Smith1983 Desvousges

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 21

The Board of Regents of the University of Wisconsin System

The Opportunity Cost of Travel Time in Recreation Demand Models


Author(s): V. Kerry Smith, William H. Desvousges and Matthew P. McGivney
Source: Land Economics, Vol. 59, No. 3 (Aug., 1983), pp. 259-278
Published by: University of Wisconsin Press
Stable URL: http://www.jstor.org/stable/3145728 .
Accessed: 10/07/2014 08:39

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

The Board of Regents of the University of Wisconsin System and University of Wisconsin Press are
collaborating with JSTOR to digitize, preserve and extend access to Land Economics.

http://www.jstor.org

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
The Opportunity Cost of Travel Time
in Recreation Demand Models

V. Kerry Smith, William H. Desvousges, and Matthew P. McGivney

I. INTRODUCTION wage rate,are not likely to resolve the prob-


lems associated with defining a measure of
The purpose of this paper is to evaluate these costs that is both theoretically con-
current proposals for valuing travel time in sistent and practically feasible with exist-
the specification of recreation demand ing datasources.
models. Based on a household production There are two reasons for evaluating the
framework, our analysis examines the im- appropriate treatment of the opportunity
plications of the treatment of travel and on- cost of travel time with informationon the
site times in recreation models. The empiri- usage of a large number of recreationsites.
cal results are based on demand functions First, our theoretical analysis suggests that
estimated for 43 water-based recreation the appropriatevaluation of time devoted
sites with the Federal Estate component of to recreationtrips depends on the natureof
the 1977 National Recreation Survey. the time constraintsfacing each individual.
Overall, our findings suggest that the It implies that the measure of the opportu-
treatment of the cost of on-site time has im- nity cost of time will depend on the types of
portant implications for any evaluation of trips undertaken-e.g., long vacations ver-
the opportunity cost of travel time in recre- sus weekend excursions-and on the indi-
ation models. Approximately half the esti- vidual's available leisure time. Second,
mated site demand models were influenced several recent studies have identified the
by the treatment of the costs of on-site time key role played by the opportunity cost of
in the specification of the site demand func- travel time in recreation demand analysis.
tions. For sites that were not affected by the Forexample, in their application of the tra-
cost of on-site time, our evaluation indi- vel cost methodology, Bishop and Heber-
cated that we could not discriminate be- lein (1979) found that the estimated con-
tween the two possible measures of the op- sumer surpluses were very sensitive to the
portunity cost of travel time. Both the wage treatmentof the opportunity cost of travel
rate-in 15 of 22 sites-and the Cesario time. Nichols, Bowes, and Dwyer (1978)
(1976) proposal-in 14 of 22 sites-could concluded their review of the availableevi-
not be rejected as appropriate measures of dence on the valuation of travel time, not-
this opportunity cost. For those sites where ing the important role of opportunity cost
the costs of on-site time were important de-
terminants of site demand, the majority re-
Centennial professor of economics, Vanderbilt
jected both hypotheses. Thus, these find- University, senior economist and economist, Re-
ings reconfirm the importance of the search TriangleInstitute, respectively. This research
selection of measures for the opportunity was supported under a U.S. EnvironmentalProtec-
cost of time involved in outdoor recreation. tion Agency ContractNo. 68-01-5835. Thanks are
due to Ann Fisher for comments throughout the re-
Moreover, they suggest that simple approx- search, to the participants in Resources for the Fu-
imations, such as treating the opportunity ture's QE Seminar Program for most constructive
cost as a fixed multiple of an individual's comments on several aspects of this research,and to
Suzanne Holt and an anonymous refereefor detailed
LandEconomics,Vol. 59, No. 3, August 1983 and constructive comments on an earlier draftof the
0023-7639/83/003-0259$1.50/0 paper.The views expressed in the paper are those of
? 1983 by the Boardof Regents the authorsand not of either their institutions or the
of the Universityof Wisconsin System funding institution.

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
260 Land Economics

measures. They observed that there was Section IIreviews the past theoreticalar-
.... a severe lack of empirical information guments that have attempted to define the
aboutthe value of travel time to recreation opportunity cost of travel time. Section III
sites." Wilman (1980) echoed these conclu- develops a general formulation of the de-
sions in a recent theoretical analysis of the mand for a recreationsite's services using a
alternative conceptual frameworks for household production framework de-
treatingtravel and on-site time. Moreover, signed to consider these opportunity costs.
she noted that empirical work was needed Section IV describes the data used in our
to clarify the appropriateassumptions and analysis and the basic structureof the esti-
values required to specify a role for time mating model. Section V reports our esti-
costs in travel cost demand models. mates of the demand functions for each site
On the whole there have been some ef- and results from hypothesis tests. The final
forts to respond to these suggestions, but section summarizesthe implications of the
the information available on this problem analysis.
remains quite limited. For example, Mc-
Connell and Strand (1981) argued that the II. MODELINGTHEROLEOF TRAVEL
TIME:A REVIEWOF THEISSUES
opportunity cost of travel time should be
treated as a fraction of the wage rate that All of the economic models that describe
may vary across recreation sites. Conse- the recreation decisions of a household as-
quently, they suggested estimating this sume an individual seeks to maximize util-
fractionby entering out-of-pocketcosts per ity subjectto time and budget constraints.3
trip and the time costs valued at averagein-
come per hour separately in each site- 'Their specific application required a number of
specific demand model. While McConnell restrictive assumptions to treat average family in-
and Strand acknowledged a number of come as a measureof the opportunitycost of time, in-
cluding: zero tax rate, constant marginal earnings,
problems with their approach (which in- and zero nonwork income. Moreover,family income
volves estimating this fraction as the ratio must be synonymous with the individual's income,
of randomvariables),they were optimistic and the individual cannot acquire utility from work.
about its potential in recreation demand Theirspecificationsomitted income as a determinant
analysis.' of site demand. While the use of average income to
measure the opportunity cost of time does not pre-
Our research extends past analyses of clude inclusion of the total income measurein the de-
the opportunity cost of travel time in three mand function, it can induce high collinearity and
ways: thereby impede precise estimates of both price and
(a) We propose an alternative method income parameters.
for estimating the wage rates used in ap- The specific nature of the effect will also depend
on how the traveltime is derived. In many cases, it is
praisingan individual's opportunitycost of estimatedfromround trip distance (d) as d/s, where s
time. This approach uses a hedonic wage = velocity. In these cases, a demandmodel includes,
model to predict, for each individual, a expressedin termsof travelcost, the opportunitycost
of travel time, and income would include variables
wage rate based on the individual's per- with similar arguments, (i.e., cd (I/H), (d/s), and I,
sonal, job, and residential site characteris- where c = travelcost per mile, I = family income, H
tics.2 = numberof hours of work per year).Since these var-
(b) We specify the travel cost demand iables include several constants (i.e., c, H, and s), the
models so thatthe treatmentof the opportu- degree of collinearity between them would depend
on the extent of independent variation in the other
nity cost of traveltime is a testablehypothe- variablesin these expressions.
sis. This allows the theoretical arguments 2Theuse of a projectedwage ratefor each individ-
for valuing time at "scarcity"or "commod- ual that attempts to incorporatethe attributesof the
ity" values (see Cesario 1976 and Wilman job (as leading to compensating differentials)seems
1980) to be evaluated statistically. to be consistent with McConnelland Strand's(1981)
suggestionsfor furtherresearch.
(c) We explicitly examine the potential 3Inwhat follows we use the terms "household"
role for the costs of on-site time in recrea- and "individual"as synonyms. See Becker(1974) for
tion site demand functions. theoreticaljustification.

