# RECORD 620cbd7217ca28bcea726920009e42496a17ea8b
# TITLE: The Impact of Brownfields on Residential Property Values in Cincinnati, Ohio: A Spatial Hedonic Approach
# YEAR: 2012.0  VENUE: Journal of Regional Analysis & Policy  DOI: 10.68479/001c.7941
# AUTHORS: Oana Mihaescu; Rainer Vom Hofe
# PAGES: 14  CANDIDATE TABLE PAGES (1-based): [9]
# ABSTRACT: This paper empirically assesses the impact brownfield sites have on market values of
 single-family residential properties. Using three different hedonic model specifications, Ordinary Least Squares and Spatial Autoregressive and Spatial Error Models, this study shows that properties located less than 1,000 feet from a brownfield site experience a significant depreciation in property values. As expected, spatial dependence among real estate property values is present, meaning that the two spatial hedonic model specifications are superior to the Ordinary Least Squares specification. The study contributes to the relevant real estate and urban economics literature by determining the negative impact brownfield sites have on the value of single-family residential properties in our study region in Cincinnati, Ohio, by explicitly accounting for the phenomenon of spatial dependence in the dependent variable and by addressing the declining influence of brownfield sites on property values in a nonlinear relationship


===== PAGE 1 (score 10, numeric-paren 0, table-words 1, var-words 8) =====

 
The Impact of Brownfields on Residential Property Values in 
Cincinnati, Ohio: A Spatial Hedonic Approach 
 
 
Oana Mihaescu and Rainer vom Hofe 
School of Planning, University of Cincinnati – USA 
 
 
 
Abstract.  This paper empirically assesses the impact brownfield sites have on market values of 
single-family residential properties.  Using three different hedonic model specifications, Ordi-
nary Least Squares and Spatial Autoregressive and Spatial Error Models, this study shows 
that properties located less than 1,000 feet from a brownfield site experience a significant de-
preciation in property values.  As expected, spatial dependence among real estate property 
values is present, meaning that the two spatial hedonic model specifications are superior to 
the Ordinary Least Squares specification.  The study contributes to the relevant real estate and 
urban economics literature by determining the negative impact brownfield sites have on the 
value of single-family residential properties in our study region in Cincinnati, Ohio, by explic-
itly accounting for the phenomenon of spatial dependence in the dependent variable and by 
addressing the declining influence of brownfield sites on property values in a nonlinear  
relationship. 
 
 
 
1. Introduction 
 
Brownfield sites (or simply brownfields) are 
abandoned, idled, or underused real properties that 
are currently or were formerly employed for indus-
trial or commercial purposes and whose expansion, 
redevelopment, or reuse may be complicated by the 
presence or potential presence of a hazardous sub-
stance, pollutant, or contaminant (Braswell, 1999; 
Bacot and O'Dell, 2006; US EPA, 2009, among oth-
ers).  Contaminated properties yield little revenue 
for local governments.  They are unsafe for ecosys-
tems and human health and impair the neighbor-
hoods that contain them, both socially and aestheti-
cally.  In addition, they negatively impact the hous-
ing prices of residential properties that are located in 
close proximity (Bromberg and Spiesman, 2006). 
It comes as no surprise that a large number of 
studies have strived to understand the benefits of 
brownfield redevelopment.  Attoh-Okine and Gib-
bons (2001), Amekudzi et al. (2003), and Barnett 
(2006), for instance, focused on the revitalization of 
inner city neighborhoods through housing, jobs, and 
tax revenues resulting from brownfield redevelop-
ment.  In an environmentally-oriented regional level 
study, Alberini et al. (2005) argued that redevelop-
ing brownfields helps to reduce the harmful effects 
of the site's groundwater, water, soil, and/or air pol-
lution on ecological systems as well as on human 
health1.  Akinmoladun and Lewis (1998) and Attoh-
Okine and Gibbons (2001) emphasize the im-
portance of brownfield redevelopment for combat-
ing the negative environmental effects of urban 
sprawl.  Amekudzi et al. (2003) studied the effects of 
infill development to address urban sprawl.  Gist 
(1999), Amekudzi et al. (2003), and Bacot and O'Dell 
(2006) focused on the impacts that redevelopment of 
contaminated properties has on neighborhoods and 
their citizens. 
In this study, we use the spatial hedonic pricing 
framework to empirically estimate the impacts 
                                                 
1 See also Bartsch, 1997; Attoh-Okine and Gibbons, 2001; Bacot 
and O'Dell, 2006. 
JRAP 42(3): 223-236.   © 2012 MCRSA. All rights reserved. 
                                                                


