Академический Документы
Профессиональный Документы
Культура Документы
R C
.
value differential, V yV is greater than or equal to zero.
REDEVELOPMENT AND CONTAMINATION RISK 417
The first empirical exploration of the optimal redevelopment rule was
w x
undertaken by Rosenthal and Helsley 12 in an examination of the
redevelopment of single family dwellings in Vancouver, BC. In addition to
finding strong empirical support for this theory in residential redevelop-
ment, Rosenthal and Helsley argue that if net demolition costs are zero,
then a new method for determining the value of vacant land emerges that
avoids some of the problems of hedonic estimation methods. These co-
authors also present the most generalized expression of the optimal
redevelopment rule: that the marginal costs of delaying redevelopment
must equal or exceed the marginal benefits of delaying redevelopment.
Incorporating the assumptions that:
.
1 landowners are myopic; that is, they act as if the level of income
obtainable from redevelopment will remain constant into the future,
.
2 capital costs are constant,
.
3 no structural depreciation occurs, and
.
4 demolition costs are zero,
the optimal redevelopment rule reduces to the form derived by Brueckner
and Wheaton.
w x
Munneke 10 presents empirical support that the optimal redevelop-
ment rule describes the redevelopment of commercial and industrial
properties in the City of Chicago during the decade of the 1980s. Addition-
ally, Munneke asserts that the assumption of a zero demolition cost might
not be appropriate for industrial properties and makes an argument that if
demolition costs are a significant component of the redevelopment deci-
sion, then these costs must be incorporated into the redevelopment rule.
In Munnekes analysis, the structural demolition costs are incorporated
R C
.
outside the value differential term V yV , and the parcels capital-to-
land ratio SrL is found to be a statistically significant proxy variable for
the structural demolition costs.
2.B. Contamination Liability as Demolition Costs
Assume that the contamination liability for a given industrial parcel can
be modeled as two separate demolition costs. The first is associated with
the existing structure, as would be the case for asbestos or lead contamina-
tion. The second is associated with the land, as would be the case for PCB
w x
or heavy metal soil contamination. As identified by Munneke 10 , the
structural demolition cost D
S
would be a function of the existing capital-
to-land ratio for the parcel, SrL. However, it makes intuitive sense that
structures of a certain vintage andror prior industrial use would have a
higher likelihood of contamination and thus have a higher structural
demolition cost. Increases in demolition cost associated with any possible
DANIEL T. MCGRATH 418
structural contamination would be readily apparent to a redeveloper
through inspection of the existing structure. Therefore, in addition to
being a function of the parcels capital-to-land ratio, the structural demoli-
tion cost would also be a function of the conditionrdepreciation level d, or
S S
.
D sD SrL, d .
The estimation of remediation costs associated with land contamination
is more uncertain for a redeveloper, and this fact is at the center of the
brownfields debate. However, over the 19 years since the passage of
CERCLA,
2
as industrial redevelopers have gained more experience with
the legal requirements and subsequent financial implications of this act,
one might reasonably expect that contamination liability would, over time,
become to some extent predictable and capitalized into parcel bid value.
This paper takes the view that there exists an observable environmental
variable E that can accurately determine a priori the land contamination
liability associated with a given parcel to be redeveloped. Thus, the land
demolition costs would become a function of the environmental variable,
L L
.
or D sD E .
2.C. Value Equations and Functional Forms
Every industrial location can be represented by a quantity of industrial
real estate R, which is a function of the land area L and the amount of
w x
capital on the site, S 8 . The value of any given parcel of industrial real
C
0
q
C
X
C
i
.
s =R L , S , 3 .
.
i i
i
where P
C
is the price per real estate unit of industrial real estate in its
current use, R is the units of industrial real estate, L is the amount of land
area, S is the measure of existing capital in units of floor space, i is the
interest rate, and X
C
is the vector of explanatory variables other than land
i
2
The Comprehensive Environmental Response, Compensation, and Liability Act, other-
wise known to as Superfund.
