(i)
and s(i) are the listing date and sales date for house i, respectively,
so that time-on-market is s(i) - l(i) + 1. O(i, j) represents the
overlapping marketing duration for contemporaneously listed houses i and
j and is defined as O(i, j) = min[s(i), s(j)] - max[l(i), l(j)] + 1.
D(i, j) is the distance in miles between houses i and j. The calculation
of this variable for each house in the data set includes all applicable
competing house sales, including houses in areas geographically
neighboring our sample as well as any house listed before our sample
period with a time-on-market that overlaps with our sample period. We
also include the variables New Density and Vacant Density, which are
constructed similar to Listing Density but are based on neighboring
newly listed (14 days or less) and vacant houses, respectively.
Finally, the vector of school characteristics, S, includes summary
measures of student test performance (in the first sample, 1994-1998) or
school performance score (in the second sample, 1999 2002), as well as
peer attributes, percentage of the student body that is black, and
percentage taking free or subsidized lunches.
Fischel (2001) argues that partial capitalization of school quality
or other local goods can usually be explained by two factors:
buyers' expectations of low quality and data or econometric
limitations. While there has been little empirical work directly
addressing expectation effects in the school quality literature, the
econometric limitations of such studies have been discussed at length.
Even with the most accurate measure of school quality, it is well known
that a reliable estimate of the value of a school cannot be separated
from other location attributes unless the latter are adequately
controlled in the model. Unfortunately, most are unobservable and others
are difficult to quantify. Most studies use census tracts to provide
neighborhood demographics, and while there is a lot of demographic
information by tracts, they are relatively large geographic areas.
Black (1999) finds that inadequate neighborhood controls inflate
the school quality capitalization estimates because better public
schools tend to be located in better neighborhoods. And when researchers
use data composed of more than one school district, the estimated
differences in house values represent the combined effect of differences
in school quality and taxes. Like us, Black uses data for houses within
the same community but in different school attendance zones. Her method
focuses solely on houses on different sides of elementary school
attendance boundary lines within the same district. Thus, homes
presumably have the same neighborhood effect, and the only difference
between the homes is the elementary school that children attend. She
finds that the coefficient on the test scores decreases by half because
of the inclusion of neighborhood effects, as captured by the boundary
indicators.
In a different vein, Lacombe (2004) and Brasington and Haurin
(2006) argue that spatial correlation in housing market data can affect
capitalization estimates (although Haurin and Brasington [2007] find
that their estimates are not affected by the presence of spatial
correlation in the data). These studies argue that spatial correlation
arises from neglected neighborhood market conditions. We follow Turnbull
and Dombrow (2006) and include neighborhood market conditions variables
in the model to directly control for these effects.
4. The Initial Creation of Attendance Zones
The first part of our study focuses on the initial establishment of
neighborhood schools, changing from randomized to more traditional
attendance zones in 1996. To examine the consequences of this change, we
use the percentage of students at the proficiency level on standardized
tests, the percentage of students qualifying for the free-lunch program,
and the school racial composition as the school attribute vector, S. The
student test data are from the State of Louisiana Progress Profiles.
These variables take on a value of zero for transactions completed prior
to September 1996 because the parish was under random school assignments
and students were not assigned to elementary schools based on their
residence location (the final attendance zones were announced to the
public in August 1996).
Table 2 reports summary statistics over the years 1994-1998 of the
variables that enter the empirical models. The dependent variable, house
sales price, is adjusted for inflation and the mean of $123,597 in 1999
dollars. House characteristics include Bedrooms, number of bedrooms
(3.27); Full Baths, number of full bathrooms (2.03); Living Area, living
area in thousand [feet.sup.2] (1.90); and Net Area (.69). School
attributes are the percent of students passing on standardized tests,
Test (mean of 89.62%, standard deviation of 6.36%); the percent of
blacks, non-Hispanic, Black School (mean of 51.04%, standard deviation
of 25.0%); and the percent of students qualifying for free or subsidized
lunches, Free Lunch (mean of 49.02%, standard deviation of 21.60%). Even
after decades of court-ordered desegregation, the percentage of black
students in schools ranges from 7.6% to 100%. The range of free-lunch
students is similar to the percentage of black students, with a minimum
of 6.6% and a maximum of 94.4%.
The variables used to describe neighborhood characteristics include
median household income in thousands of 1999 dollars (mean of $50.05);
percent black (mean of 22.48%); and percent of owner occupied housing
units (mean of 66.86%). In addition, the average percent enrolled in
private schools in the census tract is 5.2%, with standard deviation of
8.3%. (9)
Table 3 reports estimates of models using the year-specific
school-level attributes, such as school performance score, percent
black, and percent receiving free lunch, as our measures of school
attributes. The results show that housing characteristics enter the
price equation with expected signs. Looking at the effects of school
performance, however, we find no capitalization of performance scores
using any of the specifications. The value of peer composition varies
depending upon the estimation method used. Model 1 estimates the
baseline hedonic model by extending the across-the-street estimation
approach, first applied by Gill (1983) and Cushing (1984) and later
popularized by Black (1999), to correct for omitted variable bias. This
model includes a set of dummy variables for attendance zone boundaries,
constructed such that the dummy variable pertaining to the nearest
boundary to house i takes a value of 1, and all others take a value of 0
for that house. Following the methods of Black (1999) and Kane, Staiger,
and Samms (2003), only houses within 0.3 miles of a boundary are
included in the sample. Introducing these attendance zone boundary dummy
variables into the model is equivalent to calculating differences in
house prices on opposite sides of attendance boundaries, while
accounting for house characteristics and relating the differences in
prices to school quality information. In this approach, the boundary
dummies allow us to account for any unobserved neighborhood
characteristics of houses on either side of an attendance boundary.
Model 2 represents the baseline hedonic model without local market
conditions, estimated on the full sample of house transactions over the
years 1994-1998. Higher proportions of non-Hispanic blacks in the
student body lead to higher house prices; higher proportions of students
receiving subsidized lunches have the opposite effect. The housing
market does not discount larger minority composition, but it does
discount a greater presence of students in poverty. The model 2 point
estimates on school attribute coefficients do not differ much from those
found using model 1; the peer composition effects, however, are no
longer significant.
