Intra-Regional Amenities, Wages, and Home Prices

17 Pages·2008·7.09 MB·English

, ^7
Intra-Regional Amenities, Wages, and Home Prices:
The Role of Forests in the Southwest
Michael S. Hand, Jennifer A. Thacher, Daniel W. McCollum, and
Robert P. Berrens
A I3STRACT. loresis /110 tide n n-oiatkei goods and economic decisions over, such as abundant
vertices that people are iotplkiilt tilling to pat for recreation opportunities, open space, and
i/trough hedonic /toiooig and labor ntarlu'ts. But it is the availability of wilderness experiences.
tote/ear if compensating di//ereniial.v arise in there Further, forests provide several non-market
nearkets at the regional level. This empirical t/ueriioit
ecosystem services, such as watershed and
is add essed in a SItU/I' of Arizona and New Mexico.
airshed nmnagement, soil quality protec-
Hedonic regressiowv of ltottsing pines (1)1(1 lragev using
census and geographic data shall that totest area tion, and habitat protection. Individuals
carries (IIt implicit price of heiti'eeit $27 cettcI 536 per who value these characteristics may be
sc/mite mile annual/v. Compensating d,//ereniiair al willing to pay. through compensating dif-
the regional level suggest i/tal cute titust he taken ferentials in labor and housing markets, for
when app/vutg the travel cost titeiltod to value access to these non-market goods and
regiottuellm' delineated characteristics. (J EL Q23. R 14) services. A portion of' the first paycheck
(i.e., money income) may be forfeited in
I. INTRODUCTION
favor of a larger second paycheck.
Compensating differentials have been
As traditional resource extraction has
shown to exist in labor and housing markets
become relatively less important in regional
for large-scale amenities (e.g., climate) at
economies of the American West. the
the inter-regional level (Roback 1982;
amenity values of forests and other natural
Hoehn, Berger. and Blomquist 1987; Blom-
features have received more attention. In-
quist. Berger, and Hoehn 1988; Costa and
stead of seeking out areas that, for example,
Kahn 2003), and in housing markets for
hold the promise of a large paycheck from
local or neighborhood amenities (Ridker
togging or mining jobs, migrants may seek
and Henning 1967; Cheshire and Sheppard
out areas that offer a large so-called "second
1995; Acharya and Bennett 2001). Studies
paycheck" derived from the value of the
that include forest resources have also
natural landscape. This economic behavior
shown that proximity to forests is capital-
can have observable effects in the markets
ized into local housing prices (Shultz and
for housing and labor.
King 2001; Kim and Johnson 2002; Kim
As argued by some authors (e.g., Niemi.
Whitclaw, and Johnston 1999). a second
paycheck may be derived from forests
The authors are, respectively, research economist.
because these areas contain many of the
USDA Economic Research Service; assistant professor of
characteristics that people value and make economics. Department of Economics, University of New
Mexico; research economist. Rocky Mountain Research
See Nierni, Whitelaw, and Johnston (1999) for a Station, USDA Forest Service; and, professor of eco-
discussion of the second paycheck. Also. Garber-Yonts nomics. Department of Economics. University of New
(2004) provides a thorough review of literature related to Mexico. Funding support for this research was partially
the impact of natural characteristics on migration. provided by joint venture agreements (06-JV-199 and 05-
hedonic prices, and local economic conditions. JV-257) between the Rocky Mountain Research Station
(U.S. Department of Agriculture. Forest Service) and the
Department of Economics. University of New Mexico.
Land Economics. November 2008 • 84 (4): 635-651 The authors wish to thank Carl Edminster and Carolyn
ISSN 0023-7639; E-ISSN 1543-8325 Sieg for their support. The views expressed are those of
© 2008 by the Board of Regents of the University the authors, and do not necessarily reflect those of the
of Wisconsin System Department of Agriculture.