REDEVELOPMENT AND CONTAMINATION RISK 419
and capital for the ith current-use parcel that determines the temporal
and spatial variations in the unit real estate price within the urban area.
Another simplifying assumption is that the elasticity of substitution of L
for S is equal to 1, which allows the use of a CobbDouglas functional
form, or R sL
variations in the parcels unit price and is not necessarily the same vector
C
.
of variables that determine V , L is the land area of the parcel, S rL is
i i i
the existing capital-to-land ratio, d is some measure of the depreciation of
i
the existing capital on the parcel, and E is the a priori measure of land
i
contamination for the parcel. Assuming a log-linear specification for the
function g gives
R
V
i
R R
ln s q Y , 9 .
0 i
L
i
where Y
R
is defined as the full vector of independent variables expressed
i
.
on the left side of Eq. 8 . This approach is justified as a Taylor series
. .
expansion of Eq. 7 . In the actual estimation of Eq. 9 , the expansion is
carried out as far as it has explanatory power. The rationale here is to
linearize and produce an easily estimatable and comparable functional
expression for V
R
that can be utilized in the two-stage method to correct
for possible selection bias between the redeveloped and current-use prop-
erties and thus obtain consistent estimates for the value differential,
R C
.
V yV , for each parcel in the dataset.
3. ECONOMETRIC METHODOLOGY
The industrial redevelopment decision can be tested empirically through
the use of a structural probit model. Following closely the estimation
w x w x
procedure used by Rosenthal and Helsley 12 and by Munneke 10 , this
model is written as
R
s V
R
yV
C
q i s1, 2, . . . , N, 10 .
.
i i i i
where
R
represents the criterion function or index which indicates when
i
R C
.
redevelopment will take place upon sale, V yV is the parcels value
i i
2
.
differential, and is the error term which is assumed to be N 0, . A
i
parcel is redeveloped upon sale if
R
)0 and remains in its current use if
i
R
F0. In the probit estimation over the entire sample,
R
is assigned
i i
the value of 1 for parcels that were redeveloped upon sale and 0 for
REDEVELOPMENT AND CONTAMINATION RISK 421
parcels remaining in their current use. If the criterion for redevelopment is
determined by a parcels value differential, the probit estimation proce-
dure will identify the coefficient, , as significant and of a positive value.
.
R
To accomplish the probit estimation of Eq. 10 , estimates for both V
and V
C
are required to calculate the value differential for each parcel.
However, only one value, either V
R
or V
C
, is ever observed at the time of
a parcels sale. It is assumed that in a competitive land market, the sale
prices of industrial parcels sold for redevelopment will reveal V
R
and that
the sale prices for current-use parcels will reveal V
C
. The estimation of
both V
R
and V
C
for each parcel is accomplished by separating the dataset
into these two groups and then regressing the observed sale prices on the
vectors of observable spatial and temporal variables that determine value
for each group respectively. Using the estimated coefficients from each
value function, an estimate for the value differential is thus obtained for
each parcel in the combined dataset.
If one assumes that these estimated value equations take on a linear
form, then the value differential can be written as
V
R
yV
C
s
R
Y
R
y
C
X
C
q y , 11 . .
.
i i i i i i
where Y
R
and X
C
are the vectors of the observable spatial and temporal
i i
characteristics determining redeveloped and current-use value, respec-
tively,
R
and
C
are the respective coefficients of the redeveloped and
current-use value equations, and the error terms and are assumed to
i i
2
.
2
.
be ;N 0, and ;N 0, .
i i
. .
Substituting Eq. 11 into the original structural probit model 10 , the
reduced-form probit model is derived as
sZ y , 12 .
i i i
where is the vector of all coefficients, Z is the vector of the union of all
i
.
explanatory variables, and the error term s y q .
i i i i
.