Model 3 adds the local market conditions variables to the baseline
hedonic model estimated on the full (not just the boundary) sample.
While Lacombe (2004) and Brasington and Haurin (2006) conclude that
ignoring neighborhood market conditions inflates school performance
value estimates, we find that including these explicit controls has no
effect on our estimates; regardless of whether local market conditions
are included in the model, school quality, as measured by student test
scores, does not affect house value in the market during this period.
The coefficient on Test changes sign, but it remains insignificant. The
estimated peer effects do not change either. (10)
We do not find the zero school quality capitalization estimates at
all surprising. One explanation arises from the approach used in the
analysis. The most important feature of attendance boundaries that make
them useful for this estimation is that they are unchanging because this
is what home owners use when forming their expectations about the local
school. But how likely were the attendance zones to remain in effect?
The attendance boundaries represented a "reasonable good-faith
effort" to desegregate the system while considering the size of the
school, the distribution of students by grade level, natural boundaries,
and, in some cases, family economics and neighborhoods. Anecdotal
evidence points out that the boundaries, once drawn, were not meeting
the requirements spelled out in the Consent Decree. For example, on
September 27, 1996, shortly after the Consent Decree was implemented,
the school board sought permission to exceed the proposed enrollment in
17 schools. Similar motions were filed on September 24, 1997, and
October 23, 1998. It quickly became apparent to close observers that the
boundaries would require adjustments in order to comply with Consent
Decree requirements. Whether the federal court would adjust the
attendance zone boundaries or would revert to past practices and order a
wholesale change in policy direction was not at all clear in light of
the previous court actions. In sum, it appears reasonable to infer that
residents had little faith that the new attendance zone system would be
retained. It would be surprising to find significant test score
capitalization under these conditions; we believe the results in Table 3
illustrate the importance of expectations in the market valuation of
school quality.
It is also likely that the general climate of litigation was simply
not conducive to optimistic expectations. Pervasive federal court
control over key management decisions created a political environment
with a limited "voice" for residents. Coupled with lack of
exit options, aside from private schools, this lack of any sense of
local control by voters likely translates into low expectations that
observed improvement by individual schools is even sustainable. (11) The
education policy of the court to that point reflected efforts to
equalize racial composition of individual schools rather than to sustain
systematic quality improvements.
Finally, we note that while the Test variable ranges from 61 to
100, the standard deviation of about 7% of the mean is modest. It is
therefore possible that our insignificant coefficient estimates are
being driven by too little variation in this variable. Nonetheless, our
premise is that this measure of school quality is treated by households
as if it were uninformative--a conclusion that remains valid even if the
reason households find the measure uninformative is that there is too
little variation in the measure as reported by the government.
5. Change in Established Attendance Zones
Five years later, in 2001, the school district undertook a broad
revision of school attendance zones. During the spring and summer of
2001, litigants and other interested parties proposed five separate
plans for revising school attendance zones, affecting up to 5452
students (10.4% of enrollment). The last of these plans was completed in
August 2001. The presiding judge, however, resigned shortly thereafter,
one week before the school year started; the new judge chose to
implement a plan that combined features of several of the proposed
plans. The resulting reassignment was implemented so quickly that the
start of the school year was postponed for many of the affected
students. We argue that, for the purposes of this study, the 2001
reassignment episode is an unanticipated random event for the affected
households.
This section examines the effect of the redistricting on house
values, using the full sample of housing transactions from 1999 to 2002,
including both houses with original school assignments and houses that
were reassigned. The models used here resemble those used in the
previous section, except that Louisiana State categorical measures of
school quality are used in lieu of student test scores, which the State
did not report in full for the earlier period. Also, we do not pursue
the boundary analysis in this section because it is not relevant for
changing attendance zones.
Table 4 reports the summary statistics for the 1999-2002 sample
pertaining to the change in attendance zones. (12) The dependent
variable, house price, has a mean of $129,250 in 1999 dollars. Variables
under the heading "House Attributes" include the number of
bedrooms (3.25), number of full bathrooms (2.04), living area in
thousand [feet.sup.2] (1.88), and net area in thousand [feet.sup.2]
(0.68).
School attributes include the SPS, school performance score (mean
of 82.5 and standard deviation of just over 17 points); the percent of
blacks, non-Hispanic (mean of 61.90%, standard deviation of 19.22%); and
the percent of students qualifying for free lunch (mean of 54.22%,
standard deviation of 17.36%).
The dummy variable, SPS Improve, uses the information about the
school's performance category at the end of two accountability
cycles. Therefore, SPS Improve equals 1 for houses with a school whose
performance category improves between two accountability cycles. (Note
that not all increases in SPS scores lead to improvements in categorical
rankings; see Table 1, for example. The empirical results, however, are
the same when replacing the categorical improvement measure SPS Improve
with an improvement measure based on the numerical change in the SPS
score.) We draw these school performance measures from the newspaper of
record for the jurisdiction, The Morning Advocate, which is the same
source for the general public.
In cases in which houses have been affected by reassignment, this
requires comparing the ranking of schools under the 1996 school
assignment and under the 2001 school assignment. We also construct
another binary variable, Reassign, that is equal to 1 if the house has
been reassigned to a different school in the 2001 attendance zone
change. Almost 19% of houses see an improvement in their school's
categorical ranking, while only a little over 6% see a decline in their
school's standing. (13) The improved ranking is due to reassignment
in only 10% of the sample. Over 70% of our sample does not see any
changes in their school's categorical ranking, even though over 11%
are reassigned to different schools.