636 Lam! Economics Aorc,iiber 2008
and Wells 2005) and wage differences Southwestern region, this planning process
between cities (Schmidt and Courant involves regional teams formed to ensure
2006). Lacking in the literature, however, coordinated revisions of all individual
S.4
is an understanding of how forests (and forest plan Knowledge of the importance
other natural characteristics) impact eco- of forest resources at the regional level
nomic behavior within a particular region. would aid in determining the economic
Whether or not forest amenities are impor- impact of regional forest policy decisions
tant in the housing or labor markets (or and how management decisions conform to
both) within a region remains an empirical the public's preferences.
qnestion. Beyond the implications for forestry
The purpose of this paper is to examine policy, the answer to the empirical question
compensating differentials for forest ame- posed here may indicate a role for forest
nities in housing and labor markets in the amenities in rural development policy. For
Southwestern United States, composed of example, if forests and other natural
Arizona and New Mexico. This region characteristics generate wage differentials
provides an excellent opportunity to answer within the region (e.g.. between rural and
the research questions asked in this paper. urban areas), attributing observed wage
The region is characterized by widevaria- differences to other factors can lead to
tion in the size. type, and location of forest erroneous policies. It is also possible that
characteristics and their proximity to large jurisdictional boundaries prevent local
population centers. A unique geographic communities from internalizing the market
information system (GIS) data set from the effects of forest policies (Schmidt and
U.S. Forest Service Southwesterii region Courant 2006). Recognition of these roles
allows these characteristics to be precisely for forests, if they exist, suggests that an
measured. Hedonic regressions are estimat- important component of rural economic
ed for housing prices and wages using development policy may include preserva-
publicly available Census data, GIS calcu- tion of the natural features that people find
lations of landscape features, and data amenable.
enumerating outdoor recreation sites. An The results of this study may also be
empirical framework pioneered by Roback important for the application of non-
(1982) and developed by Hoehn. Berger, market valuation techniques within the
and Blomquist (1987) is used to determine region. Two issues arise if the characteristic
the existence of hedonic premiums for forest to he valued is generating compensating
amenities in housing and labor markets and differentials. First, an endogeneity problem
the implicit prices faced by Southwest arises if labor market earnings are used
region residents for these amenities. (e.g., as an independent variable) to esti-
Compensating differentials imply that mate a demand relationship for a charac-
the supply of forest amenities at the regional teristic that generates a wage differential
level is policy relevant. National forests in the preferences an analyst seeks to measure
the United States, under the management of with non-market valuation techniques may
the U.S. Department of Agriculture's For- determine an individual characteristic
est Service, undergo a periodic regional thought to be an exogenous demand shifter.
planning process.3 In the Forest Service's Second. the data-generating process be-
hind some valuation techniques may itself
be endogenous to the preferences of inter-
2 Hand (2006) investigated labor market compensat-
ing differentials for linguistic enclave amenities in the est. An example of this problem is outlined
Southwest, but did not incorporate regional variation in in Randall's (1994)"Difficulty with the
natural amenities.
The regional planning process was set forth in the
final rule for National Forest System Land Management See 'Southwestern Region Revision Plan Strategy"
Planning. Federal Register, v.70 n.3. January 5. 2005. April 21. 2006. http:l/www.fs.fcd.uslr3/plan-revisionl
pg. 1055. strategy.shtml. accessed 911912006.
84(4) Jlaiic/ ci iii.: Forest Amenities in the Southwest 637
Travel Cost Method." In the context of New Mexico, (population about 5,100) has
regional forests, compensating wage or access to a different set of forest, outdoor
housing-price differentials imply that peo- recreation, and other characteristics than
ple are choosing their residential locations someone living in downtown Phoenix,
based on forests. To the degree that Arizona. Further, the labor market oppor-
preferences for forests are related to recre- tunities are unique to each place. These
ation travel, observed travel cost (i.e., differences define the bundles that people
distance) would be endogenous. Further. make decisions over and may give rise to
the travel cost would not reflect the full compensating differentials.
price of recreation site access, including the The underlying model used for the
price paid through compensating differen- empirical analysis is most similar to that
tials to live closer to a particular site. in Roback (1982). The key theoretical
Schmidt and Courant (2006) suggest that results, with notation reflecting the appli-
failing to account for location choices may cation to regional forest characteristics, are
lead to underestimates of the value of presented -here to inform the empirical
recreation access. approacli.
