# RECORD 9848e68efcee710bc05a55f8576e298c00f7e04b
# TITLE: The effect of remediating PCB-contaminated sediments on home prices in Milwaukee, Wisconsin, USA
# YEAR: 2022.0  VENUE: Agricultural and Resource Economics Review  DOI: 10.1017/age.2022.26
# AUTHORS: R. Wilkinson; Richard T. Melstrom
# PAGES: 18  CANDIDATE TABLE PAGES (1-based): [9, 10, 11, 17, 18]
# ABSTRACT: Abstract This paper estimates the effect of environmental remediation on housing prices in the Milwaukee Estuary Area of Concern (AOC) using a hedonic analysis of individual arms-length sales before and after a major remediation action between 2011 and 2015. Our design leverages this before-after comparison as well as the proximity of homes to remediation and the AOC boundary. Measuring the effect of AOCs in the housing market has always been a difficult task, given that water in AOCs can provide a mix of amenities and disamenities. Indeed, we find little evidence of a negative proximity effect when applying hedonic analysis to cross-section data. However, when we apply the analysis to a repeated cross section in a quasi-experimental framework, we find statistically significant evidence that living near the affected part of the AOC became more desirable after cleanup.


===== PAGE 1 (score 9, numeric-paren 0, table-words 0, var-words 9) =====

RESEARCH ARTICLE
The effect of remediating PCB-contaminated
sediments on home prices in Milwaukee,
Wisconsin, USA
Robin Wilkinson and Richard T. Melstrom*
Loyola University Chicago, Chicago, IL 60660, USA
*Corresponding author: Email: rmelstrom@luc.edu
(Received 27 May 2022; revised 9 November 2022; accepted 11 November 2022; first published online 22
December 2022)
Abstract
This paper estimates the effect of environmental remediation on housing prices in the
Milwaukee Estuary Area of Concern (AOC) using a hedonic analysis of individual arms-
length sales before and after a major remediation action between 2011 and 2015. Our design
leverages this before-after comparison as well as the proximity of homes to remediation and
the AOC boundary. Measuring the effect of AOCs in the housing market has always been a
difficult task, given that water in AOCs can provide a mix of amenities and disamenities.
Indeed, we find little evidence of a negative proximity effect when applying hedonic analysis
to cross-section data. However, when we apply the analysis to a repeated cross section in a
quasi-experimental framework, we find statistically significant evidence that living near the
affected part of the AOC became more desirable after cleanup.
Keywords: areas of concern; Great Lakes; hedonic method; sediment remediation
JEL Codes: Q25; Q53; Q58
Introduction
Areas of Concern (AOCs) were established in 1987 as part of the Great Lakes Water
Quality Agreement (GLWQA) between the United States and Canada. AOCs are coastal
locations around the Great Lakes region that have been highly degraded from human
activity to a point where water-related beneficial uses are significantly impaired
(EPAa). The goal of the GLWQA’s AOC program is to restore AOCs by removing benefi-
cial use impairments (BUIs), which can range from restrictions on fish and wildlife
consumption, to beach closings, to degradation of esthetics; BUIs are directly experienced
by those living in the vicinity of AOCs. An AOC can be delisted once all BUIs associated
with it have been removed. To accomplish this, AOCs have been the targets of extensive
and expensive remediation efforts. Between 1985 and 2019, the United States and Canada
spent $22.78 billion on AOC remediation efforts (Hartig et al. 2020). With currently 6 of
the original 43 AOCs delisted, substantial restoration work remains to be done.
© The Author(s), 2022. Published by Cambridge University Press on behalf of the Northeastern Agricultural and Resource
Economics Association. This is an Open Access article, distributed under the terms of the Creative Commons Attribution
licence (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use, distribution and reproduction,
provided the original article is properly cited.
Agricultural and Resource Economics Review (2023), 52, 71–88
doi:10.1017/age.2022.26
https://doi.org/10.1017/age.2022.26 Published online by Cambridge University Press