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
Smith, Desvousges, and McGivney:Travel Time 261

These models are distinguished by the links the choice variables to the house-
level of detail used to describethe interrela- hold's time and income constraints.
tionships between the numberof recreation The basis for her derivationof a different
trips, the time spent at recreationsites, the implicit valuation of travel and on-site
activities undertakenon site, and the other times is an assumption that the number of
goods and services consumed by the indi- trips and visits to a site are equal. The re-
vidual. For the most part, earlier studies sulting firstorderconditions requireequal-
have adopted one of two simple formula- ity between the sum of the marginal utili-
tions of the consumer choice process. In the ties of trips and visits, and the
first, a composite market good and recrea- correspondinggoods and time costs of each
tion trips (McConnell1975;McConnelland (weighted by the appropriate lagrangian
Strand1981) provide utility. In the second, multipliers). Equation [1] below repro-
income and the amounts of work and lei- duces a simple version of the relevant first-
sure time (Cesario 1976) enter the utility ordercondition.
function directly. With the first model, all
recreationtrips are homogeneous in length MUV+ MUT= h(Pvbv+ PTbr)+ k(aT + av)
with the time requirements fixed per trip. [1]
Time allocated to recreation must be di-
verted from other uses if the recreationtrip where:
is to be taken. While time has occasionally MUv = marginal utility of a visit (v);
been separated into components-travel MUT = marginalutility of a trip (T);
time and on-site time-the distinction is bv,bT= goods requirements for visits and
unimportant,because neither one is treated trips, respectively;
as a choice variable. PV,PT = prices of goods required for visits
The second approach for modeling the and trips;
role of time in recreation decisions con- aV,aT= time requirements for visits and
siders the general problem of allocating trips, respectively;
time between work and leisure activities. It h, k = lagrangianmultipliers.
assumes that the decision to visit a recrea- The lagrangianmultiplier, k, can be inter-
tion site is equivalent to any other leisure pretedas the scarcityvalue of time. Wilman
time decision. Unfortunately, this frame- rearrangesequation [1] as equation [2] in an
work does not provide direct guidance as to attempt to distinguish the values used for
how observable market prices-whether on-site and travel times.
explicit or implicit-influence these
choices. MUv = h(Pvbv + PTbT) + kav
Wilman's (1980) conceptual analysis
can be interpreted as an attempt to bridge + aT(k - MUT/aT) [2]
the gap between these two types of models
and provide a rationale for defining im- In this formit is arguedthat there is a dis-
plicit prices for the differentuses of leisure tinction between the appropriate implicit
time. The Wilman model specifies utility to valuation of on-site time (ar) and travel
be a function of goods and services requir- time (aT).That is, the on-site time seems to
ing time, those not requiringtime, and two be valued at the scarcityvalue (k)while tra-
measures of use of a recreation site-the vel time is valued by the scarcity value less
number of visits of a given length to a site the "commodity value of time" (MUT/aT)
and the number of round trips to that site. (see De Serpa 1971 and Cesario 1976). How-
The latteris intended to reflectany satisfac- ever, this distinction is an artificial one,
tion derived from traveling to the recrea- since it is only one of an infinite number of
tion site. By assuming that the time and cost allocations. To establish this conclu-
budget requirementsare fixed multiples of sion let c1 be a positive constant between
each of these decision variables, Wilman zero and one. A completely equiva-