These issues begin to describe a role for Individuals derive utility from goods
regionally delineated natural amenities that consumption (X, a numeraire good. sold
is more deeply ingrained in local and at a constant price), land used for housing
regional economies than the sum of their (L, sold at price p), and the amenity
market and non-market values. Answering characteristics (/1'forests, in this case)
associated with a location. An individual
the empirical question of whether forests
lives in one of the J locations in the region,
are important determinants in hedonic
housing and labor markets will indicate / = I..... J. A convenient way of thinking
about the consumer's problem is that they
the degree to which this role may impact
select Xand L. conditional onf1 and subject
regional and local forest policy decisions.
to the budget constraint, to maximize
Utility. This yields an indirect utility func-
II. HOUSING PRICES AND WAGES IN A
REGIONAL HEI)ONIC FRAMEWORK tion, V = J7(p, it, ./), where it is an
individual's wage rate (assuming that in-
come is wholly derived from wage earn-
Forest characteristics have been unam-
ings). Firms also operate in both the labor
biguously shown to affect labor and hous-
and land markets, producing X and selling
ing markets in a variety of empirical
in a perfectly competitive market at the
settings. Among these are studies of the
nunieraire price. Thus, the firm's unit cost
wage and housing-price effects of inter-
function depends on the price of land,
regional differences in forest access
wages, and forest characteristics, or
(Schmidt and Courant 2006) and local
forest access (Kim and Wells 2005). The Clfi, n,/) = 1.
In equilibrium, utility and costs must be
primary empirical question in this paper is
equalized across all J locations; a location
whether forest characteristics that vary
that offers real utility gains or cost savings
regionally are in fact amenities that signif-
compared to an individual's or firm's
icantly affect regional housing-price and
current location would induce migration.
wage differentials.
Since forest characteristics are associated
The basic idea behind compensating
with a particular location and assumed
differentials in housing and labor markets
fixed, land prices and wages must adjust in
is that individuals purchase a bundle of
characteristics tied to a heterogeneous good
in the marketplace (see Rosen 1974). In the Several theoretical advances have been made since
this model was developed (e.g.. Hoehn. Berger. and
context of regional forests, the bundle of
Blomquist 1987). but the Roback model is more
characteristics is tied to where one lives and
appropriate to this application because the study area
works. Someone living in Mora County. does not have endogenous geographic borders.
638 Land Economics November 2008
each location to equalize utility and costs study region. The second condition is
across the region. The model must satisfy. assumed as in Roback (1982). The final
condition is the most relevant for estimating
VO = 17(j), u,f) yj. [I] compensating differentials within a region.
Some areas within the study region consid-
ered in this paper will more plausibly meet
C(p1. "./) = 1 Vj. [2]
this condition than others. For example,
where V1 is a constant. and land prices and people may live and work in the greater
wages are now associated with the particu- Phoenix area but commute across geo-
lar locations to reflect the equilibrium graphic areas as defined by the available
adjustment mechanism. The first equation data: such behavior violates the underlying
defines utility equality of individuals across assumption. How this issue may affect the
all areas, while the second defines produc- empirical results is discussed as necessary
tion cost equality. Totally differentiating below.
equations [1] and [2], noting that 1V = dC The partial derivatives of [3] and [4] are
= 0 in equilibrium, and solving for the two what traditional hedonic studies seek to
unknowns Iti'Idf and d1,Idf yields partials estimate. Beginning with housing price,
that are functionally equivalent to equa- forest amenities are but one feature that
tions [4] in Roback (1982): impact the price of an individual dwelling
unit (i.e., not all dwellings in location / are
- V/cl, Vl,c/ the same price). The price of a house is a
[3]
(If - D function of its structural (square feet,
number of bedrooms), neighborhood
(crime rates, school quality), and forest
(1/) - V C, - J, C
' [4] (forest views, recreation opportunities)
D
characteristics. The hedonic price function
where D = - V1C'1, + ',,C. < 0. specifies the price of house i in location.! as
As noted in Roback (1982) and Hoehn,
Berger, and Blomquist (1987), the signs of = P1 (s. N,, F',), [5]
the partials in [3] and [4] depend on the sign
where S, N, and F represent vectors of
of C (V1 is assumed to be positive).
structural, neighborhood, and forest ame-
Characteristics that are amenable to indi-
nity characteristics, respectively. It is as-
viduals may or may not affect productivity
sumed that the neighborhood and forest
for firms. If '1 > 0 (the amenity is
characteristics are common to all housing
unproductive), then dw/df < 0 and dpldf
o or > 0. If Cl. < 0 (the amenity is units in a given geographic area ].6
productive), then dii,Idf 0 or ^! 0 and Converting equation [5] into an econo-
metric model yields
dp/df> 0. In the case that amenities do not
matter for firms, both partials are unam-
6]
biguously signed less than zero and greater
than zero, respectively.
where v, is the assumed zero-mean random
The partials of equations [3] and [4] are
error and cE, P, y, and ô are the parameter
not equal to zero only under certain
vectors to be estimated. The implicit mar-
conditions: movements across the land-
ginal housing price of a forest characteristic
scape result in changes in amenity levels,
f is PI?f= 5.
individuals cannot occupy the same space in
time, and it is costly for individuals to access
amenities in areas where they do not live.