The estimation of the structural probit equation 10 follows the two-
w x w x
stage procedure developed by Lee 5 and outlined by Maddala 7 . This
method involves the substitution of estimated endogenous variables into
the criterion function prior to its estimation via a probit analysis. The first
step in this procedure is to use the probit method to estimate the reduced
.
form probit equation 12 to obtain estimates for based on the dichoto-
mous observations of Rthat is, whether or not the ith parcel was
redeveloped.
The second step is to estimate the respective redeveloped and current-use
value functions in a way that produces unbiased estimates for the coeffi-
cients
R
and
C
. As discussed above, only one value, either V
R
or V
C
,
DANIEL T. MCGRATH 422
will ever be observed when a parcel is sold. However, the original criterion
.
function, Eq. 10 , requires the calculation of the value differential for all
observations in the dataset. The individual value functions for V
R
and V
C
are estimated separately by regressing parcel sale prices on the vectors of
observable spatial and temporal characteristics for each parcel Y
R
and X
C
,
i i
respectively. Since the data have been divided into two separate groups,
unobserved characteristics favorable to each group may exist, and selection
bias becomes a possibility. Expressed more formally, when selection bias is
present, the conditional expectation of the error terms of the value
equations will not be zero. Consistent and unbiased estimates of the value
equation parameters are obtained by defining new error terms,
U
and
i
U
, and by the introduction of new estimated explanatory variables W
R
i i
and W
C
, the Mills ratios, into their respective equations. By including the
i
Mills ratios into the value functions, the conditional expectation of both
U
i
and
U
is now zero. The values of W
R
and W
C
are computed for each
i i i
observation in the combined dataset using the estimates of obtained
.
from the initial probit estimation of the reduced form probit equation 12
identified in step 1. The respective redeveloped and current-use value
equations to be estimated are written as
z .
i
U
R R R
V s Y y q
i i i
/
z .
i
s
R
Y
R
y W
R
q
U
13 .
i i i
and
z .
i
U
C C C
V s X q q
i i i
/
1 y z .
i
s
C
X
C
q W
C
q
U
, 14 .
i i i
where
U
s q W
R
,
U
s y W
C
, is the covariance be-
i i i i i i
tween and , is the is the covariance between and , is the
Diff
s V q ,
.
i i i
Diff
R R
C C
where V sV yV s Y y X . 15 .
i i i i i
4. DISCUSSION OF DATA
The principal dataset is a group of 195 redeveloped and current-use
property sale transactions that occurred within the City of Chicago from
August 1983 through November 1993. These records were obtained from
the City of Chicagos Harris-REDI database maintained by the Citys
Department of Planning and Development. The Harris-REDI database is
a combination of the Citys Harris Land-use Data with land sale transac-
.
tions compiled by Real Estate Data, Inc. REDI , which have been
gathered from all real estate transfer declarations filed with Cook County.
The redeveloped property data subset is a group of 95 redeveloped
industrial properties. This subset was created by matching the 8043 indus-
trial land sale transactions with 1867 industrial building permit records and
the Citys 881 industrial demolition permit records that were issued by the
City of Chicago Department of Buildings from January 1984 through
December 1993. Information concerning the neighborhood racial composi-
tion for the parcels in the database was obtained by matching the records
with 1990 census information.
The criterion used to identify whether a parcel was purchased for
redevelopment is whether the sold parcel had an industrial demolition
permit andror an industrial building permit filed with the Department of
Buildings within 24 months of the sale date and is currently zoned for
industrial use. If so, then the property is considered to have been pur-
chased for redevelopment. Including industrial building permits in addition
to industrial demolition permits produces a softer redevelopment rule.
Often, existing structures are not completely removed by the new user, but
through additional construction, the existing structure is substantially
modified for a new use. This situation falls somewhere in between the
DANIEL T. MCGRATH 424
hard redevelopment criteria in which all existing capital is removed and
the case where a parcel is purchased for current use and remains essen-
tially unchanged. The intention here is to identify those industrial parcels
that show evidence of investment by a redeveloper, whether through
complete demolition or through modification of the existing structure.
The salerpermit address matching was verified by visual investigation of
each parcel in the Sanborn Fire Insurance Maps of the City of Chicago.