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
262 Land Economics

lent description of the necessary condition constrained objective function with


given in [1] using c1 is then: U(Zr,Zo) the utility function and f4(.) and
fo(.)the household productionfunctions for
MU, = h(P.b~ +PTbT) + a,(k - clMUT/av) recreationand nonrecreationservice flows,
respectively.
+ aT(k - (1 - cl)MUT/aT) [3]
G U(Zr,Zo)+ =
In this case neither type of time is valued at
the scarcity value of time (k). Rather,each \[wtw + R- PrXr - PoXo-c(di VI + d2V2)]
sharesin the loss of the commodity value of
time depending upon the value of c1. + 0I[Z,- fr(Xr, VI, V2,tvi, tv2)]
Wilman's formulation is simply a special
case of [3] with c1 = 0. + 02[Zo - fo(Xo, ta)] + (pi(1i - tw- t,)
While Wilman proposed this rearrange-
ment in the allocation condition for visits +P2[ t2
-
VI(tv + t) - V2(tv2 +
t2)] [4]
and trips in an attempt to take account of
the potential utility derived from travel where:
time, it is importantto note that trips and tw= worktime;
visits are delivered jointly on a one-to-one w = wagerate;
basis in this version of her model. They R = nonearnedincome;
must be treatedas a single commodity and Xi = marketgood(i = r,o);
any cost allocation between them is arbi- Pi = priceof Xi(i = r,o);
trary.4Indeed, once the equality assump- c = vehicle-related travel cost per
tion between trips and visits is dropped, mile;
Wilman'smodel implies that both types of V1= numberof trips to ith recreation
time should be valued at their scarcity site (i = 1,2)6;
value (k) (see Wilman 1980, equation [24]). di = round trip distance to ith recrea-
Thus,the existing recreationliteraturedoes tion site (i = 1,2);
not provide an unambiguous theoretical tj,tvi = travel time and time spent on site
justification for distinguishing the valua- per trip to ith recreation site (i =
tion assigned to the travel and on-site time
components of a recreationalexperience. 4Burt and Brewer (1971) anticipated this conclu-
sion in a discussion of the appropriate unit of mea-
PRODUCTION
III. A HOUSEHOLD sure for outdoor recreation consumption in an appen-
MODEL dix to their paper. They observed that:
FORRECREATION DECISIONS The literature on outdoor recreation economics
abounds with discussions of the cost in time and in-
To recognize Cesario'sconcerns over the convenience of traveling to the source of recreation
services. Since these costs are extremely nebulous to
types of time constraints,we propose an ex- quantify, there is considerable advantage in defining
tension to the household productionframe- the recreation unit in a way that avoids this indirect
workfor modeling outdoorrecreationdeci- cost.... Travel time can be a pleasure or a cost, de-
sions.5 Assume for simplicity that the pending on the scenery, road, traffic congestion, etc.;
but it is an inseparable part of the recreation trip. In
household produces and consumes two other words, travel time and the recreation experi-
final service flows-recreational activities, ence itself are a package of commodities in the usual
economic sense, and the consumer has no alternative
Zr,and a nonrecreation composite service to the particular package presented to him by his spa-
flow, Zo. These activities are selected to tial location (Burt and Brewer 1971, p. 826).
maximize utility subject to the constraints 5This framework was originally proposed by
of: (a) the household production technolo- Becker (1965) and has been developed for recreation
gies for Zrand Zo;(b) the budget constraint modeling by Deyak and Smith (1978) and Bockstael
and McConnell (1981).
for purchasedinputs; and (c) two time con- 8Separate terms for the visits to each of two sites
straints. are included in this specification of the production
Equation [4] presents the household's function to reflect the prospects for site substitution.

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
Smith, Desvousges, and McGivney:Travel Time 263

1,2); assumed constant for all available,but when it is free may not coin-
trips; cide with the requirements of the desired
ta = time allocated to the production activities.
of nonrecreational service flows; The use of multiple time constraints is
t, = total time available for work and an attemptto reflectall of the dimensions of
nonrecreational activities; time availability within a comparative
t< = total time available for recrea- static framework.The two constraints can
tional activities. be considered to reflect time at different
X,O9,02,0p1,4p2,are lagrangian multipliers. periods during the time horizon of the
model. For example, t2 might be inter-
The specification of two time constraints
preted as the total time available between
requires additional clarification. In de- twelve noon and five p.m.
scribing the rationale for his analysis of the To consider the implications of this for-
opportunity cost of travel time to recreation mulation of the consumer choice problem,
sites, Cesario (1976) observed that: we have solved the first-orderconditions
... it seems farfetched to assume that the for a maximum of equation [4]. This yields
recreation tripmaker is trading off time for a relationship that is comparable to the
travel with time for work. It seems much marginal conditions in a conventional
more likely that the tradeoff is between model of consumer choice, as given in
time for travel and time for leisure activi- equation [5].
ties, which we loosely define to be activi-
ties conducted during nonwork hours, MUzOMP a MUZoMPx?o MUzrMPr
whether they be in the form of rest, sleep, [51
w PO PI
gardening, outdoor sport, etc. (Cesario
1976, p. 34).
MUzr MPv
Cesario has expressed one of the reasons 2
for dissatisfaction with conventional treat- cdl + w(tvl + tl)
q9,
ments of the consumer's decisions on allo-
cating his (or her) time. Comparative static MUzr. MPr
models do not reflect the importance of 2
"the timing of the time" in allocation deci- d2 + w(tv2 + t2)
sions related to the use of an individual's
time. For example, it may be difficult for MUzr 'MPt, MUzrMPtr
some individuals to have discretion over
P2 2
the use of their time at all times of day (or -wVWV V2
week or month). These constraints may cpl
arise from occupations that limit flexibility, where
from the availability of daylight hours
which may be required for particular recre-
MUzi = marginalutility of ith final service
flow (i = r, 0);
ational activities, or from a number of other MPf = marginal product of jth input in
constraints influencing households' behav-
production of Zr (j = Xr, Vi, V2, tvl,
ioral choices. Unlike the constraints im-
42);
posed by the total amount of time available MPO= marginal product of sth input in
for leisure activities, these constraints
reflect the limits of the timing of free time. production of Zo (s = Xo,ta).
An individual working the night shift has The important elements for consider-
discretion to use his (or her) free time for an ation of the role of differenttypes of time in
afternoon baseball game, but cannot allo- equation [5] are the expressions involving
cate time for "prime time" television pro- the "marginal utility products" of visits
grams or for an evening performance of a and time on site for each of the two types of
play. The total amount of time may well be sites (i.e., the last four ratios in the equality