The first condition is met by the fact that This assumption is not necessary to carrying out a
hedonic price estimation, but is driven by the geographic
the characteristics of interest, those repre-
specification of the available micro data, described
sented by forests, vary widely across the
below.
iI

84(4) Hand ci al.: Forest Amenities in the Southnest 639
Hedonic wage estimation is conceptually productive, neutral, or unproductive ame-
similar. The traditional framework specifies nity for firms), the observed sign of the
that workers have preferences over a vector amenity wage and housing-price effect can
of job characteristics (e.g., risk of injury or be positive or negative. Thus, any calcula-
working conditions). In the case of exam- tion of the implicit amenity price requires
ining hedonic wages for forest amenities, both housing and wage differentials.
however, the concern is not with the Equations [6] and [7] can be estimated
characteristics of each job, but the land- in linear, logged. log-linear, or some other
scape characteristics that are tied to the area form.' Wage equations, for example, are
where one works. The same bundle of forest generally estimated with a log-linear spec-
amenities that may influence hedonic hous- ification (which is derived from its underly-
ing prices ma compensate workers for ing human capital investment theory (Minc-
y
lower wages. er 1974)). In contrast, Hoehn. Berger. and
Estimation of the hedonic wage must Blomquist (1987) and Blomquist. Berger,
incorporate other factors that affect the and Hoehn (1988) use a Box-Cox search
wage, namely productivity factors. The over possible specifications to determine the
most common method is to extend the specification for both equations that best
Mincer (1974) wage equation to include the fits the data. This latter method is adopted
vector of job characteristics. In this case, a here in pursuit of flexibility and goodness of
vector of landscape characteristics is added
fit.
that includes forest amenities. The hedonic
wage equation is.
III. DATA
+
= m + OV }' 115'\' + 'l + p. 17 The housing-price and wage regressions
are estimated on a matched sample of wage-
where fVii is the natural log of individual i's earner housing units. That is, each obser-
hourly wage, Y is a vector of individual
vation has a reported wage, rent or house
human capital characteristics and N and F
value, individual characteristics, and hous-
are defined above. Neighborhood and
ing characteristics. Estimates of equa-
forest amenity characteristics should be
tions [6] and [7] are conducted with indi-
included to the extent that the geographic
vidual-level data from the 2000 Public Use
areas correspond to labor markets. Note
that in the econometric model i indexes Microdata Series (PUMS), 5'Y0 sample for
Arizona and New Mexico (Ruggles et al.
individual wages and characteristics, rather
2004). This data gives individual responses
than particular jobs.
to the long form of the 2000 decennial
The empirical approach adopted in this
census. Detailed data is available for each
paper is most similar to Hoehn, Berger. and
Blomquist (1987) and Blomquist, Berger. respondent on income and earnings, edu-
and Hoehn (1988), where hedonic regres- cation attainment, race and ethnicity, hous-
sions for the housing market (equation [6]) ing characteristics, labor market participa-
and wage earners (equation [7]) include a tion. and a variety of other variables.
common bundle of amenities. The focus of Individuals are identified to geographic
this paper is the importance of forest areas, called Public Use Microdata Areas
amenities, so the bundle includes measures (PUMAs), with a population of at least
of forest area, recreation sites, and other 100,000. PUMAs are used to calculate the
natural amenities that vary across the vector of site-specific characteristics that
region. vary across the landscape and to compile
Hoehn. Berger. and Blomquist (1987) other geographic-based data. The study
show that the hedonic results from only one
market can be misleading. Depending on
the specific assumptions made about the
See Freeman (2003, ch.l 1) for a discussion of
production effects of amenities (i.e., a different functional forms for housing models.