Additionally, to ensure that the identified address match represents a
property that is currently in industrial use and was not purchased for
conversion to either residential or commercial use, the current zoning of
each property was verified as industrial by site identification in the 1994
Chicago Zoning Ordinance zoning maps.
The current-use group of sales is a set of 100 observations chosen
randomly from the industrial sale record. These properties were verified to
have no building or demolition permits associated with them any time after
their sale and are also currently zoned for manufacturing.
The environmental variable E, which has been hypothesized to be a
significant determinant of the land contamination liability faced by a
buyerrredeveloper and subsequently a determinant of the land demolition
costs, is represented by the a priori probability of the parcels contamina-
tion. This continuous variable was identified for specific land uses by
w x
Noonan and Vidich 11 in a survey of completed cleanups by 17 environ-
mental engineering firms in the northeast U.S. Table 1 outlines the
individual a priori probabilities of contamination for 25 specific categories
of land use. Denoted as PROBCON in the redeveloped and current-use
valuation models, this variable is used to signal higher liability costs
because, at present, there are no publicly available remediation cost
databases which might be used to produce an accurate remediation cost
estimate or to produce statistically significant contamination risk factors.
In an effort to duplicate the process a buyer would have undertaken in a
Phase I environmental investigation, a historical land-use investigation was
undertaken for each of the 195 properties in the database. To determine
this land use, each property was researched in both the 1949 and 1975
versions of the Sanborn Fire Insurance Maps. These maps, available on
microfilm, provide a wealth of land-use information, identifying the spe-
cific company using the site at the time and often the specific nature of the
industrial activity. From this investigation, each property has been assigned
an SIC-code interpretation which best represents the historical land use.
From this SIC-code interpretation, a value of the a priori probability of
contamination has been identified from Noonan and Vidichs classifica-
tion. The definitions and the summary of the statistics of the variables
used in this analysis are presented in Table 2.
REDEVELOPMENT AND CONTAMINATION RISK 425
TABLE 1
The A-priori Probability of Contamination Based on Historical Land Use
Probability of
Commercial and Industrial Land Use Categories Contamination
1. Former coal gas plants, fuel distributors, chemical distributors, .99
airports, incinerators
2. Auto salvage yards, plastic manufacture, electric utility, refining, .95
hazardous waster storagertransfer
3. Oil and other petroleum storage .92
4. Metal plating, landfills, chemical manufacture, metal finishingr .90
tool & dye, laboratories
5. Heavy industrial manufacturing, power plants, paper manufacturing, .88
gas stations
6. Tanneries .87
7. Urban vacantrabandoned land, furniture repair and stripping, .85
circuit board manufacturers, tank farms, waste treatment plants
8. Metal working and fabrication .83
9. Railroad yards and right of ways, vehicle maintenance facilities .82
10. Refuse recycling facilities, machine shops, electronics assembly .80
facilities, agricultural mixersrformulators, high technology
manufacturing
11. Junkyards, electronics manufacture .79
12. Industrial parks, automotive assembly facility, light industrial .75
manufacturing
13. Dry cleaners .74
14. Auto repair shops .72
15. Chemical research facility .70
16. Trucking terminal, textile printing and finishing .65
17. Resource recovery facilities, electricalrplumbingrHVAC service .60
18. Photographic .53
19. Auto dealerships, fabric dyeing establishments, pharmaceutical .50
establishments
20. Highways, research facilities .40
21. Warehouses .35
22. Gas utilities .35
23. Retail property .25
24. Residential, rural vacant property, hospitals .20
.
25. Offices non-manufacturing .13
w x
Source: Noonan and Vidich 11 .
DANIEL T. MCGRATH 426
TABLE 2
Statistical Summary of the Land Use Data
Full Redeveloped Current-Use
Sample Parcels Parcels
. . .
Ns150 Ns95 Ns100
Std. Std. Std.
Mean Dev. Mean Dev. Mean Dev.