===== PAGE 9 (score 122, numeric-paren 86, table-words 10, var-words 16) =====

Spatial Hedonic Analysis of Brownfield Impact on Property Values                                                               
 
   231 
 
 
variables remained in the final model specifications.  
Among others, the total finished square footage of 
the house (HOUSESQFT) and the total number of 
rooms (TOTALROOMS) were highly correlated with 
the other structural variables and were therefore 
removed from the model.  Though multicollinearity 
may not reduce the usefulness of the model as a 
whole, it does have severe consequences for the in-
terpretation of the marginal effects of individual re-
gression coefficients (Chatterjee and Hadi, 2006), for 
example, the effect of the distance of residential 
properties on nearby brownfield sites (DISTBR).  
The problem of ambiguous regression results has 
been avoided by removing some of the severely cor-
related explanatory variables, while keeping almost 
all of the important and relevant variables in the 
model. 
In addition, for reasons of parsimony the median 
household income by ZIP code (MEDHHINC) and 
the distance to the CBD (DISTCBD) variables were 
removed due to statistically insignificant parame-
ters.  Of the remaining fifteen variables, all calculat-
ed coefficients have the expected sign, an indication 
that no severe multicollinearity problems are pre-
sent (see Table 4).  Based on the Breusch-Pagan test, 
we reject the null hypothesis of homoskedastic error 
at the 99% level of confidence for both spatial mod-
els.  This is a commonly observed and documented 
phenomenon of spatial regression models. 
 
 
Table 4.  Estimation results for OLS, SAR, and SEM model specifications. 
 
 
           OLS 
             SAR 
              SEM 
Variable 
Coefficient 
p-value 
Coefficient 
p-value 
Coefficient 
p-value 
Constant 
10.51091** 
(0.000) 
10.42543** 
(0.000) 
10.55953** 
(0.000) 
AREALOT 
3.7870E-06** 
(0.003) 
3.3899E-06** 
(0.007) 
4.82604E-06** 
(0.001) 
FRONT 
8.7561E-06** 
(0.001) 
1.0600E-05** 
(0.000) 
7.42170E-06** 
(0.003) 
ATTIC 
0.00014 
(0.067) 
0.00013 
(0.098) 
0.00014 
(0.065) 
BEDROOMS 
0.08504** 
(0.000) 
0.08581** 
(0.000) 
0.07047** 
(0.000) 
FULLBATHS 
0.15120** 
(0.000) 
0.16015** 
(0.000) 
0.14032** 
(0.000) 
HALFBATHS 
0.10480** 
(0.002) 
0.11255** 
(0.000) 
0.10348** 
(0.001) 
AGE 
-0.00698** 
(0.000) 
-0.00672** 
(0.000) 
-0.00643** 
(0.000) 
GARAGE 
0.12720** 
(0.000) 
0.12730** 
(0.000) 
0.12189** 
(0.000) 
FIREPLACES 
0.23866** 
(0.000) 
0.23163** 
(0.000) 
0.20293** 
(0.000) 
CVPOOR 
-0.83248** 
(0.000) 
-0.80300** 
(0.000) 
-0.86721** 
(0.000) 
AIRCOND 
0.36702** 
(0.000) 
0.35357** 
(0.000) 
0.28278** 
(0.000) 
HEATING 
0.31162 
(0.062) 
0.26759 
(0.107) 
0.31981* 
(0.035) 
EDUCATION 
0.01346** 
(0.000) 
0.01360** 
(0.000) 
0.01346** 
(0.000) 
EXP-DISTBR 
-0.22867* 
(0.026) 
-0.22844* 
(0.025) 
-0.25361* 
(0.030) 
DISTHW 
-6.6047E-06* 
(0.017) 
-5.6457E-06* 
(0.041) 
-6.7431E-06* 
(0.029) 
ߩ 
 
 
0.01007** 
(0.001) 
 
 
ߣ 
 
 
 
 
0.32641** 
(0.000) 
** significant at the 99% confidence level; * significant at the 95% confidence level. 
 
The univariate global Moran's I test statistic of 
0.4919 is statistically significant at the 99% level  
(p-value = 0.001), which indicates the presence of 
spatial dependence in housing prices.  This is ex-
pected, as the price of a residential property is not 
solely determined through an array of housing and 
neighborhood characteristics, but also depends on 
the prices of neighboring properties.  As for the OLS 
estimators, they are biased and inconsistent in the 
presence of spatial correlation.  Accordingly, the 
discussion of our regression results will focus on the 
spatial model specifications.  The presence of spatial 
dependence is further supported as the additional 
spatial parameters ߩ = 0.01007 and ߣ = 0.32641 in the 
SAR and SEM specifications are both statistically 
significant at the 99% level (Table 4).  The relatively 
low ߩ value further implies that, though highly sig-
nificant, the spatial spill-over effect of neighboring 
residential properties is relatively small.  This find-
ing can be explained by the fact that residential 
properties around brownfield sites do not show 
large variations in their values.  Our sample data 
indicate that 80% of the property values fall between 
$30,464 and $163,800, compared to $67,500 to 
$319,970 citywide.  As a matter of fact, most proper-
ties included in our study lie in the Mill Creek  


===== PAGE 10 (score 33, numeric-paren 0, table-words 6, var-words 21) =====

232 
  
Mihaescu and vom Hofe 
Corridor, an old and depressed industrial area 
known for its low property values. 
The fact that the SAR specification contains a 
spatial lag of the dependent variable requires an ad-
ditional step to allow interpretation of the coeffi-
cients from the model.  We follow LeSage and Pace 
(2009), who suggest estimating the summary 
measures of the total impacts of each explanatory 
variable on property values as: 
 
ߚ்ൌሺ1 െߩሻିଵߚ 
(7) 
 
The estimated total impacts are shown in Table 5 
below. 
 