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
264 Land Economics

given in [5]).The denominatorof each ratio the perceived importanceof on-site time in
can be interpretedas an "implicit" price of recreation demand models. For example,
each choice variable.Thus, this framework most recreationists' trips to areas that at-
clearly illustrates that, assuming no entry tract national attention involve different
fees, there are are three components of the types of time constraintsthan trips to local
cost of a visit to a site: vehicle-relatedtravel areas.Generally,the time availablefor trips
costs (cdi),the cost of travel time, to national sites is constrained by the indi-
vidual's annual vacation. In contrast, the
( -2W ti) demands for recreational areas attracting
users within its immediate vicinity are
and, because the model restrictsall trips to more likely to reflect weekend or short-trip
the same amount of time on site, the on-site visits. Our frameworkimplies that no one
time cost of one more trip approachfor estimating the costs of travel
time will suit all types of recreationequally
( 2wtvj)
(Pi
well. Thus, there is theoretical support for
assessing the performance of the current
In general,the relevant opportunity cost of proposals for measuring the opportunity
travel time will differ from the wage rate costs of time with informationon a number
(except where (92 = (pi). Since (p and 92 are of recreationsites.
lagrangianmultipliers and can be expected
to be functions of all the parametersto the IV. DATAAND MODELSPECIFICATION
consumer's optimization problem, includ-
ing the wage rate, the most appropriatein- The Federal Estate component of the
terpretationof [5] implies that the opportu- Heritage Conservation and Recreation
nity cost of time is best treated as a Service's 1977 National Outdoor Recrea-
nonlinear function of wages. Moreover,for tion Survey provided the visitor informa-
our particularformulation of the decision tion to estimate our travel cost demand
process, the opportunity cost is the same functions. Although 13,729 recreationists
function of the wage for both the time on were interviewed at 155 sites, our analysis
site and the travel time per trip. was confined to 43 U.S. Army Corps of En-
Based on this model, the Cesario (1976) gineers sites with consistent visitor infor-
and McConnell-Strand(1981) approaches mation. These sites were selected because
to valuing travel time can be interpretedas they provide a fairly comparable range of
practical attempts to approximate this water-basedrecreationalactivities. Table 1
more complex-and theoretically appro- identifies the sites we selected and pro-
priate-expression for the opportunity vides informationon some featuresof each
costs. Unfortunately, our model does not site and of the survey respondents.
suggest an empirically feasible approach The Federal Estate survey includes de-
for treatingthese time costs, because more tailed socioeconomic information on each
detailed information on the time con- respondent, the travel time incurred to
straints facing recreationists would be re- reach the site, a variety of variables indica-
quiredthan is available. ting the respondents' attitudes toward out-
Inaddition, our model implies that some door recreation, and the length of the trip
consideration should also be given to the during which they were surveyed. Round
role of on-site time. On-site time is both a trip distances were not included. However,
component of the cost of a trip and a choice it was possible to measure these distances
variable.In the absence of specific informa- because the zip codes for surveyed individ-
tion on these time constraints, we can ex- uals' residences were reported. The dis-
pect their effects will be reflectedin the per- tance measurements assumed that respon-
formanceof the methods for approximating dents used interstateand primaryroads to
opportunitycost. They may also influence reacheach site. The travel costs per mile (c)

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
Smith, Desvousges, and McGivney:Travel Time 265

were assumed to be 8?, based on the 1976 (ourCPSsample contained 9,077 for males
estimate of the averagecost of operatingan and 7,067 for females) and a small errorvar-
automobile (source: U.S. Statistical Ab- iance for the estimated model, the bias in
stract1978). our estimates will be small. A 10%discrep-
As in most recreation surveys, the ques- ancy would be a generous outer bound on
tionnairedid not ask the respondent's wage the magnitude of the percentagedifference
rate. Since this variable plays an essential between our direct predictions of these
role in any evaluation of the opportunity wage ratesand the estimates based on Gold-
costs of travel time, it is important to de- berger'smethod. Indeed, in most largesam-
scribe how these wage rates are estimated ple applications (see the examples in Gold-
in our empirical results. Past studies have berger 1968 and Giles 1982), the actual
either assumed a constant wage rate for all differences are under 5%. Thus, despite
individuals (see Cicchetti, Fisher, and this limitation, we feel these estimates pro-
Smith 1976) or used an averageincome rate vide a betterset of proxy measuresfor wage
per hour based on reported family income rates than the available alternatives in that
(see McConnell and Strand 1981). Both of they takeexplicit account of individual and
these practices have been acknowledged to job characteristics. To consider the pro-
provide very crude proxy estimates for spective differences,we have comparedav-
wage rates. Our alternative approach uses erage income rates (using family income)
the predictions from hedonic wage models by income class and by occupation with the
estimatedseparatelyfor males and females. average wage predictions for these same
These models were estimated with the categories.Table 2 indicates that while the
1978 CurrentPopulation Survey (the sur- two types of estimates generally move in
vey was actually undertakenin 1977), spec- the same direction over occupation classes,
ifying the nominal wage rate to be a semi- there are substantial differences for the
log function of a local cost of living index upper-income classes, where family in-
and variables measuring:(a) the character- come may reflect the effects of substantial
istics of each surveyed individual, includ- non-wage income and the impact of dual
ing age, education, job experience, sex, earnerhouseholds.9
race, occupation, and other indicators of Equation[6] specifies the generalformof
family status;7(b) the attributes of the re- the travel cost demand model used to eval-
spondent's job and industry; and (c) the uate the alternative methods for treating
characteristicsof the individual's residen- travel time. Based on its performancewith
tial location, such as climatic conditions, site demand models and on the growing
crime rates, air quality, etc.8 support for this form from earlier findings
The Federal Estate survey provided suf- (see Smith 1975; Ziemer et al. 1980,,,and
ficient information (i.e., on individual, job,
and residential site characteristics)to use
these wage models to predict wage ratesfor
7In practice, these data sets do not have a direct
each individual. When data for variablesin measure of experience and utilize a proxy-age mi-
these wage equations were not available, nus education and minus sex. Thus, both the proxy
the sample means from the 1978 CPS sur- for experience and age are not included as separate
vey were used. Of course, our estimates determinants of individual wage rates.
8Formore details on the model and the wage equa-
should be regardedas proxy measures for tions using the real wage as the dependent variable,
the actualwage rates.Since the wage model see Smith (1983). The estimated wage models with
is a semi-log, we can expect the predictions the nominal wage are available on request from the
to understatethe estimated conditional ex- first author.
9The CPS survey also limits the reported usual
pectation for a wage rate. While Gold-
weekly earnings to $999. Hence, there is some censor-
berger's(1968) proposed unbiased estima- ing in the wages reported for individuals likely to
tor for this conditional expectation would have these higher annual incomes. This may also par-
be superior, for large degrees of freedom tially account for these findings.