===== PAGE 9 (score 31, numeric-paren 0, table-words 3, var-words 22) =====

and the census tract shapefile for the City of Milwaukee to determine the tract associated
with each property. Homes sales occurred in 213 tracts.
Results and discussion
Table 2 presents the main coefficients of interest in the three models. For a complete set of
coefficient estimates, see the Appendix. These coefficients measure the percent change in
sales price associated with a 1% increase in distance (a decrease in proximity) to the AOC,
depending on whether a home is closest to the Milwaukee, Menominee, or Kinnickinnic
Rivers, or the treated area. For the sake of brevity, we will not describe the other coeffi-
cients in the model (i.e. on the structural attributes and lake and downtown distance vari-
ables), except to note that the estimates are in-line with expectations and consistent with
prior research. Model (1) provides no evidence that proximity affects price. In fact, the
Table 1. Summary statistics of Milwaukee home sales in the pre- and postproject phases
Variable
Mean
Std.
dev.
saleprice
Sales price
167051 121629
finsqft
Total finished square footage in thousands
1.584
0.830
acres
Acreage of lot
0.097
0.067
bdrms
Number of bedrooms
2.963
1.330
hbath
Number of half bathrooms
0.250
0.465
fbath
Number of full bathrooms
1.480
0.642
age
Age of property (2018 – year built)
77.666
32.510
distance
Straight-line distance to nearest part of AOC (100 s of meters)
20.894
14.784
distance to LP
Straight-line distance to Lincoln Park (100 s of meters)
93.029
39.304
downtown
Straight-line distance to downtown (100 s of meters)
55.652
28.414
lake
Straight-line distance to Lake Michigan (100 s of meters)
43.091
30.717
Dummy Variables
Count
Mean
ranch
Ranch-style house
444
0.113
colonial
Colonial-style house
199
0.051
capecod
Cape Cod-style house
650
0.166
condo
Condominium
851
0.217
Milwaukee
Nearest the Milwaukee River in the AOC
1539
0.392
Menomonee
Nearest the Menomonee River in the AOC
1328
0.338
Kinnickinnic
Nearest the Kinnickinnic River in the AOC
1058
0.267
Treated
Nearest the Milwaukee River segment with water quality
improvements, downstream of the remediation at Lincoln Park
833
0.212
Total sales
3,925
Agricultural and Resource Economics Review
79
https://doi.org/10.1017/age.2022.26 Published online by Cambridge University Press


===== PAGE 10 (score 40, numeric-paren 22, table-words 6, var-words 3) =====

coefficients on ln distance


 × Menominee and ln distance


 × Kinnickinnic are negative,
which runs contrary to expectations, though neither is statistically significant. The
coefficient on ln distance


 × Milwaukee is also not significantly different from zero.
Now consider the coefficients in model (2), which are estimated using both the
pre-
and
postproject
data.
The
coefficients
on
ln distance


 × Menominee
and
ln distne


 × Kinnickinnic are significantly negative, which implies that homebuyers place
a premium on living near the Menominee and Kinnickinnic Rivers. The coefficients on
ln distance


 × Milwaukee and ln distance to LP


 distances are both statistically insignifi-
cant. However, the coefficient on ln distance to LP


 × Post is significantly negative.
This indicates that a proximity premium for homes near Lincoln Park developed after
remediation.
Finally, consider the coefficients of model (3), which are also estimated using the
pre-
and
postproject
data.
The
coefficients
on
ln distance


 × Menominee
and
ln distance


 × Kinnickinnic are significantly negative, similar to model (2). The coefficient
on ln distance


 × Milwaukee is not significantly different from zero. However, the
Table 2. Hedonic models – coefficients of interest
Model
(1)
(2)
(3)
ln distancei


×Menomineei
−0.0275
−0.0705*
−0.0789**
(0.0707)
(0.0361)
(0.0366)
ln distancei


×Kinnickinnici
−0.0551
−0.0923**
−0.1006**
(0.0701)
(0.0364)
(0.0368)
ln distancei


×Milwaukeei
0.0023
0.0069
0.0448
(0.0654)
(0.0355)
(0.0374)
Postt
−0.0141
−0.0141
(0.0122)
(0.0130)
ln distance to LPi



0.0143
(0.0177)
ln distance to LPi


×Postt
−0.0231*
(0.0129)
ln distancei


×Treatedi
−0.0532*
(0.0295)
ln distnci


×Treatedi×Postt
−0.0231*
(0.0139)
Housing structure characteristics
Yes
Yes
Yes
Month fixed effects
Yes
Yes
Yes
Tract fixed effects
Yes
Yes
Yes
R-squared
0.8296
0.7966
0.7971
Observations
1,033
3,925
3,925
Standard errors in parentheses below coefficients. * and ** indicate significance at 10% and 5% levels, respectively.
80
Robin Wilkinson and Richard T. Melstrom
https://doi.org/10.1017/age.2022.26 Published online by Cambridge University Press


===== PAGE 11 (score 45, numeric-paren 25, table-words 6, var-words 5) =====

coefficient on ln distance


 × Treated is significantly negative, which indicates a price
premium for homes in the treated area relative to those in proximity to other parts of
the Milwaukee River. Turning to the effect of remediation in the treated area, the coeffi-
cient on ln distance