LNUP 2.5811 1.1494 2.5150 1.2315 2.6438 1.068
Natural Log of the
Parcel per sq. ft.
sale price
LNA 10.3329 1.1841 10.5885 1.1977 10.09 1.1239
Natural Log of the
land area in sq. ft.
LNSPACE 9.9283 2.9283 9.6997 3.9631 10.1455 1.3391
Natural Log of the
building area in
sq. ft.
LAND 69,341 142,584 82,905 158,991 56,546 124,471
Parcel land area in
square feet
CBD 5.6384 2.8056 5.7687 2.8966 5.5146 2.7251
Distance in miles
from the central
business district
intersection of
LaSalle and
.
Jackson Streets
LNORTHD 0.5538 0.4984 0.6316 0.4849 0.4800 0.5021
1 if the parcel is
north of Lake St.,
0 otherwise
AGE 54.33 23.74 54.29 23.98 54.37 23.65
Age of the building
in years
COND 1.5897 0.6255 1.5158 0.5809 1.6600 0.6547
Condition Code of
the building based
on external survey
by City of Chicago:
1 sexcellent,
2 sgood, 3 sfair,
4 spoor
DAYS 2018.9 966.9 1877.3 862.2 2153.3 1043.4
Number of days
from the initial
saledate in
the dataset,
8r1r83
REDEVELOPMENT AND CONTAMINATION RISK 427
TABLE 2Continued
Full Redeveloped Current-Use
Sample Parcels Parcels
. . .
Ns150 Ns95 Ns100
Std. Std. Std.
Mean Dev. Mean Dev. Mean Dev.
PCNTAFAM 0.2953 0.3804 0.2184 0.3197 0.3683 0.4188
Percentage of
African-American
population within
census tract of
parcel
CAPINT 1.4526 2.1187 1.2632 2.0079 1.6324 2.2139
Capital Intensity as
measured by the
building floor area
divided by parcel
land area
PROBCON 0.6778 0.2240 0.6785 0.2248 0.6770 0.2243
The a-priori
probability of
parcel
contamination
based on
historical
land use
VALDIFF y24.493 634.603 190.511 569.232 y228.747 628.497
The estimated value
differential
R C
.
V yV
of the parcel in
000s of 1995$
5. EMPIRICAL RESULTS
5.A. The Reduced Form Probit Results
As previously discussed in Section 3, the two-stage method requires the
.
initial estimation of the reduced form probit equation 12 . Although the
purpose of this initial step is to facilitate the calculation of the Mills ratios
W
R
and W
C
, which are then included as estimated variables in their
respective value functions, the estimation results are by themselves of
particular interest. The estimation results, presented in Table 3, identify
two separate estimations, Model 1 and Model 2. Model 2 includes the
structural and land demolition variables, CAPINT, the capital intensity of
the parcel, and PROBCON, the a priori probability of contamination
DANIEL T. MCGRATH 428
TABLE 3
Estimation Results of the Reduced Form Probit
Model 1 Model 2
Intercept y3.616 y3.650
. .
2.533 2.540
LNA 0.652 0.778
. .
Natural Log of land area in sq. ft. 3.876 3.805
LNSPACE y0.287 y0.414
. .
Natural Log of the building area in sq. ft. 2.499 2.567
LAND y1.743E-06 y1.841E-06
. .
Parcel land area in square feet 1.830 1.915
CBD y0.345 y0.038
. .
Distance in miles from the central business 8.745 0.949
R
.
non-linear relationships to ln V rL . The redeveloped value function
exhibits statistically significant cubic relationships with respect to both
. .
land area LAND and distance from the central business district CBD ,
w x
which is consistent with the results presented by Munneke 10 . The cubic
relationship with respect to CBD identifies a rapid decline in value as
parcel distance from the CBD increases with a local minimum at 3.7 miles
.