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
TABLE 1
THE CHARACTERISTICS OF THE SITES AND THE SURVEY RESPONDENTS SELECTED FROM THE FEDERAL

Characteristics of Survey Respondents

Site Characteristics Predicted


Wage Rate Household Income Visits (T + M
Property Recreation Shore Area
Project name code days miles acres X c X Ca X o- X

Allegheny River 300 - - - 5.45 1.65 15,667 8,625 2.6 2.5 45.19
System, PA
Arkabutla Lake, MS 301 2,011,700 134 52,549 5.23 1.45 13,184 8,974 5.4 2.7 20.04
Lock &Dam No. 2
(Arkansas River
Navigation
System), AR 302 343,700 96 32,415 5.24 1.03 10,409 3.991 6.8 2.0 3.04
Beaver Lake, AR 303 4,882,600 449 40,463 5.59 1.70 18,150 9,946 3.5 3.0 94.55
Belton Lake, TX 304 2,507,000 136 30,789 5.52 1.51 17,279 11,913 6.0 2.8 33.18
Benbrook Lake,
TX 305 1,978,000 37 11,295 5.00 1.21 19,135 10,065 2.3 1.2 30.23
Berlin Reservoir, OH 306 1,179,000 70 7,990 5.44 1.24 16,459 10,161 5.2 2.9 21.15
Blakely Mt. Dam,
LakeOuachita, AR 307 2,104,300 690 82,373 5.24 1.53 17,144 9,524 4.3 2.8 45.39
Canton Lake, OK 308 3,416,500 45 19,797 5.09 1.54 17,392 10,553 4.6 3.2 32.30
Clearwater Lake, MO 309 888,000 27 18,715 5.43 1.38 17,943 8,456 4.0 2.7 50.51
Cordell Hull Dam
&Reservoir, TX 310 2,167,900 381 32,822 5.43 1.58 15,491 9,215 5.7 2.9 29.65
DeGray Lake, AR 311 1,659,700 207 31,800 5.17 1.58 19,235 10,612 4.8 2.7 42.04
Dewey Lake, KY 312 1,116,800 52 13,602 5.83 2.10 18,021 9,559 2.4 2.0 90.75
Fort Randall, Lake
Francis Case, SD 313 4,756,000 540 133,047 5.43 1.69 20,696 11,705 3.3 3.1 100.29
Grapevine Lake, TX 314 5,139,100 60 17,828 5.20 1.58 19,309 10,992 6.3 2.6 38.45
Greers Ferry
Lake, AR 315 4,407,000 276 45,548 5.15 1.45 15,890 8,562 4.7 3.0 54.16
Grenada Lake, MS 316 2,553,900 148 86,826 5.13 1.56 9,199 4,833 6.4 2.6 24.57
Hords Creek Lake,
TX 317 359,500 11 3,027 5.26 1.42 16,263 9,699 4.4 3.0 39.46
Isabella Lake, CA 318 1,489,200 38 15,977 5.64 1.48 15,938 11,445 3.3 2.5 55.59
Lake Okeechobee
and Waterway, FL 319 2,894,584 402 451,000 5.38 1.20 13,849 9,541 4.1 3.0 24.91

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
Lake Washington
Ship Canal, WA 320 712,900 80 169 6.26 2.07 16,686 5,815 3.3 3.0 98.63
Leech Lake, MN 321 950,600 316 162,100 5.90 1.40 18,886 10,986 2.5 1.8 104.08
Melvern Lake, KS 322 2,034,600 101 24,543 5.69 1.65 18,087 9,015 4.3 3.0 31.48
Millwood Lake, AR 323 2,042,300 65 142,100 5.49 1.87 18,630 1,319 5.6 3.0 37.62
Mississippi
River Pool
No. 3, MN 324 1,323,700 37 20,350 6.36 2.23 29,571 10,895 3.0 2.4 99.20
Mississippi
River Pool
No.6, MN 325 645,500 55 11,292 5.79 1.42 19,589 10,693 4.8 3.0 52.23
Navarro Mills
Lake, TX 327 1,111,500 38 14,286 5.16 1.41 13,739 4,652 4.6 2.8 27.68
New Hogan
Lake, CA 328 335,200 44 6,162 5.57 1.28 18,954 11,270 4.0 3.1 34.10
New Savannah
Bluff
Lock &Dam, GA 329 207,600 32 2,030 5.28 1.13 12,609 9,414 5.8 2.7 18.65
Norfork Lake, AR 330 3,066,500 380 54,193 5.65 1.61 17,667 8,889 3.2 2.5 94.89
Ozark Lake, AR 331 1,102,000 173 39,251 5.02 1.22 12,654 7,568 4.9 3.0 58.71
Perry Lake, KS 332 3,388,000 160 41,769 5.52 1.48 16,565 6,925 4.7 2.7 28.79
Philpott Lake, VA 333 1,454,900 100 9,600 5.33 1.55 14,268 6,668 5.8 2.6 26.09
Pine River, MN 334 1,615,100 119 22,177 5.95 1.80 20,097 9,370 2.1 1.4 69.80
Pokegama Lake, MN 335 948,300 53 66,542 5.70 1.46 18,816 9,476 3.3 2.7 100.63
Pomona Lake, KS 336 1,460,400 52 12,301 5.42 1.36 17,265 7,330 5.4 2.8 25.38
Proctor Lake, TX 337 975,200 27 15,956 5.49 1.63 17,510 11,167 5.4 2.9 46.08
Rathbun
Reservoir, TX 338 2,332,200 156 36,072 5.74 1.56 20,543 7,473 4.3 2.9 41.78
Sam Rayburn Dam &
Reservoir, TX 339 2,728,700 560 176,869 5.32 1.35 19,515 11,331 4.1 2.7 40.23
Sardis Lake, MS 340 2,488,900 110 98,590 5.41 1.31 13,141 7,223 6.5 2.3 36.08
Waco Lake, TX 343 3,371,600 60 21,342 5.46 1.25 16,396 12,454 6.9 2.2 33.02
Whitney Lake, TX 344 1,976,400 170 53,230 5.25 1.29 18,688 11,651 5.0 2.8 35.40
Youghiogheny River 345 1,122,600 38 4,035 5.56 1.59 16,682 11,051 5.4 2.9 24.67
Lake, PA