Table 5 reports estimates of the relationship between school
quality measures and house prices for the 1999-2002 time period. Model 4
represents the baseline hedonic specification with location controls for
543 individual subdivisions, in which school quality is measured solely
by year-specific categorical performance. Model 5 adds local market
conditions to the baseline model. Model 6 introduces the school
performance growth variable, and models 7 and 8a-c allow for interaction
between dummy variables for categorical ranking improvement and
reassignment. Models 8a-c present the estimates using the location
controls based on the 12 MLS areas. As an additional robustness check
for this specification, models 8b and 8c correct for robust standard
errors and clustering at the census tract level. (14)
Model 4 in Table 5 shows that school performance scores are valued
by the housing market when no attempt is made to control for
neighborhood market conditions. Recall that Lacombe (2004) and
Brasington and Haurin (2006) find that ignoring neighborhood market
conditions inflates school performance value estimates. Once again,
however, we find no significant change in the market valuation of school
quality once these variables are added to model 5. Models 4 and 5
estimates indicate that an increase in one standard deviation in the
school performance score increases a house price by about 3.08%, or an
increase of approximately $3,900 at the mean house price ($129,250).
(15)
Introducing the change in school performance classifications in
model 6, however, alters our conclusions considerably. The SPS Improve
coefficient indicates that an improvement in the categorical ranking of
school performance is associated with a 3.8% increase in house prices
($4,900 at the mean). Our results indicate that the housing market
appears to value improvement more than relatively higher performance,
per se. (16) At the same time, the coefficient on reassignment becomes
statistically significant, or the market shows a discount for the houses
affected by reassignment. Interestingly, once we account for the
interaction between dummy variables for categorical ranking improvement
and reassignment in model 7, the SPS coefficient is no longer
statistically significant, while the SPS Improve coefficient is
associated with a 6.05% increase in house prices ($7,800 at the mean).
The correlation between SPS and SPS Improve is only -0.10, so the loss
in significance for SPS in model 7, as compared with model 6, is not
likely the result of collinearity. This impact is somewhat greater than
that found in model 6, but less than that found by Figlio and Lucas
(2002), who show that an assigned letter grade is associated with an
approximately 10% increase in a house price for each full grade
increment in the months directly following the release of the school
report cards.
The Reassign and the SPS Improve*Reassign coefficients together
indicate that the market values improvement by a given school but does
not value improvement from being reassigned to a better school (note
that the net effect of SPS Improve, Reassign, and SPS Improve*Reassign
is not significantly different from zero). (17)
Finally, it is possible, given the short time horizon, that there
could be a period of confusion immediately before or after the
reassignments were announced. To test robustness of our results, we also
introduced controls for houses that sold within six weeks of the
announcement. The results remained unchanged. (18)
We also present the model 8a-c estimates using the dummy variables
for 12 MLS areas as location controls. The major difference between our
models with individual subdivisions and MLS location controls is the
relationship between student body characteristics and house price. Table
5 illustrates that the representation of black students in local public
schools leads to an increase in property values after controlling
directly for school performance: models 8a and 8c.
Models 8b and 8c provide estimates for errors clustered by census
tract and bootstrapped errors, respectively. The bootstrapped errors
alter none of the previous conclusions. The Reassign variable, however,
becomes insignificant when correcting for clustered errors, as do the
student composition variables Black School and Free Lunch. The clustered
error estimates indicate that reassignment per se has no effect on house
price. Nonetheless, the set of coefficients for SPS Improve, Reassign,
and SPS Improve*Reassign are jointly, not significantly, different from
zero. Coupled with the positive and significant SPS Improve coefficient,
these results yield the same conclusion as model 7; home owners value
improvement in a given school but not improvement that comes from being
reassigned to a better school.
Nonetheless, in all models, the housing market is not directly
discounting schools based on race, a result similar to that of Norris
(2002) for six other Louisiana parishes. We conclude that property
values are not systematically influenced by school racial integration. A
similar conclusion holds for the proportion of the student body eligible
for free lunch; the effect is negative and significant only in models 8a
and 8c.
In conclusion, while we find that the housing market reveals no
stable capitalization of year-specific test scores, we show that an
improvement in a school's categorical ranking has a large and
significant impact on house price. The state focuses a large share of
the financial rewards on the schools that meet or exceed growth targets.
We cannot ascertain whether our results show that the housing market
shares state policymakers' enthusiasm for these measures or that
the housing market values the incremental state resources for schools
whose categorical rankings improve.
6. Voting Behavior and Capitalization
Fischel's (2001) homevoter hypothesis hinges on the assumption
that home owners have an incentive to vote for policies that increase
their property values and to vote against policies that reduce it. This
behavior is driven by the unique role that home ownership plays in the
typical family's finances. An owner of a home has a large fraction
of his/her wealth tied up in the property. To sell the property at a
loss has potentially much larger consequences than selling any other
financial asset. Similarly, home owners cannot diversify their
portfolios by spreading out ownership of their assets among more
risk-neutral investments. As a result of this large concentration of
wealth in one asset, people who buy houses are more careful about it
than they are about almost any other type of transaction. Thus, they are
willing to prevent the unfavorable events that reduce their home values,
assumed here to be school quality. In the case of East Baton Rouge
Parish, this implies that after home owners purchase their property, the
best hope of maintaining or improving one's investment is to
exercise "voice" through the political processes that address
the school quality. The problem confronting residents of the parish,
however, is that the federal court controlled virtually all important
management aspects of the school system leaving residents no opportunity
to express their preferences through the usual political channels. Only
after the Consent Decree was established in 1996 and voters experienced
school management stability did voters signal a fundamental change in
their expectations about the public school system. The $280 million
school tax referendum in November of 1998 provides an opportunity to
test Fischel's (2001) maintained hypothesis regarding voting
behavior, expectations, and capitalization. This tax proposal was the
first such tax plan approved by the electorate in more than 25 years.
In order to further examine the tie between observed capitalization
and expectations, we reestimate the capitalization models in the
following manner. We use precinct-level voting data to split the housing
market into houses located in areas that supported the tax initiative
and those in areas that did not. These samples are indicated by the
"Vote Yes" and "Vote No" columns in Table 6. We
estimate four hedonic price functions for each of these subsamples,
using the individual subdivisions as location controls with robust
standard errors and using the MLS area with location controls with
robust standard errors. We then repeat the MLS model, but consider
robust standard errors corrected for clustering at the census tract
level and bootstrapped standard errors.