Focusing on the results for Model VC2, the coefficient of LNA is negative
and significant, indicating that, for current-use parcels, the per-square-foot
unit value declines with increasing land area. The coefficient of LNA is
.
equal to y1 , where represents the elasticity of land value to land
area in a CobbDouglas functional form. is estimated within the
current-use data group to be equal to 0.269. The coefficient of LNSPACE
in the model is equal to , which represents the elasticity of land value to
floor space. In Model VC2, the coefficient of LNSPACE is positive and
significant and identifies that the elasticity of land value to building space
is equal to 0.426.
Of the locational variables, the influence of CBD and LNORTHD on
unit bid value are significant for current-use parcels, with unit bid value
exhibiting a quadratic relationship with respect to CBD. The average unit
bid value declines to a minimum reduction of about 70% at 7.5 miles from
the CBD, rising thereafter. A north location relative to a south location
gains 35% in unit bid value. Similar to the redeveloped group of parcels,
the sector with highest average unit value is the northwest sector of the
city, outside nine miles from the central business district.
Both of the site-specific explanatory variables, COND and AGE, are
statistically significant. Each unit increase in condition code decreases unit
bid value by 39% and each additional 10 years to the structures age
decreases unit value by about 8%. The racial population variable PC-
NTAFAM, which identifies the percentage of African-American popula-
tion within the census tract of the parcel, also has a large and significant
effect on current-use unit value. A 10% increase in census tract African-
American population reduces parcel unit bid value about 6%. Lastly, the
coefficient of the Mills ratio is not significant, again identifying no selec-
tion bias between the redeveloped and current-use data groups.
5.D. Structural Probit Results
The redeveloped and current-use value equations estimated above are
used to calculate the value differential for each property in the full
dataset. The value differential for each property is the difference in the
estimated value identified by the redeveloped value equation minus the
R C
.
predicted value identified by the current-use value equation, or V yV .
This measurement of estimated parcel value differential is then used in the
.
structural probit model, Eq. 15 , to test the hypothesis that redevelopment
REDEVELOPMENT AND CONTAMINATION RISK 437
occurs when the value of parcel converted to a new use exceeds the
parcels current-use value.
Similar to the previously mentioned situation caused by the inclusion of
.
an estimated variable the Mills ratio in the value equations, since the
probit model is estimated using a variable that is itself estimated from the
value equations, there may be bias in the standard errors of the probit
model. In this probit analysis, the standard errors shown are uncorrected,
so there is a possibility that they may be biased downwards.
7
The results of the various specifications of the structural probit equa-
tions are presented in Table 6. The dependent variable of the probit model
is REDEV, which takes on the value of 1 if a parcel is among the
redeveloped properties and 0 if it is among the current-use properties. The
estimated values of the dependent variable REDEV are interpreted as the
probability of redevelopment occurring contingent upon a sale. The ex-
planatory variable in the structural probit model is VALDIFF, which is the
estimated value differential calculated as the difference between the
predicted redeveloped value and the predicted current-use value for all
observations.
For Probit Model 1 in Table 6, which uses as a measure for VALDIFF
the difference between VR1 and VC1 model specifications that exclude
7
Recognizing the possibility of standard error bias but leaving the standard errors uncor-
rected is the standard approach to this problem, due to the complexity of the correction
w x w x
procedure. This approach is the same as used by Munneke 10 . See also Lee 4 .
TABLE 6
Estimation Results of the Structural Probit
Value
differential
.
Constant VALDIFF LLF LRI
Model 1 0.000698 y127.748 0.054
.
Excluding demolition variables 3.342
VALDIFFsVR1 yVC1 y0.119778 0.000809 y127.039 0.060
. .
1.185 3.398
Model 2 0.001470 y118.076 0.126
.
Including demolition variables 4.426
VALDIFFsVR2 yVC2 y0.027682 0.001469 y118.034 0.126
. .
0.290 4.437
Note: The absolute values of the t statistics are presented in parentheses. The dependent
variable in the structural probit is REDEV which takes on the value of 1 if the parcel was
redeveloped and 0 otherwise. The LRI is analogous to the R
2
statistic in an OLS model. See
w x
Greene 2, p. 651 .