aOne-way distance to the site.


bNumber of observations are based on the final models estimated for site.
NOTES:Xis the arithmeticmean.
cais the standard deviation.
(T + M) cost is the sum of vehicle and time-related costs of a visit.

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
268 Land Economics

TABLE 2
A COMPARATIVEANALYSIS OF PREDICTEDWAGE RATES

PredictedWageRate
Featureof Averageincome
Sample per hour (a) Male Female
A. Annual Household Income
Under5,999 1.50 5.79 4.06
6,000 to 9,999 4.00 5.49 4.10
10,000to 14,999 6.25 6.01 4.38
15,000to 24,999 10.00 6.70 4.39
25,000 to 49,999 18.75 7.17 4.65
50,000 or more 25.00 6.53 4.65
B. Occupation
Professional,technical 10.17 7.89 5.65
&kindredworkers
Farmers 8.34 5.71 2.75
Manager,officials 10.92 7.74 4.94
&proprietors
Clerical&kindredworkers 9.13 5.94 4.10
Sales workers 9.30 6.24 3.29
Craftsmen,foremen 8.17 6.05 4.31
Operatives 7.99 5.15 3.56
Serviceworkers 7.38 4.71 3.18
Laborers,except
farm&mine 7.13 4.74 3.11
aThese estimatesare derived using the midpoint of the rangeof incomes and an assumption of 2,000 hours of
worktime per year.

Vaughan et al. 1982), a semi-log specifica- The variable for the cost of on-site time
tion was selected. includes endogenous and exogenous com-
ponents. It is the product of the wage rate
and the time-on-site during the trip when
In V., = ao + aljMC,, + a2jTCen
each respondent was interviewed. Thus,
when it is included in the model, a simulta-
+ a3 Y, + a4SC'n + EbsjXsn+ E,, [6]
neous equation estimator is required for [6].
We have used nonlinear two stage least
where: squares (N2SLS) (Kelejian 1971). Models
number of trips to site j by indi-
Vin = including and deleting on-site time in this
vidual n; format have been considered for each site.
jMCjn= vehicle-related travel cost of a trip In addition, a wide variety of socioeco-
to site j by individual n (i.e., c d, in nomic variables were considered for inclu-
terms of our earlier theoretical sion in the site demands (see Desvousges,
model); Smith, and McGivney 1982 for details). Ap-
TCjn= travel time costs of a trip to site j proximately half of the "best" specifica-
by individual n; tions for the sites' estimated demand mod-
Yn = family income of individual n; els excluded on-site time costs (i.e., 22 of
SCjn = on-site time costs of a trip to site j the 43 sites). The balance favored a model
by individual n; that included the cost of on-site time. These
Xsn = sth socioeconomic feature of indi- "best" specifications were selected based
vidual n; on the size of each equation's standard er-
Ejn = stochastic error for jth demand ror of estimate and on the agreement of the
function and associated with nth signs of the estimated parameters with a
observation. priori expectations.

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
Smith, Desvousges, and McGivney:Travel Time 269

The demand model in the form given in the survey respondents from a particular
[6] also permits each of the most popular site have the same demand equations. RE-
approaches for estimating the opportunity CIMPcan be interpretedas one measure of
cost of travel time to be represented. TCin individual's taste for water-based recrea-
can be described in either of two forms:the tion. It is a role analogous to age, sex, or
product of the predicted wage and travel other demographicvariables in that it pro-
time, or one-third this quantity (based on vides some basis for relaxing this assump-
Cesario's(1976) proposal). Since our theo- tion of identical demand functions by in-
reticalmodel suggested thattreatingoppor- corporatingproxy variables assumed to be
tunity costs as a multiple of the wage is sim-associated with individual tastes as deter-
ply an approximation,the relevant issue is minants of demand.
finding the specification most consistent While the results for income are not sur-
with the empirical data. prising, given past empirical evidence on
site demands for these types of recreational
V. MODELESTIMATESAND experiences, they nonetheless deserve
TESTRESULTS some comment. Our theoretical model
treats income as full income-including
Our results are divided into two sets both earned income and the monetary
based on whether the demand model value of time that could be "sold" in the
judged to be "best" included the costs of marketfor the wage. Thus, under ideal con-
on-site time. We present the site demand ditions we would like to include measures
estimatesfor each groupseparately.Table 3 of the nonwage income, the wage rate,
reportsthe estimates for the 22 sites whose workingtime (to the extent it is not subject
demand functions were judged to be unim- to individual choice), or the total time
proved by our measure of the costs of on- available for working (if working time is a
site time. They are ordinary least squares choice variable).
(OLS)estimates of each model and main- One significant advantage of the house-
tain that the wage rate is the relevant mea- hold production frameworkis that it high-
sure of the opportunity cost. These final lights the importanceof the institutional ar-
models include, in addition to the travel rangements governing the time and
cost variables and family income, the re- monetary constraints facing individuals.
spondent's age (in years), sex (male = 1), The variables we have identified as supe-
and a measure of the individual's attitude rior to family income would be preferable
toward outdoor recreation, RECIMP(with because they provide some limited insight
one designating individuals who regarded into the natureof these arrangements.Until
outdoor recreation as very important in this type of detailed information is availa-
comparison to their other interests, and ble, we can expect that conventional in-
zero otherwise). However, as the equations come variablesmay not be judged to be sta-
in Table 3 indicate, not all site demands in- tistically significant determinants of the
cluded all of these variables. demandfor goods, such as the use of recrea-
The travel cost variable (including both tion sites, where nonmonetary constraints
vehicle and time costs) is the most consist-
ent determinant of the demand for each l?These estimates imply fairly large site values-
site's services.10 As a rule, income and as measured by the Marshallian consumer surplus
(integrating from the average travel costs to the maxi-
other socioeconomic variables were not mum observed among site's respondents). While
consistently significant determinantsof the these may indicate the possibility of a specification
demands for these sites. Of the remaining error or the presence of an additive rather than a mul-
variables,the attitudinalvariable,RECIMP, tiplicative stochastic error, thereby leading to biased
parameter estimates, it is difficult with existing infor-
appears to influence the demands for a mation to an unambiguous explanation. We
numberof sites. This should not be surpris- are gratefulprovide
to an anonymous referee for bringing
ing. Our model implicitly assumes that all these possibilities to our attention.