640 Land Economics ;Voi'enther 2008
region is composed of 51 PUMAs, 36 in The hourly wage is constructed from
Arizona and 15 in New Mexico. reported earnings and work behavior. That
is,
Dependent Variables and Population
= uav,'cinv1/ (u/nv x tjsnrk1 ). [8]
The two dependent variables are housing
price and wages. Housing price is defined as where lt,ageinc1 is reported yearly wage
the annual contract rental cost or imputed income, uhrs1 is usual hours worked per
annual rent. For those households in the week, and wksivi-ki is number of weeks
worked in 1999.
sample who rent their dwelling unit, the
reported contract monthly rent is multiplied The sample for the wage equation is
restricted to wage-earning men and women
by 12 to obtain annual rent. Households
between the ages of 18 and 64. Those who
with owner-occupants report the current
did not earn wage income, worked fewer
value of the dwelling. Annual rent is
than four weeks in 1999, and usually
imputed by multiplying this value by an
worked fewer than ten hours per week were
annual discount rate of 7.51/0; annual rent
eliminated from the sample. This sample
for owners is estimated as the income they
was created to capture only workers who
are earning on that property as an invest-
are closely tied to the wage-earning labor
ment, which is assumed to be approximate-
market. Like the housing sample, observa-
ly equal to what they could earn if it was
tions with imputed data were dropped.
being rented to tenants." Top-coded obser-
The final sample used to estimate [6] and
vations of rent and house value ($1,700 per
[7] is the intersection of the housing and
month and $999,998 respectively) are ex-
wage samples. That is. only those observa-
cluded from the sample.
tions with complete data under the criteria
The sample for the housing equation is all
in both samples are retained for estimation;
households from the 2000 PUMAs in
all observations are household heads under
Arizona and New Mexico who either own
the Census definition. This results in a
or rent their residence (excluding group
sample of 36.375 male and 15,285 female
quarters). Households with imputed obser-
household heads.
vations for ownership status (owner-occu-
pied versus rental), housing value, or rent
Indepen clan! Variables
are excluded.9 One departure from previous
studies is the inclusion of households with
The primary interest in the empirical
10 or more acres of property. Previous work
estimates is measures of natural character-
was generally concerned with urban and
istics, and specifically those measures that
suburban residents, so excluding these
relate to forest resources. A comprehensive
households ensured that the sample reflect-
vector of site-specific characteristics has
ed the population and behavior of interest.
been gathered, including different measures
Although a small portion of our sample
of forest area, recreation sites, water fea-
(about 1.2%). the population of interest in
tures, hazardous waste sites, recreation-
the intra-regional case includes those living
related businesses, and community charac-
in rural areas who may have dwellings on teristics. However, several of these measures
more than 10 acres.
may exhibit collinearity when included in
empirical estimates (see Section 4). Table 1
describes the geographically coded indepen-
The 7.5% figure was used in Costa and Kahn (2003).
Hoehn, Berger, and Blomquist (1987) and Blomquist. dent variables retained for estimation.
Berger, and Hoehn (1988) use a discount rate of 7.85% The independent variables that measure
estimated from a separate source. forest characteristics in a PUMA are the
The PUMAs data often uses reported characteristics
proportion of PUMA land in U.S. Forest
to impute other missing data. These are dropped since the
characteristics used to impute data are often the same Service (USFS) area (AREA_FS) and
ones used in the hedonic regressions. Congressionally designated wilderness area
84(4) Hand Cf al.: Forest Amenities in the Southwest 641
TABLE I
Dr.scRIIrrIoNs OF GEOGRAPHIC TNDEPENI)I1' I VARIABLES
Variable Description (Source) Mean Std. Dcv.
AREA_FS Percentage PUMA area in USFS National Forests and .098 .172
Grasslands )GIS. USFS)
AVQFS Mean percentage USFS area in contiguous PUMAS (UIS. .17 .082
USFS)
AREA-WILD Percentage PUMA area in Congressionally designated .022 .054
wilderness areas )GIS, USFS)
AVGWILD Mean percentage wilderness area in contiguous l'UMAs .024 .023
(015. USFS)
SURFACE Area of surface water bodies )excludes streams), square miles 16.02 46.72
National Atlas)
11 AZ_COUNT Number of EPA-designated CERCLA hazardous waste sites 6.45 7.74
(EPA)
[SR EC Number of USFS recreation sites )NORSIS) (7.0 9.12
IED_RE(' Number of non-USFS federal recreation sites (NORSIS) 20.17 36.47
FSFED_R EC Number of USFS and other federal rec. sites )NORSIS) 37.16 40.08
URBBAANN Urban residence: Maricopa. Arizona Pima, Arizona: .667 .471
ltcrnalillo. New Mexico counties (binary. 015)
DENSITY Persons per square mile in residents PUMA .658.0 1.969,1
DENS2 Square of density 6.625.957 1(000,000
,Suiiria's: GIS indicates calculation of polygon- or linc-l'onntii 015 tiles for each PUMA: USES: United States Forest Service
Southsscsterti Region, hit p:lI'a ss 55 .fs.fcd.us/r3/gis/dalascts.shtnd: National Atlas; hitp://nationalatlas.gos latlasl tp.htittl lts drogtn: EPA:
U.S. Environrnetiial Protection Agenc C'ERCI .A database. [iiiii://svwNk.el)a.gov/en%ii-o/htiiil/ccrclis/in(lex.lititil: NORSIS: National
Outdoor Recreation Suppt\ Inlorntaitoti Svsteni, USES Southern Research Station.