Under Fischel's (2001) hypothesis, we expect to find positive
capitalization of school quality in the sample of precincts supporting
the tax and weaker or no capitalization in the sample voting against the
referendum. The table only reports the coefficients on the key school
characteristic variables. Several interesting results are apparent.
First, Fischel's (2001) homevoter hypothesis is broadly supported
by the capitalization pattern in the table: Supporters of the tax plan
exhibit stronger positive school quality capitalization than do
opponents. Opponents, on the other hand, exhibit significantly negative
capitalization in the robust standard error and bootstrapped standard
error models, while exhibiting insignificant school quality
capitalization in the clustered error model. In addition, the sample of
precincts supporting the school tax indicates that this part of the
market values both differences in level of performance (SPS) as well as
improvement in performance (SPS Improve). Once again, improvement from
reassignment is not as highly valued as improvement of a particular
school, a result consistent with the pooled sample estimates.
7. Conclusion
The uniqueness of the Baton Rouge experience enables us to deal
effectively with a number of issues with which the housing literature is
typically concerned. First, public schools are only one of the public
services attached to any particular location. Because we consider a
single school district that lies entirely within a single city-parish
government jurisdiction, we can adequately control for local taxes and
the provision of public services other than elementary education.
Second, the specific changes in school assignment observed in this case
provide natural experiments in education policy, rare events that allow
us to examine housing market responses to direct and indirect changes in
school quality and peer effects under vastly different expectation
conditions.
Even though the local housing market reveals very small
capitalization of year-specific school scores measuring broader
performance than simple student test scores in Baton Rouge, improvements
in categorical rankings have a more profound impact on house prices.
This resembles the results of Figlio and Lucas (2002), who find some
evidence that public school report cards affect house values in Florida.
The results also indicate that parents care about how school improvement
occurs; improving a given school is valued more highly than being
reassigned to a better school. Louisiana focuses a large share of the
financial rewards on the schools that meet or exceed growth targets. At
face value, state resources appear to be allocated consistently with
measures most valued by the housing market. Alternatively, it could be
that the housing market simply recognizes the incremental flow of
resources that measured improvement attracts from the state. The latter
view, of course, also means that the housing market expects any
subsequent increase in school resources to be applied in a way that
yields a valued improvement in school quality.
References
Black, S. 1999. Do better schools matter? Parental valuation of
elementary education. Quarterly Journal o/" Economies 114:579-99.
Brasington, D. M., and D. R. Haurin. 2006. Educational outcomes and
house values: A test of the value added approach. Journal of Regional
Science 46:245-68.
Conley, J., and M. Dix. 2004. Beneficial inequality in the
provision of municipal services: Why rich neighborhoods should get
plowed first. Southern Economic Journal 70:731-45.
Cushing, B. J. 1984. Capitalization of interjurisdictional fiscal
differentials: An alternative approach. Journal of Urban Economics
15:317-26.
Davis, Jr.. et al. v. East Baton Rouge Parish School Bd.. et al. D
La., Baton Rouge Div., Civ. #1662. 1956.
Davis, Jr.. et al. v. East Baton Rouge Parish School Bd., 514 F.
Supp. 869 (1981), p. 877.
Downes, T. A., and J. E. Zabel. 2002. The impact of school
characteristics on house prices: Chicago 1987-1991. Journal of Urban
Economics 52:1-25.
Figlio, D. N. 1999. Functional form and the estimated effects of
school resources. Economics of Education Review 18:241-52.
Figlio, D. N., and M. E. Lucas. 2002. What's in a grade?
School report cards and house prices. American Economic Review
94:591-604.
Fischel, W. A. 2001. The Homevoter Hypothesis. Cambridge, MA:
Harvard University Press.
Gill, H. L. 1983. Changes in city and suburban house prices during
a period of expected school desegregation. Southern Economic Journal
50:169-84.
Hanushek, E. A., J. R. Kain, and S. G. Rivkin. 2002. New evidence
about Brown v. Board of Education: The complex effects of school racial
composition on achievement. NBER Working Paper No. 8741.
Hanushek, E. A., and L. L. Taylor. 1990. Alternative assessments of
the performance of schools: Measurement of state variations in
achievement. The Journal of Human Resources 25:79-201.
Haurin, D. R., and D. Brasington. 1996. School quality and real
house prices: Inter- and intrametropolitan effects. Journal of Housing
Economics 5:351-68.
Haurin, D. R., and D. Brasington. 2007. Parents, peers, or school
inputs: Which components of school outcomes are capitalized into house
value? Working Paper, University of Cincinnati.
Hayes, K. J., and L. L. Taylor. 1996. Neighborhood school
characteristics: What signals quality to homebuyers? Economic and
Financial Policy Review, Federal Reserve Bank of Dallas, QIV:2-9.
Hoxby, C. M. 2000. Does competition among public schools benefit
students and taxpayers? American Economic Review 90:1209-38.
Kane, T., D. O. Staiger, and G. Samms. 2003. School accountability
ratings and housing values. In Brookings-Wharton papers on urban
affairs. Washington, DC: Brookings Institution Press.
Lacombe, D. 2004. Does econometric methodology matter? An analysis
of public policy using spatial econometric techniques. Geographical
Analysis 36:105-18.
Louisiana. Department of Education. Annual Financial and
Statistical Report, 2002-2003. 154th Edition. Baton Rouge, LA: State of
Louisiana Department of Education. (2004). Print.
Norris, D. N., Jr. 2002. Schools, race and housing values; The
effect of school quality improvements on residential housing prices in
Louisiana. Unpublished paper, Southern Economic Association Conference.
Reback, R. 2005. House prices and the provision of public services:
Capitalization under school choice programs. Journal of Urban Economics
57:275-301.
Ries, J., and T. Somerville. 2004. School quality and residential
property values: Evidence from Vancouver rezoning. Working Paper,
University of British Columbia.
Ross, S., and J. Yinger. 1999. Sorting and voting: A review of the
literature on urban public finance. In Handbook of Regional and Urban
Economics, edited by Paul Cheshire and Edwin S. Mills. Amsterdam:
Elsevier, pp. 2001-60.