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
TABLE 3
TRAVEL COST DEMAND MODELS: ORDINARY LEAST SQUARES
Dependent Variable is Ln (Visits)

T&Mb Income
Site Intercept Cost Income Squared Sex R

Allegheny River 0.53 -0.001 8.2 x 10-6


System, PA' (2.04) (-0.13) (0.74)
Arkabutla Lake, MS 1.58 -0.009 6.2 x 10-6
(9.99) (-3.09) (0.67)
Lock &Dam No. 2 1.94 -0.013 -1.5x10-5 0
(Arkansas River (6.02) (-2.46) (-0.88) (1
Navigation System), AR
Belton Lake, TX 1.69 -0.005 2.6x10-6
(9.38) (-2.47) (0.29)
Berlin Reservoir, OH 1.40 0.001 -4.1 x 10-7
(8.47) (0.43) (-0.05)
Clearwater Lake, MO 1.51 -0.003 -1.Ox 10-5
(5.97) (-1.42) (-1.21)
DeweyLake, KY 0.08 -0.003 2.5x10-5 0.498
(0.36) (-3.67) (2.74) (2.89)
Isabella Lake, CA 1.26 -0.007 7.9x 10-6
(5.55) (-3.15) (0.81)
Lake Okeechobee 1.68 -0.027 1.9X10-7
and Waterway, FL (3.68) (-1.72) (0.01)
Melvern Lake, KS 1.87 -0.008 -6.7x10-5 1.6X10-9
(3.93) (-1.60) (-1.36) (1.50)
Mississippi River 2.12 -0.005 -7.2 x 10-5 1.4X10-9
Pool No. 3, MN (3.90) (-4.22) (-1.63) (1.78)
Navarro Mills Lake, TX 1.66 -0.006 -1.4x 10-5
(6.40) (-1.39) (-1.14)
New Hogan Lake, CA 1.04 -0.004 7.1 x 10-6
(2.58) (-0.41) (0.60)
New Savannah Bluff 1.88 -0.007 -9.8X10-6
Lock &Dam, GA (8.39) 1.44) (0.70)
Ozark Lake, AR 1.42 -0.005 -7.1 x 10-6 0
(5.11) (-4.58) (-0.53) (1
Perry Lake, KS 1.50 -0.004 -1.0x 10-5
(4.17) (-0.74) (-0.68)

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
Smith, Desvousges, and McGivney:Travel Time 271

N in e Go
can be as importantas the monetary in ex-
r-4 CfO eC eC o)co
plaining behavior.
r-4
o CO Before turning to our evaluation of the
rs
N
CO
co
CM
CO
co
CM 00
co co
approaches for measuring the opportunity
costs of time and to the associated test
co
O
C17
r-
CM
0r0
CO
-
O
0
c
r
results, it should be acknowledged that this
o 6 66 6d formulation of the site demand model is
basedon a numberof implicit assumptions.
Space limitations prevent a detailed de-
o scription of the analyses conducted to
c,
gauge the implications of each, but we can
u
summarize the results of our appraisal of
u,U,
one of the most importantof these assump-
U, tions. Travel cost demand models assume
CO
that the trips to each site are single-
.4-4
o
purpose. When these trips are not single-
, ." 2 purpose,the travelcost cannot be treatedas
O .0 the implicit price of one "unit" of the site's
services (i.e., a visit). Rather,it is the cost of
·" 0
all the objectives of the trip. Haspel and
Johnson (1982) have considered this issue
CO for a survey of users of the Bryce Canyon
National Park and found that for this site,
the assumption of single-purpose trips for
0 visitors was inappropriateand would lead
oo
to substantial differences in the estimated
, c. , travel cost demand functions. They pro-
O0 0 0 0 0
pose that distance averaged over the sites
x
CM r
CrvX Cs X C4 X O X e` x e visited in a single trip be used as the dis-
(CD R :" C n e
cO O CO
tance component of the travel cost. These
findingsconfirmresults by Smith and Kopp
(1980) which indicated that estimates of
1a
the travelcost model become unstable with
increases in the distances traveled by visi-
<U
v -r.0au
i -t§ §
Co.- tors."
COcD
OclOOT O OOCDMCM
dCO dco dddc oddd r
The Federal Estate survey does not pro-
I I ^I I I I I 0 ° vide trip itineraries, and therefore it was
CO O0
not feasible to consider the Haspel-Johnson
proposal. However, it was possible to in-
-rC _COh C7O r_ H_O CCD vestigate the implications of distance for
> cc H o c r;o o C' o D
_ d
the stability of the estimated parametersin
ED-cS
these site demand models. The single ob-
(DI >C ·
OCd9 servationregressiondiagnostics developed
by Belsley, Kuh, and Welsch (BKW)(1980)
were calculated for each estimated parame-
P4t
ter in all site demand models. Using the
BKW interval scaling and their DFBETA

"Smith and Kopp(1980)arguedthatthe prospects


X X formultiple destination were an importantreason for
this instabilityof travel cost demand models.