(AREA—WILD). USFS area is generally TY and DENS2), and urban status (UR-
expected to be an amenity, but its designa- BAN). The forest area measures and surface
tion for multiple uses (including recreation water area are calculated for each PUMA
and extractive uses) distinguishes USFS using GIS calculations. Hazardous waste site
area from wilderness area. Wilderness area, counts and recreation sites are available at
like USFS area, is expected to carry a the county level: PUMAs are mostly defined
positive implicit price reflective of recrea- as one of, more counties, so the county-level
tion, ecosystem services, and passive use data is aggregated across the appropriate
values (Phillips 2004). Although these counties. Recreation site data is gathered
measures are generally thought to be from the National Outdoor Recreation
amenable, wildfire risk may offset the Supply Information System (NORSIS).
positive implicit prices to some extent which lists the number of several different
(Kim and Wells 2005). types of recreation sites and opportunities
Recreation access is measured by the for every county in the United States.]()
number of recreation sites managed by the There are three counties that are some-
USFS (FS_REC) and those managed by what problematic for dealing with county-
other federal agencies (FEDREC). The level data: Maricopa County and Pima
sum of these two measures (FSFED_REC) County in Arizona, and Bernalillo County
is used to estimate shadow prices for in New Mexico. These three urban centers
recreation sites in general. Recreation sites are separated into several PUMAs (22 for
include camping, hiking, boat launches, and Maricopa, seven for Pima, and five for
picnic sites. Bernalillo county). Given data availability,
The other site-specific data used in esti-
mation include square miles of surface water
0 The NORSIS data is maintained by Carter Betz at
area (SURFACE). number of CERCLA
the USFS Southern Research Station. Contact informa-
hazardous waste sites (HAZ_COUNT), tion available at http://www.srs.fs.usda.gov/trends/betz.
population density and its square (DENSI- html,
642 Land Economics No rent/jet' 2OO5
TABLE 2
DrscRt1moNs oF HOt SING AND WAGE STRUCTURAL VARIABLES

Men Women
Variable Description Mean Std. Dev. Mean Std. Dcv.
Housing I aria ble.v

ANN RENT Annualized contract lent or dwelling value, S 9,544.4 7.33)).0 7.554.3 5.117,3
ACR E9 Property size. 1 9 acres .122 .328 .087 .282
ACRE 10 Property size, greater than 9 acres .013 .114 .01(1 .098
CONDO Cond am jot u m Status. binary .016 .126 .044 .205
BEDS Number of bedrooms 2.76 1.08 2.43 .991

AGES Structure age. 2-5 years .163 .369 .131 .338
AGEI I) Structure age. 6 10 years .111 .315 .098 .297
AGE2O Structure age. II -20 years .229 .420 .249 .433
AG[30 Structure age, 21 31) years .221 .415 22) .42))
AGE4O Structure age. 31 40 years .096 .295 .101 .301
AG [50 Structure age. 41-50 years .076 .264 .084 .278
AGE6O Structure age. 51-60 years .026 .160 .033 .179
AGE6I Structure age. 61 or more years .026 .158 .034 .182
iI'tge J'artal,le.t
LNWAGE Natural log of hourh wages 2.76 .685 2.5)) .646

EDLIC Years of education 3.9 2.82 14.] 2.40

EXP Potential years of experience. =age-EDUC-6 21.3 11.0 20.1 11.5

EXP2 Square of EXP 575.9 508.6 537.0 505.5
MARRIED Marital status, binary .729 .445 .218 .413

IM MIG Immigrant status, binary .143 .351 .092 .289
BLACK Race: Black, binary .020 .138 .030 .171
ASIAN Race: Asiatt/Paci lie islander. binary .01S .133 .014 .117
NATIVE Race: Native American/Alaska Native, .037 .189 .066 244
binary
RACE2 Race: Two or more races, binary .138 .345 .123 .329
NOENG Speaks English not well or not at all, binary .046 .210 .024 .152
TRANTIME Transit time to work. minutes 22.2 19.4 18.8 17.1
.'Vo's.- ANN RENT and I.NWAGF. are the dependent '.ariahles. ANN_RENT is the rtn.translorrned value: results are reported with
the appropriate Box-Cox lrans(onnatlon.