Turnbull, G. K., and J. Dombrow. 2006. Spatial competition and
shopping externalities: Evidence from the housing market. Journal of
Real Estate Finance and Economics 32:391-408.
Weimer, D. L., and M. J. Wolkoff. 2001. School performance and
house values: Using noncontiguous district and incorporation boundaries
to identify school effects. National Tax Journal 54:231-54.
(1) Note that this situation differs from the public school choice
(open-enrollment) program studied in Reback (2005). Unlike the
open-enrollment program, which gives students the opportunity to enroll
in school districts other than the one in which they reside, the Baton
Rouge school district imposed random school assignments, which resulted
in mandatory busing for its students. In an effort to achieve racial
balance, formerly white and formerly black schools were paired or
clustered, and students were bused to their clusters based on the need
to create racial balance.
(2) Parishes in Louisiana are political subdivisions equivalent to
counties in other states. Conley and Dix (2004) conclude that central
city governments may have incentives to vary service quality across
neighborhoods in order to retain mobile higher income residents.
Nonetheless, it seems reasonable to expect that services vary less
within the jurisdiction than across jurisdiction boundaries. In any
case, including a neighborhood income variable in the empirical
capitalization model should control for the Conley-Dix effect.
(3) See Ross and Yinger (1999) and Fischel (2001) for overviews of
the property tax capitalization literature and empirical public service
quality.
(4) Students were randomly assigned to schools within
court-designated "clusters," each composed of schools with
three or four different races. This created an environment in which it
was not unusual for a given student to attend a different school each
year and in which siblings often attended different schools. See Davis
et al. v East Baton Rouge Parish Sch. B (1981) for the plan details.
(5) Enrollment data are from Louisiana Department of Education,
Annual Financial Report, various years.
(6) These numbers are for elementary schools only. Schools with
grades 9-12 and 9-12 portions of K-12 schools (i.e., high school and
combination schools) officially entered the Louisiana School
Accountability System in 2001.
(7) Most of the literature investigating school quality effects on
the housing market uses the hedonic price approach. See Ries and
Somerville (2004) for an example of the repeat sales methodology. The
small repeat sample of 256 observations, of which only 28 were directly
reassigned, precludes using the repeat sales methodology for Baton
Rouge.
(8) The price adjustment uses the CPI for South Urban, obtained
January 18, 2006, from http://data.bls.gov/, Series Id: CUUR0300SA0,
CUUS0300SA0.
(9) Private school enrollment data come from the National Center of
Education Statistics' Common Core Data.
(10) While there is high correlation among Test, Black School. Free
Lunch, Black Pct, and Income, the relevant variance inflation factors
are below 20, the cutoff normally used to indicate possible deleterious
multicolinearity effects.
(11) East Baton Rouge Parish covers much of the closely settled
region of the Baton Rouge metropolitan area. In addition, the peculiar
geography of the metropolitan area limits the number of good substitute
locations for households employed in Baton Rouge. The parish is bounded
by poorly drained areas subject to frequent flooding on the northeast
and the south and major rivers on the west (Mississippi River) and east
(Amite River). The few bridges across the Mississippi and Amite limit
accessibility across those boundaries.
(12) The school quality measures were initiated in 1998. Following
convention, the relevant school measures are lagged one year. This
ensures full exposure of the house seller and buyer to the school
ranking at the time the property is put on the market.
(13) Louisiana Department of Education reports that elementary
schools improved 22 points on the score basis from 1999 to 2004.
However, the initial spike in the average SPS from 1999 to 2001 is
followed by steady decreases through the third accountability cycle in
2004. Note that the changes in SPS scores do not necessarily imply a
change in categorical ranking. Recall that SPS Improve only reflects an
improvement in the categorical ranking.
(14) The model 7 estimates for bootstrapped standard errors are the
same and therefore are not reported in the table. However, in order to
obtain estimates using robust errors clustered at the census tract level
in the subdivision model, we had to drop subdivisions with fewer than 12
transactions from the sample. The estimates, however, yield the same
conclusions as model 7.
(15) All dollar amounts are in 1999 dollars.
(16) Recall, however, that most of the public schools in this
district are performing at a low level in any event.
(17) Reassignment is not only associated with changes in school
performance scores or categorical rankings but also with changes in
other school characteristics, such as the percent of the student body
that is black, Black School, and the percent of the student body
receiving free or reduced-price lunches, Free Lunch. About 78% of
schools that have recorded an increase in their SPS score due (not to a
change in categorical ranking) to reassignment also recorded a decrease
in percent Black School. All of the schools that have recorded an
improvement in school categorical ranking also recorded a decrease in
percent Black in the neighborhood of 15%. Similarly, schools that
dropped in a categorical ranking at the end of the accountability cycle
as a result of reassignment recorded an increase in percent Black
School. The largest drop in SPS score was about 30 points, accompanied
by an increase in Black School (by 23%) and an increase in Free Lunch
(by 33%). The largest increase in SPS score was about 33 points, and it
was accompanied by 14% and 16% drops in percent Black School and Free
Lunch, respectively. More specifically, there were 233 observations
originally assigned to Wedgwood Elementary School. About one third of
these observations recorded a significant increase in SPS scores of 24
points but did not see a change in categorical rankings. These
observations related to houses redistricted from Wedgwood Elementary to
Shenandoah Elementary.
(18) These results are available by request.
Velma Zahirovic-Herbert, Department of Housing and Consumer
Economics, University of Georgia, 213 Dawson Hall, Athens, GA
30602-3622, USA, E-mail
[email protected]; corresponding author.
Geoffrey K. Turnbull, Department of Economics. Andrew Young School
of Policy Studies, Georgia State University, P.O. Box 3992, Atlanta, GA
30302-3992, USA: E-mail
[email protected].
We would like to thank Christopher Bollinger and two anonymous
referees for their helpful comments. The authors are responsible for any
errors.
Received August 2007: accepted June 2008.