DANIEL T. MCGRATH 438
.
the demolition variables , the impact of VALDIFF on the probability of
redevelopment is positive and significant. The significance of VALDIFF in
Probit Model 1 is maintained for probit model specifications that both
include and exclude a constant term, although the constant is statistically
insignificant. The insignificance of the constant term is an interesting
result, given the fact that the demolition variables have been excluded
R
w x
from the redeveloped value function V . Munneke 10 argues that if the
value differential is the only determinant in the criterion function, then a
significant constant term would represent a fixed demolition cost for
industrial property. However, Munneke finds in his probit analysis of
industrial properties that the inclusion of a constant term, while statisti-
cally significant, eliminates the significance of the value differential vari-
able for industrial properties.
8
In this analysis, the insignificance of the
constant term in the Probit Model 1 supports the view that demolition
costs are likely to be systematically related to some characteristic of the
parcel that would certainly be capitalized into land value and thus become
part of the parcels value differential.
Model 2 uses the value equations specified with the demolition variables
VR2 and VC2 to calculate VALDIFF. The mean value for VALDIFF is,
as expected, positive for the redeveloped subset of parcels, and, on
of the other systematic determinants of land value and thus value differ-
.
ential for the parcel in question. Intervention would be optimal on
marginal properties that have the locational and other site characteristics
which, but for the existence of contamination risk, would clearly be
desirable to an industrial user. For such properties, the gain in land value
and the subsequent increase in value differential resulting from a site
remediation might indeed bring the parcel into a more competitive posi-
tion in the private industrial real estate market and could be viewed as a
viable strategy to attract jobs to a central-city location. Therefore, under-
standing the spatial patterns of value differential for industrial parcels
within the urban core could be a very important strategy for local govern-
ments to optimize their efforts, specifically publicly funded site cleanups,
to attract new industrial users for brownfield properties within the urban
core.
REFERENCES
1. J. K. Brueckner, A vintage model of urban growth, Journal of Urban Economics, 8,
.
389402 1980 .
2. D. Coursey et al., Environmental Racism in the City of Chicago: The History of EPA
Hazardous Waste Sites in African-American Neighborhoods, unpublished paper,
.
University of Chicago 1994 .
.
3. W. H. Greene, Econometric Analysis, MacMillan, New York 1993 .
4. L. F. Lee, Unionism and wage rates: A simultaneous equations model with qualitative
.
and limited dependent variables, International Economic Reiew, 19, 415433 1978 .
.
5. L. F. Lee, Identification and estimation in binary choice models with limited censored
.
dependent variables, Econometrica, 47, 977995 1979 .
6. L. F. Lee, G. S. Maddala, and R. P. Trost, Asymptotic covariance matrices of two stage
probit and two state tobit methods for simultaneous models with selectivity, Econo-
.
metrica, 48, 491503 1980 .
7. G. S. Maddala, Limited Dependent and Qualitative Variables in Econometrics, Cam-
.
bridge Univ. Press, Cambridge, UK 1983 .
8. J. F. McDonald, The substitution of land for other inputs in urban areas, Papers of the
.
Regional Science Association, 48, 3952 1981 .
9. E. S. Mills, The value of urban land, in The Quality of the Urban Environment H. S.
. .
Perloff, Ed. , Johns Hopkins, Baltimore, MD 1971 .
DANIEL T. MCGRATH 442
10. H. J. Munneke, Redevelopment decisions for commercial and industrial properties,
.
Journal of Urban Economics, 39, 229253 1996 .
11. F. Noonan and C. A. Vidich, Decision analysis for utilizing hazardous waste site
.
assessments in real estate acquisition, Risk Analysis, 12, No. 2, 245251 1992 .
12. S. S. Rosenthal and R. W. Helsley, Redevelopment and the urban land price gradient,
.
Journal of Urban Economics, 35, 182200 1994 .
13. W. C. Wheaton, Urban spatial development with durable but replaceable capital, Journal
.
of Urban Economics, 12, 5367 1982 .