the approach adopted here is to assign the IV. EMPIRICAL RESULTS
county-level value to all of the PUMAs in
that county. Estimation issues
The independent variables for the hous-
ing-price equation include categorical var- Any hedonic estimation involves several
iables for property acreage (ACRE9, econometric issues that must be addressed
ACRE 10). condominium status (CONDO), in order to obtain efficient and reliable
number of bedrooms (BEDS), and struc- results. As mentioned above, a flexible Box-
ture age (AGE). The wage equation inde- Cox search is used to determine the most
pendent variables include education efficient functional form. Using maximum
(EDUC), potential experience and its likelihood estimation, the generic Box-Cox
square (EXP, EXP2), marital status (MAR- search estimates,
RIED), immigration status (IMMIG), race
indicators (BLACK, NATIVE. ASIAN, Y"1 =
RACE2). English language proficiency + I + r:. 191
(NOENG), and transit time to work
(TRANTIME). Table 2 describes these where V is a dependent variable, X is a
variables. vector of k independent variables, and .
84(4) IIa,uI et al.: Forest Amenities in the Southwest 643
a, and fi are parameters to be estimated. geographic independent variables exhibited
Estimates of ).i = 0 and 2 = 0 correspond these problems and correlation coefficients
with a log-log specification, while 2 = between the independent variables con-
and 22 = 1 is the linear model. Any firmed this suspicion. To reduce the conse-
functional form where the Box-Cox param- quences of multicollinearity. a subset of
eters are between - I and 1 is possible. geographic variables is sought that are
12
The transformation parameter for the jointly independent. The independent geo-
continuous independent variables was esti- graphic variables described in the previous
mated as 27 = 1. For the housing-price section, AREA ES or AREA _WILD, SUR-
equation, /'l was estimated as 0.25 for men FACE. HAZ COUNT, and FSFED_REC.
and 0.38 for women. The wage equation satisfy the condition of joint independence.
estimates of /'.1 ranged between 0.007 and Finally, any econometric exercise that
0.05, but the logged form of the wage deals with geographic data must consider
variable (i.e., 2 = 0) is used to avoid the possibility of spatially dependent rela-
additional computation costs and to allow tionships. Spatial econometric problems
for simple comparisons of the human arise when data is autocorrelated in space
capital model to other Mincer-type wage (Anselin 1988). As is the case with time-
equations RI the literature. series data that is autocorrelated, estimates
There also exists the possibility that that do not correct for autocorrelation will
wages affect housing prices and housing be inefficient.
prices affect wages. As in Bloniquist, The spatial nature of the amenity data
Berger, and Hoehn (1988). a fully simulta- collected by PUMA or county suggests that
neous model is not estimated because spatial relationships could be present in the
calculations of implicit prices using estima- PUMAs data. Several methods are available
tions from both markets require reduced to correct for this potential problem. One
form estimates, where the effect of wages on option is to specify a spatially autocorrelated
housing prices (and vice versa) is implicit. model, where error correlation between
However, the possibility remains that there observations is not zero (i.e.. the off-diagonals
are unobserved factors that affect the error of the error covariance matrix are not zero).
terms in both equations. Estimates using a In this case, where the data set yields tens of
seemingly unrelated regression (SUR) re- thousands of observations, sitch an approach
vealed significant, if small, correlation is not feasible because of the size of the spa-
between the error variances of the wage tial weights matrix required for estimat ion. t3
and housing-price equations (see Griffiths,
Hall, and Judge 1993. 550); for this reason,
12 The method for selecting variables follows the
SUR estimates are used throughout.'' suggestion of Gujarati (1978. 184), where ordinary least
Given the large vector of site-specific squares regressions of one independent variable on the
measures collected as independent variables. other variables are used to identify significant linear
relationships between the independent variables. A set of
another concern is that the data will exhibit
variables is acceptable if the regression of each variable on
multicollinearity. In the case of imperfect the other variables all do not reject joint independence.
multicollinearity, estimates of coefficient All F-tests for the combinations in the results did not
standard errors will be large and coefficients reject independence at the 957; confidence level.