Table 1. 2000-2001 Louisiana School Performance Labels
School Performance Label SPS Range
School of Academic Excellence 150.0 or above
School of Academic Distinction 125.0-149.9
School of Academic Achievement 100.0-124.9
Academically Above the State Average 79.9-99.9
Academically Below the State Average 30.1-99.9
Academically Unacceptable School 30 or below
Table 2. Summary Statistics for the 1994-1998 Sample
Variable (description) Obs
Dependent variable
Price (sold price in 1999$) 9821
School attributes (a)
Test (percent students passing CRCT) 5363
Black school (percent black 5363
students in school)
Free lunch (percent students on 5363
free lunch)
House attributes
DOM (days on market) 9821
Bedrooms (number of bedrooms) 9821
Full bath (number of full bathrooms) 9821
Living area (living area in 9821
thousand [feet.sup.2])
Net area (total area under roof 9821
minus living area)
Neighborhood attributes
Owner Pet (percent of owner 9821
occupied housing)
Black Pet (percent black in tract) 9821
Income (median household income 9821
in thousand 1999$)
Private Pet (percent of children 9821
enrolled in private schools)
Local market conditions
Listing density 9821
New density (new listing density) 9821
Vacant density (vacant 9821
listing density)
Variable (description) Mean
Dependent variable
Price (sold price in 1999$) 123,597.3
School attributes (a)
Test (percent students passing CRCT) 89.62227
Black school (percent black 0.5104318
students in school)
Free lunch (percent students on 0.4901844
free lunch)
House attributes
DOM (days on market) 65.30537
Bedrooms (number of bedrooms) 3.269219
Full bath (number of full bathrooms) 2.03187
Living area (living area in 1.90040
thousand [feet.sup.2])
Net area (total area under roof 0.686461
minus living area)
Neighborhood attributes
Owner Pet (percent of owner 0.6686114
occupied housing)
Black Pet (percent black in tract) 0.2247747
Income (median household income 50.04975
in thousand 1999$)
Private Pet (percent of children 0.052336
enrolled in private schools)
Local market conditions
Listing density 3.642229
New density (new listing density) 1.774355
Vacant density (vacant 0.5578697
listing density)
Variable (description) SD
Dependent variable
Price (sold price in 1999$) 56,276.25
School attributes (a)
Test (percent students passing CRCT) 6.357115
Black school (percent black 0.2500155
students in school)
Free lunch (percent students on 0.2159264
free lunch)
House attributes
DOM (days on market) 43.03807
Bedrooms (number of bedrooms) 0.6281082
Full bath (number of full bathrooms) 0.4789908
Living area (living area in 0.5631569
thousand [feet.sup.2])
Net area (total area under roof 0.286104
minus living area)
Neighborhood attributes
Owner Pet (percent of owner 0.1879812
occupied housing)
Black Pet (percent black in tract) 0.2380711
Income (median household income 15.30714
in thousand 1999$)
Private Pet (percent of children 0.0830685
enrolled in private schools)
Local market conditions
Listing density 2.656127
New density (new listing density) 1.515996
Vacant density (vacant 1.131285
listing density)
Variable (description) Minimum
Dependent variable
Price (sold price in 1999$) 40,000
School attributes (a)
Test (percent students passing CRCT) 60.75
Black school (percent black 0.0764706
students in school)
Free lunch (percent students on 0.0658824
free lunch)
House attributes
DOM (days on market) 14
Bedrooms (number of bedrooms) 1
Full bath (number of full bathrooms) 1
Living area (living area in 0.62
thousand [feet.sup.2])
Net area (total area under roof 0.1
minus living area)
Neighborhood attributes
Owner Pet (percent of owner 0.0506403
occupied housing)
Black Pet (percent black in tract) 0.0096137
Income (median household income 11.397
in thousand 1999$)
Private Pet (percent of children 0
enrolled in private schools)
Local market conditions
Listing density 0
New density (new listing density) 0
Vacant density (vacant 0
listing density)
Variable (description) Maximum
Dependent variable
Price (sold price in 1999$) 358,146.5
School attributes (a)
Test (percent students passing CRCT) 99.75
Black school (percent black 1
students in school)
Free lunch (percent students on 0.94375
free lunch)
House attributes
DOM (days on market) 180
Bedrooms (number of bedrooms) 6
Full bath (number of full bathrooms) 4
Living area (living area in 4.46
thousand [feet.sup.2])
Net area (total area under roof 1.995
minus living area)
Neighborhood attributes
Owner Pet (percent of owner 0.908982
occupied housing)
Black Pet (percent black in tract) 0.9836207
Income (median household income 78.509
in thousand 1999$)
Private Pet (percent of children 0.3732262
enrolled in private schools)
Local market conditions
Listing density 18.79724
New density (new listing density) 11.88706
Vacant density (vacant 10.48087
listing density)
Obs, observations; SD, standard deviation.
(a) School averages are calculated using only houses sold after the
publication of the Consent Decree document containing school
attendance zones. These cover sales made after June of 1996.
Table 3. Regression Results Dependent Variable: In(sold price in
1999$) (Standard Errors in Parentheses)
(1) (2)
Error estimate Robust Robust
Local market
conditions No No
Location controls School boundaries (b) MLS areas
School attributes
Test (x) 0.000904 (0.00087) 0.0000326 (0.00063)
Black school 0.0590 (0.041) 0.0550 ** (0.026)
Lunch school -0.0506 (0.052) -0.0655 ** (0.031)
Constant 10.71 *** (0.069) 10.69 *** (0.017)
Observations 6284 9821
Adjusted [R.sup.2] 0.87 0.86
(3a) (3b)
Error estimate Robust Clustered (a)
Local market
conditions Yes Yes
Location controls MLS areas MLS areas
School attributes
Test (x) -0.0000264 (0.00064) -0.0000264 (0.0010)
Black school 0.0566 ** (0.026) 0.0566 (0.059)
Lunch school -0.0637 ** (0.031) -0.0637 (0.075)
Constant 10.71 *** (0.017) 10.71 *** (0.045)
Observations 9821 9821
Adjusted [R.sup.2] 0.86 0.86
(3c)
Error estimate Bootstrapped
Local market
conditions Yes
Location controls MLS areas
School attributes
Test (x) -0.0000264 (0.00062)
Black school 0.0566 ** (0.025)
Lunch school -0.0637 ** (0.030)
Constant 10.71 *** (0.017)
Observations 9821
Adjusted [R.sup.2] 0.86
Coefficients for House Attributes, Neighborhood Attributes, and Local
Market Conditions as well as house age range, year, and season sold
are not reported here.