 × Treated × Post is significantly negative. To be specific, the coef-
ficients on ln distance


 × Treated and ln distance


 × Treated × Post imply that before
remediation a 100% increase in distance (e.g. 0.5 km to 1 km) from the AOC (in the treated
area) was associated with a 5.3% reduction in price, which is $8,880 at the average price,
and that after remediation the same change was associated with a 7.6% reduction, which is
$12,740 using the average sale price for properties in our analysis (Table 1).
Robustness checks
We performed three robustness checks to test the sensitivity of the estimates to the defi-
nition of the treated area, controlling for time-varying unobservables, and the timing of
capitalization effects. These results are shown in Table 3. The first robustness check revises
the definition of the treated area in model (3) by reducing the area from 8 to 5 km
Table 3. Robustness checks applied to model (3)
Treated area 5 km
downstream
Add tract-by-post fixed
effects
Mature-project
ln distancei


×Menomineei
−0.0797**
−0.0795**
−0.0257
(0.0393)
(0.0379)
(0.0311)
ln distncei


×Kinnickinnici
−0.1008**
−0.0966**
−0.0573*
(0.0391)
(0.0380)
(0.0318)
distancei


×Milwaukeei
0.0282
0.0730*
0.0771**
(0.0367)
(0.0382)
(0.0318)
Postt
0.0282
–
–
(0.0367)
–
–
ln distancei


×Treatedi
−0.0300
−0.0239
−0.0048
(0.0529)
(0.0432)
(0.0407)
ln distancei


×Treatedi×Postt
−0.0303**
−0.0718
−0.1078**
(0.0133)
(0.0471)
(0.0414)
Housing structure
characteristics
Yes
Yes
Yes
Month fixed effects
Yes
Yes
Yes
Tract fixed effects
Yes
Yes
Yes
Tract-by-post fixed effects
No
Yes
Yes
R-squared
0.7968
0.8079
0.8124
Observations
3,925
3,925
4,437
Note that the effect of Postt is not identified when the regression includes tract-by-post effects.
Standard errors in parentheses below coefficients. * and ** indicate significance at 10% and 5% levels, respectively.
Agricultural and Resource Economics Review
81
https://doi.org/10.1017/age.2022.26 Published online by Cambridge University Press


===== PAGE 12 (score 20, numeric-paren 0, table-words 6, var-words 8) =====

downstream of Lincoln Park. These results are shown in the first column of Table 3. This
results in a more precisely estimated, larger (in absolute value) treatment effect than in
Table 2. This should increase confidence that remediation affected sales prices, particularly
for homes just downstream of Lincoln Park. The second column returns to the
original definition of the treated area and adds tract-by-post fixed effects to correct for
the influence of time-varying neighborhood characteristics that could potentially be corre-
lated with the timing and location of remediation (e.g. major capital improvements or
changes in land use). The addition of these fixed effects increases the coefficient on
ln distance


 × Treated × Post; however, the estimate is less precise and no longer signifi-
cantly different from zero at conventional levels. Finally, the last robustness check uses an
alternative pre-post comparison, by replacing the 2,892 postproject sales with 3,404
“mature-project” sales that occurred between August 2017 and July 2018, in case the effect
of cleanup takes longer to capitalize than a single year after the project ends. With tract-by-
post fixed effects, the coefficient on ln distance


 × Treated × Post is significantly negative
and substantially larger relative to the earlier estimates. This suggests the effect of reme-
diation could take several years after a project end to fully materialize.4
Additional discussion
The results indicate that hedonic analysis based on distance cannot always estimate the
effect of an AOC on home prices. The coefficients on the proximity variables in model
(1) provide no evidence that homebuyers either discount or place a premium on a home’s
proximity to the Milwaukee Estuary AOC. The variation in sign between different rivers in
the AOC adds further uncertainty as to the actual proximity effect. Prior research has
found that home values tend to rise with distance (Braden et al. 2008a, Braden et al.
2008b), a result which we are unable to replicate. This does not mean that buyers are indif-
ferent to the presence of AOCs. Rather, our results suggest that, at least in some housing
markets, the proximity effect of an AOC can be tempered or entirely offset by amenities
near the water, as appears to be the case in Milwaukee.
The overall effects of proximity notwithstanding, the results imply that buyers care
about the presence of PCBs and remediation because cleaning up part of the
Milwaukee Estuary led to capital gains that increased with proximity. Both the coefficient
on distance to remediation at Lincoln Park in model (2) and the coefficient on distance to
the downriver segment in model (3) were significantly negative, implying that living closer
to these areas became more valuable after remediation. We can use the change in the price
gradient associated with the cleanup effect to estimate the capital gain from remediation,
using the formula 100 × βTP
ln
distancei  ln
dig
stance
~
, where dig
stance
~
is the hypo-
thetical boundary distance of the effect. Given we trimmed the data to sales within
5 km, and measured distance in hundreds of meters, dig
stance
~
= 50. Using the coefficients
in model (3) (the estimates are similar using model (2)), for a property in the treated area
adjacent to the water, which we define as within 150 m, the implied capital gain is 8%, or
$13,538 using the average property sales price. For properties affected by the improvement
but located 3 km away, the implied capital gain is just 1%, or $1,972. Across all homes in
the treated area, the implied capital gain is 3%, or $5,683. Note that the gain is larger if we
were to use the coefficients in the robustness checks. For example, the implied capital gain
4An important caveat, though, is that a dam in the treated area was removed in 2018, and the mature-
project estimates may pick up the effect of removal. The direction of any bias, though, is unclear, because
removal was expected to improve water quality and restore habitat (a benefit) but reduce water levels for
upstream recreation (a potential loss).
82
Robin Wilkinson and Richard T. Melstrom
https://doi.org/10.1017/age.2022.26 Published online by Cambridge University Press