Table 5.  Calculated total SAR impacts. 
 
Variable 
Estimated SAR 
coefficients 
Total SAR 
impacts 
Constant 
10.42543 
10.53148 
AREALOT 
3.3899E-06 
3.4244E-06 
FRONT 
1.0600E-05 
1.0708E-05 
ATTIC 
0.00013 
0.00014 
BEDROOMS 
0.08581 
0.08668 
FULLBATHS 
0.16015 
0.16178 
HALFBATHS 
0.11255 
0.11369 
AGE 
-0.00672 
-0.00679 
GARAGE 
0.12730 
0.12860 
FIREPLACES 
0.23163 
0.23399 
CVPOOR 
-0.80300 
-0.81117 
AIRCOND 
0.35357 
0.35717 
HEATING 
0.26759 
0.27031 
EDUCATION 
0.01360 
0.01374 
EXP-DISTBR 
-0.22844 
-0.23076 
DISTHW 
-5.6457E-06 
-5.7031E-06 
 
Most of the variables in each of the three model 
specifications are statistically significant at the 99% 
level of confidence.  The exception is the attic square 
footage (ATTIC) variable, which is only significant at 
the 10% level in each of the three models.  Further, 
the presence of a heating system (HEATING) ap-
pears to be significant at the 95% level in only the 
SEM model specification.  In addition, both distance 
measures, i.e., distance to brownfield sites and 
highways, are significant at the 95% level as well. 
The interpretation of the spatial dependence in 
the SAR specification is a structural process, based 
on the assumption of omitted variables (LeSage and 
Pace, 2009).  As such, the SAR approach does ac-
count to some extent for these unobservable latent 
influences, such as noise and crime issues and other 
excluded variables.  The total SAR impacts in Table 
5 show that one additional bedroom adds 8.67% to 
the market value of a house, while a full bath and a 
half bath add 16.18% and 11.37%, respectively.  Hav-
ing a central air conditioning system increases the 
property value by 35.72%, while a heating system 
adds 27.03% and a garage adds 12.86%.  As  
expected, the age of the property and educational 
attainment level of the residents impact residential 
property prices to a lesser extent.  An increase in the 
age of the house by one year causes a decrease in the 
property value by only 0.68%, and a 1% increase of 
the population aged 25 or older holding a bachelor's 
degree or higher by ZIP code increases the property 
values by 1.37%.  Also of note is that properties in 
very poor and unlivable condition are devalued by 
81.12%. 
The total estimated impact with respect to the 
distance from a property to its closest brownfield  
is -0.2308.  Relying on a non-linear distance function, 
as shown in Figure 2, means that the estimated  
decrease in 
property value 
with 
respect 
to  
the distance to the closest brownfield site needs to 
be 
calculated 
using 
the 
exponential 
form 
݂ሺ݀݅ݏݐܾݎሻൌ݁ݔ݌ሺି೏೔ೞ೟್ೝ
ఱబబሻ%, where distbr denotes the 
distance to the closest brownfield in feet. 
Referring to Figure 2 we see that decreasing the 
maximum distance allowed to a brownfield site, i.e., 
from 1,000 to 0 feet in our study, is equivalent to a 
reduction 
of 
the 
exponential 
function 
of 
݁ݔ݌ሺିଵ଴଴଴/ହ଴଴ሻ to ݁ݔ݌ሺି଴/ହ଴଴ሻ, which is equivalent to a 
reduction from 0.1353 to 1.00 on the vertical axis.  
This translates into a maximum possible reduction 
in property value of ሺ1 െ0.1353ሻ∗23.08% or 
19.96%.  In other words, given that the maximum 
change in the explanatory variable is ሺ1 െ0.1353ሻൌ
0.8647, the largest change possible in a residential 
property value is 19.96%.  The relationship between 
selected changes in actual distances in feet to the 
closest brownfield site, the equivalent changes in the 
exponential function, and the reductions in property 
values are shown in Table 6 below. 
As indicated in Table 6, reducing the distance 
from 500 to 0 feet results in a decrease of residential 
property values of 14.59%.  Analogously, reducing 
the distances from 1,000 to 500 feet, from 500 to 100 
feet, and from 100 to 0 feet result in property value 
reductions of 5.37%, 10.40%, and 4.18%, respectively.  
As indicated in Figure 2, the largest change in prop-
erty values occurs in close proximity to the brown-
field sites, while reductions in property values level 
off when moving beyond the 1,000 feet cut-off dis-
tance used in our study.  For instance, reducing the