(a) Robust standard errors adjusted for clustering at the census tract
level.
(b) This specification includes location controls based on 159 school
attendance boundaries.
(x) Test represents the percent of students passing a CRCT test. That
is the percent scoring at or above the performance standard that the
state has set in that subject area.
** p < 0.05.
*** p < 0.01.
Table 4. Summary Statistics for the 1999-2002 Sample
Variable (Description) Obs
Dependant variable
Price (sold price in 1999$) 6169
School attributes
SPS (School Performance Score) 6169
Black school (percent black 6169
students in school)
Free lunch (percent students 6169
on free lunch)
SPS improve (school improved 6169
ranking dummy)
Reassign (reassignment dummy) 6169
House attributes
DOM (days on market) 6169
Bedrooms (number of bedrooms) 6169
Full bath (number of full bathrooms) 6169
Living area (living area in 6169
thousand [feet.sup.2])
Net area (total area under roof 6169
minus living area)
Neighborhood attributes
Black Pet (percent black in tract) 6169
Income (median household income 6169
in thousand 1999$)
Private Pet (percent children 6169
enrolled in private schools)
Local market conditions
Listing density 6169
New density (new listing density) 6169
Vacant density (vacant 6169
listing density)
Variable (Description) Mean
Dependant variable
Price (sold price in 1999$) 129,250
School attributes
SPS (School Performance Score) 82.50089
Black school (percent black .61903
students in school)
Free lunch (percent students .5422346
on free lunch)
SPS improve (school improved .1778246
ranking dummy)
Reassign (reassignment dummy) .123845
House attributes
DOM (days on market) 68.56865
Bedrooms (number of bedrooms) 3.246069
Full bath (number of full bathrooms) 2.036148
Living area (living area in 1.875477
thousand [feet.sup.2])
Net area (total area under roof 0.679019
minus living area)
Neighborhood attributes
Black Pet (percent black in tract) 0.205967
Income (median household income 50.7434
in thousand 1999$)
Private Pet (percent children 0.049926
enrolled in private schools)
Local market conditions
Listing density 3.826291
New density (new listing density) 1.756177
Vacant density (vacant 1.900903
listing density)
Variable (Description) SD
Dependant variable
Price (sold price in 1999$) 54,838.95
School attributes
SPS (School Performance Score) 17.10345
Black school (percent black .1921673
students in school)
Free lunch (percent students .1735621
on free lunch)
SPS improve (school improved .382396
ranking dummy)
Reassign (reassignment dummy) .3294314
House attributes
DOM (days on market) 44.50294
Bedrooms (number of bedrooms) 0.6225428
Full bath (number of full bathrooms) 0.490992
Living area (living area in 0.5437634
thousand [feet.sup.2])
Net area (total area under roof 0.2799634
minus living area)
Neighborhood attributes
Black Pet (percent black in tract) 0.2296404
Income (median household income 14.69561
in thousand 1999$)
Private Pet (percent children 0.0779826
enrolled in private schools)
Local market conditions
Listing density 2.518737
New density (new listing density) 1.460711
Vacant density (vacant 1.747426
listing density)
Variable (Description) Minimum
Dependant variable
Price (sold price in 1999$) 40,000
School attributes
SPS (School Performance Score) 40.9
Black school (percent black .2795486
students in school)
Free lunch (percent students .1873258
on free lunch)
SPS improve (school improved 0
ranking dummy)
Reassign (reassignment dummy) 0
House attributes
DOM (days on market) 14
Bedrooms (number of bedrooms) 1
Full bath (number of full bathrooms) 1
Living area (living area in 0.703
thousand [feet.sup.2])
Net area (total area under roof 0.11
minus living area)
Neighborhood attributes
Black Pet (percent black in tract) 0.009614
Income (median household income 11.397
in thousand 1999$)
Private Pet (percent children 0
enrolled in private schools)
Local market conditions
Listing density 0
New density (new listing density) 0
Vacant density (vacant 0
listing density)
Variable (Description) Maximum
Dependant variable
Price (sold price in 1999$) 320,000
School attributes
SPS (School Performance Score) 123
Black school (percent black .9987374
students in school)
Free lunch (percent students .9306282
on free lunch)
SPS improve (school improved 1
ranking dummy)
Reassign (reassignment dummy) 1
House attributes
DOM (days on market) 180
Bedrooms (number of bedrooms) 5
Full bath (number of full bathrooms) 5
Living area (living area in 4.435
thousand [feet.sup.2])
Net area (total area under roof 1.995
minus living area)
Neighborhood attributes
Black Pet (percent black in tract) 0.983621
Income (median household income 78.509
in thousand 1999$)
Private Pet (percent children 0.373226
enrolled in private schools)
Local market conditions
Listing density 18.30214
New density (new listing density) 11.30623
Vacant density (vacant 13.59951
listing density)
Obs, observations; SD, standard deviation.