===== PAGE 17 (score 67, numeric-paren 48, table-words 1, var-words 8) =====

Appendix
Table A1. Hedonic models – all coefficients
Model
(1)
(2)
(3)
Finsqft
0.283044
0.280172
0.280050
(0.024368)
(0.010226)
(0.010258)
Bdrms
−0.039027
−0.037703
−0.037248
(0.016597)
(0.007050)
(0.007058)
Fbath
0.175602
0.141640
0.140987
(0.026654)
(0.011502)
(0.011494)
Hbath
0.126607
0.124649
0.125128
(0.026258)
(0.011963)
(0.011927)
Age
−0.012805
−0.011156
−0.011028
(0.002421)
(0.001245)
(0.001255)
Age2
0.000076
0.000067
0.000066
(0.000016)
(0.000008)
(0.000008)
Waterfront
4.498826
0.987184
0.980320
(1.325469)
(0.383445)
(0.383740)
Acres
−11.702370
−1.001644
−1.014829
(3.848793)
(0.705101)
(0.698689)
Acres2
0.159765
0.142691
0.142736
(0.038555)
(0.024794)
(0.017641)
Cape Cod
0.153071
0.186078
0.188487
(0.051420)
(0.017676)
(0.024864)
Colonial
0.145395
−0.003397
−0.007865
(0.1270)
(0.052669)
(0.052892)
Condo
0.120879
0.081467
0.081110
(0.048037)
(0.021277)
(0.021237)
Ranch
−0.070964
0.142348
0.162981
(0.084647)
(0.049215)
(0.049872)
DistancetoDowntown
−0.256703
−0.322434
0.241554
(0.070046)
(0.038786)
(0.092125)
DistancetoLake
0.260477
0.211159
0.241554
(0.130816)
(0.088278)
(0.092125)
ln distance


×Menomineei
−0.027456
−0.070494
−0.078889
(0.070704)
(0.036149)
(0.036620)
(Continued)
Agricultural and Resource Economics Review
87
https://doi.org/10.1017/age.2022.26 Published online by Cambridge University Press


===== PAGE 18 (score 20, numeric-paren 12, table-words 1, var-words 4) =====

Table A1. (Continued )
Model
(1)
(2)
(3)
ln distancei


×Kinnickinnici
−0.055137
−0.092289
−0.100555
(0.070136)
(0.036414)
(0.036838)
ln distncei


×Milwaukee
0.002291
0.006906
0.044785
(0.065402)
(0.035543)
(0.037371)
Postt
−0.014117
−0.014066
(0.012180)
(0.013048)
ln distance to LPi



0.014304
(0.017707)
ln distance to LPi


×Postt
−0.023126
(0.012902)
ln distancei


×Treatedi
−0.053156
(0.029504)
ln distancei


×Treatedi×Postt
−0.023111
(0.013905)
Month fixed effects
Yes
Yes
Yes
Tract fixed effects
Yes
Yes
Yes
Cite this article: Wilkinson, R. and R. T. Melstrom (2023). “The effect of remediating PCB-contaminated
sediments on home prices in Milwaukee, Wisconsin, USA.” Agricultural and Resource Economics Review 52,
71–88. https://doi.org/10.1017/age.2022.26
88
Robin Wilkinson and Richard T. Melstrom
https://doi.org/10.1017/age.2022.26 Published online by Cambridge University Press