Kim, Phipps. and Anselin (2003) estimated spatial
sensitive to empirical specification changes
hedonic models with spatial data organized similarly to
(Gujarati 1978, 178). Preliminary regressions
the data used here. The authors specified a contiguity-
using the full vector and sub-vectors of the based spatial weights matrix, where all observations
within a sub-district (like a PUMA in our data) were
considered neighbors. This same approach is conceptu-
The results reported in Blomquist. Berger. and ally possible with the data in this paper, but the 609 total
Hoehn (1988) do not use a SUR model, although the observations in Kim. Phipps. and Anselin (2003) are
authors note that the cross-equation error correlations significantly more manageable than the 50.000-plus
are significant. The coefficient estimates reported here observations used here. A promising related method that
were more sensitive to the restriction of error indepen- deserves attention in this Context has been developed by
dence than in Blomquist, Berger. and Hoehn (1988). Conley (1999).
644 Land Economies November 2008
The second method is to control explic- For both men and women, the propor-
itly for those spatial characteristics that tion of the PUMA in forest area is positive
generate autocorrelation. This approach is and significantly associated with housing
more feasible with the current dataset values, and a larger proportion of forest
because it only requires additional vectors area in contiguous PUMAs is associated
of independent variables. For example, it with higher housing prices. Further, it
may be that the characteristics of nearby appears that forest area in contiguous
places affect housing prices and wages in a PUMAs is more important than forests in
particular area: Albuquerque, New Mexico, the household's own PUMA: the housing
has very few forest amenities within the city equation coefficients for AVG_FS and
limits, but large tracts of wilderness and AVGWILD are larger than AREAFS
National Forest areas just outside the city. and AREA—WILD for men, and AVG
As a First attempt to control for amenities in WILD is larger than AREA—WILD for
close places, the average measure of forest women.
area (AVGFS or AVG_WILD) in contig- Own-PUMA forest area does not appear
uous PUMAs is added as an independent to directly affect wages, but nearby forest
variable for observations in a given PUMA. area is associated with lower wages (a
negative coefficient indicates that larger
SUR Results values of that characteristic compensate
for lower wages). For both men and
The seemingly unrelated regression women, the coefficients for AVG_FS and
(SUR) in this context works well in AVG_WILD are negative and significant
explaining housing prices and wages in the while AREA_FS and AREA WILD are
region. Equations [6] and [7] are estimated positive and not significant. In other words,
and reported separately for men and people do not trade wages for forest area
women and reported in Tables 3 and 4, where they work, but they will accept lower
respectively.14 Two primary specifications wages to live and work in an area where
that are representative of the best fit and forests are closely accessible.
have been purged of multicollinearity are The coefficients for the forest area
reported: one using USFS area measures measures unambiguously indicate that for-
(AREA_FS and AVG_FS) and the other est area is amenable. But the estimates for
using wilderness area measures (AREA the other geographic features are more
WILD and AVG_WILD). difficult to interpret. Hazardous waste sites
The structural variables in the housing and (HAZ_COUNT) are negatively associated
wage equations are all of the expected sign with housing prices, which is the expected
and highly significant. For example, property sign, but the sign is also negative in the wage
acreage and number of bedrooms increases equation. The coefficients for surface water
housing value, while years of education and area (SURFACE) and recreation sites
experience increase wages. A test for error (FSFED_REC) are also the same sign in
correlation between the equations showed both equations. Proper interpretation and
that the SUR estimation is indeed justified a
any meaningful perspective of the magni-
correlation of about 0.20 for men and 0.18 tude of the results in a SUR framework
for women was present for all specifications require calculations of the total implicit
and was always significant. 15 prices.
'
Interacting a categorical variable for gender with all Total Implicit Prices
of the other independent variables confirmed that both
the wage and housing-price equation should be estimated The above discussion highlights the
separately for men and women (i.e.. all of the gender
necessity of considering both housing and
interactions were jointly significant).
15 See Griffiths. Hall. and Judge (1993) for the labor markets in tandem. For example, how
relevant Breusch-Pagan test statistic. do we interpret coefficient estimates that are