Table 5. Regression Results Dependent Variable: ln
(Sold Price in 1999$) (Standard Errors in Parentheses)
(4) (5)
Error estimate Robust Robust
Local market conditions No Yes
Location controls Subdivisions (b) Subdivisions (b)
School attributes
SPS 0.00181 * (0.00093) 0.00179 * (0.00093)
Black school 0.0448 0.0527
Free lunch 0.194 0.188
SPS improve
Reassign -0.0129 (0.0089) -0.0134
SPS improve * reassign
Constant 10.68 *** (0.18) 10.67 *** (0.18)
Observations (c) 6169 6169
Adjusted R2 0.92 0.92
(6) (7)
Error estimate Robust Robust
Local market conditions Yes Yes
Location controls Subdivisions (b) Subdivisions (b)
School attributes
SPS 0.00160 * (0.00095) 0.00152
Black school 0.0682 0.0863
Free lunch 0.172 0.117
SPS improve 0.0388 *** (0.013) 0.0605 *** (0.017)
Reassign -0.0210 ** (0.0091) -0.00508
SPS improve * reassign -0.0765** (0.031)
Constant 10.65 *** (0.18) 10.68*** (0.19)
Observations (c) 6169 6169
Adjusted R2 0.92 0.92
(8a) (8b)
Error estimate Robust Clustered (a)
Local market conditions Yes Yes
Location controls MLS areas MLS areas
School attributes
SPS 0.000605 (0.00054) 0.000605 (0.00090)
Black school 0.113 *** (0.040) 0.113
Free lunch -0.154 *** (0.058) -0.154
SPS improve 0.0570 *** (0.0072) 0.0570 *** (0.019)
Reassign 0.0265 *** (0.0064) 0.0265
SPS improve * reassign -0.0928 *** (0.018) -0.0928 ** (0.039)
Constant 10.81 *** (0.075) 10.81 *** (0.13)
Observations (c) 6414 6414
Adjusted R2 0.86 0.86
(8c)
Error estimate Bootstrapped
Local market conditions Yes
Location controls MLS areas
School attributes
SPS 0.000605 (0.00052)
Black school 0.113 *** (0.039)
Free lunch -0.154 *** (0.054)
SPS improve 0.0570 *** (0.0070)
Reassign 0.0265 *** (0.0061)
SPS improve * reassign -0.0928 *** (0.018)
Constant 10.81 *** (0.073)
Observations (c) 6414
Adjusted R2 0.86
Coefficients for House Attributes, Neighborhood Attributes, and Local
Market Conditions, as well as the dummy variables For house age range,
year, season, and month sold are not reported here.
(a) Robust standard errors adjusted for clustering at the census
tract level.
(b) This specification includes location controls based on 543
subdivisions.
(c) Sample size differs because not all of the observations have
information on individual subdivisions, while all of the
observations in our data are assigned to MLS areas.
* p < 0.1.
** p < 0.05.
*** p < 0.01.
Table 6. Sub-Sample Regression Results 1999-2002: Dependent Variable:
ln (sold price in 1999$) (Standard Errors in Parentheses)
Error estimate Robust
Local market conditions Yes
Location controls Subdivisions
School attributes
SPS 0.00103 (0.0012)
Black school 0.211 ** (0.11)
Lunch school -0.152 (0.17)
SPS improve 0.0782 *** (0.020)
Reassign -0.00916 (0.012)
SPS improve -0.0883 ** (0.035)
Constant 10.91 *** (0.21)
Observations 4369
Adjusted R(2) 0.92
Error estimate Robust
Local market conditions Yes
Location controls MLS areas
School attributes
SPS 0.00215 *** (0.00065)
Black school 0.193 *** (0.064)
Lunch school -0.00367 (0.080)
SPS improve 0.0670 *** (0.0085)
Reassign 0.00632 (0.0073)
SPS improve -0.0999 *** (0.022)
Constant 10.65 *** (0.097)
Observations 4600
Adjusted R(2) 0.86
(9) Vote Yes
Error estimate Clustered (a)
Local market conditions Yes
Location controls MLS areas
School attributes
SPS 0.00215 *** (0.00093)
Black school 0.193 (0.13)
Lunch school -0.00367 (0.17)
SPS improve 0.0670 *** (0.021)
Reassign 0.00632 (0.013)
SPS improve*reassign -0.0999 *** (0.033)
Constant 10.65 *** (0.14)
Observations 4600
Adjusted R(2) 0.86
Error estimate Bootstrapped
Local market conditions Yes
Location controls MLS areas
School attributes
SPS 0.00215 *** (0.00064)
Black school 0.193 *** (0.060)
Lunch school -0.00367 (0.077)
SPS improve 0.0670 *** (0.0086)
Reassign 0.00632 (0.0074)
SPS improve*reassign -0.0999 *** (0.022)
Constant 10.65 *** (0.093)
Observations 4600
Adjusted R(2) 0.86
Error estimate Robust
Local market conditions Yes
Location controls Subdivisions
School attributes
SPS -0.000477 (0.0058)
Black school -0.0496 (0.23)
Lunch school -0.252 (0.64)
SPS improve 0.0736 (0.057)
Reassign 0.0387 (0.071)
SPS improve*reassign -0.208 * (0.11)
Constant 10.84 *** (0.91)
Observations 1072
Adjusted R(2) 0.90
(10) Vote No
Error estimate Robust
Local market conditions Yes
Location controls MLS areas
School attributes
SPS -0.00398 ** (0.0018)
Black school 0.0495 (0.063
Lunch school -0.585 *** (0.15)
SPS improve -0.0216 (0.028)
Reassign 0.08l3 *** (0.021)
SPS improve*reassign -0.175 *** (0.046)
Constant 12.33 *** (0.33)
Observations 1083
Adjusted R(2) 0.81
Error estimate Clustered
Local market conditions Yes
Location controls MLS areas
School attributes
SPS -0.00398 (0.0024)
Black school 0.0495 (0.099)
Lunch school -0.585 *** (0.18)
SPS improve -0.0216 (0.037)
Reassign 0.08l3 ** (0.034)
SPS improve*reassign -0.175 ** (0.063)
Constant 12.33 *** (0.36)
Observations 1083
Adjusted R(2) 0.81
Error estimate Bootstrapped
Local market conditions Yes
Location controls MLS areas
School attributes
SPS -0.00398 ** (0.0018)
Black school 0.0495 (0.062)
Lunch school -0.585 *** (0.15)
SPS improve -0.0216 (0.027)
Reassign 0.0813 *** (0.020)
SPS improve*reassign -0.175 *** (0.047)
Constant 12.33 *** (0.33)
Observations 1083
Adjusted R(2) 0.81
Coefficients for House Attributes, Neighborhood Attributes, and
Local Market Conditions, as well as the dummy variables for house
age range, year, season, and month sold are not reported here.
(a) Robust standard errors adjusted for clustering at the census
tract level.
* p <0.1.
** p < 0.05.
*** p < 0.01.