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
272 Land Economics

statistics for each site demand model, our formulate these hypotheses as parameter
results indicated that single observations restrictions. The full-cost hypothesis ((a)
had little influence on the models' esti- above) implies that aij and a2jfrom equa-
mates and did not seem to be related to the tion [6] are equal. The Cesario hypothesis
distance individuals traveled to the sites. implies that a2j = (1/3)alj. Table 4
These findings should not be surprising, reportsthe significance level at which each
because the Corps of Engineers recreation hypothesis can be rejectedusing an F statis-
sites involved would not be considered tic. Seven of the twenty-two sites rejectthe
unique. As Haspel and Johnson acknowl- full-cost hypothesis at a 10% significance
edge, the multiple destination problem is level. The Cesario hypothesis is rejected
especially crucial for recreationalsites that with approximately the same frequency
attractvisits from aroundthe nation.12 (i.e., in nine of the twenty-two site de-
The first two columns in Table 4 report mands' estimates.)13Thus, neither method
the results for our tests of two hypotheses:
(a) the treatmentof the opportunity cost as 12Based on the trip itineraries of the users in their
the wage rate;and (b) the Cesariohypothe- sample, Haspel and Johnson (1982) report that 71% of
sis thatthe opportunitycost is one-thirdthe the Bryce Canyon visitors also visited the Zion Na-
tional Park, 42% visited the Grand Canyon's North
wage rate. Following the McConnell- Rim, and 16% the Grand Canyon's South Rim.
Strand(1981) interpretationof the opportu- "3These results do not include two cases (one for
nity cost as an approximatelyconstantmul- each hypothesis) where the test statistic equals the
tiple of the wage rate, it is possible to critical value.

TABLE 4
TEST RESULTSFORALTERNATIVETREATMENTSOF OPPORTUNITYCOST OF TIME: OLS ESTIMATES

Site Test Significance Level McConnell-


Full-cost Cesario- Strand
hypothesis hypothesis estimates

Allegheny River System, PA .53 .53 -.528


Arkabutla Lake, MS .14 .50 + .046
Lock & Dam No. 2 (Arkansas
River Navigation System), AR .03 .01 -1.465
Belton Lake, TX .04 .04 -.401
Berlin Reservoir, OH .43 .56 -.568
Clearwater Lake, MO .40 .64 -.189
Dewey Lake, KY .16 .03 + 4.385
Isabella Lake, CA .61 .99 +.369
Lake Okeechobee &
Waterway, FL .59 .26 + 1.831
Melvern Lake, KS .41 .87 +.189
Mississippi River Pool
No.3, MN .99 .10 +1.059
Navarro Mills Lake, TX .64 .50 -2.585
New Hogan Lake, CA .72 .66 -.785
New Savannah Bluff
Lock & Dam, GA .49 .51 -.273
Ozark Lake, AR .03 .01 + 80.000
Perry Lake, KS .73 .89 +.213
Pine River, MN .02 .01 -2.000
Pokegama Lake, MN .35 .11 + 2.542
Pomona Lake, KS .13 .18 -1.818
Rathbun Reservoir, TX .01 .03 -.039
Sardis Lake, MS .01 .01 + 28.833
Waco Lake, TX .01 .01 -9.417
Youghiogheny Lake, PA .10 .04 + 4.322

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
Smith, Desvousges, and McGivney:Travel Time 273

is unambiguously superior to the other as ditional visit. This relationship would im-
an approach for approximatingthe oppor- ply a negative association. Finally, we
tunity cost.14 should acknowledge that our model is an
We can also use the estimates derived for approximation,and the assumption of con-
our test of the full-cost hypothesis to esti- stant on-site time for all visits may well be
mate the multiple relating the opportunity inconsistent with the actual behavior of
to the wage rate. Following McConnell and many respondents. In view of these quali-
Strand, this is given by (a2j/a1j).The last fications, the sign of the parameterfor the
column in Table 4 reports these estimates. costs of on-site time cannot be determineda
Our results contrast with the McConnell- priori.Ourresults are consistent with these
Strandestimates for one site as well as with possibilities and exhibit both positive and
their experimental findings. Twelve of the negative associations, with all of the statis-
twenty-two estimated values for this multi- tically significantestimates implying a neg-
ple were negative. Moreover,of the remain- ative relationship.
ing estimates, a majoritywere greaterthan Testingthe full cost and Cesariohypoth-
unity. In some cases, the estimates would esis within a simultaneous equations for-
seem completely implausible (e.g., the mat is not a direct application of the con-
Ozark Lake and Sardis Lake sites). These ventional F-statistic. These hypotheses
results may well be due to the high degree have been evaluated using a test statistic
of collinearity between our estimates of thatfollows on asymptotic F distribution.It
vehicle-relatedtravel costs and the costs of is derived from the generalized least
traveltime. Indeed, as a rule, it was not pos-
sible to obtain precise estimates of the ef- 14There is an additional problem with the use of
fects of these individual components of the the conventional travel cost methodology with these
travelcost-often the sign of one of the esti- data. Since the survey describes the behavior of a
mated parameters(i.e., a1jand a2i) contra- sample of users of each site, we do not observe indi-
viduals with less than one visit. This implies that our
dicted theoretical expectations. semi-log dependent variable is truncated at zero and
Table 5 reports the nonlinear two-stage that OLS estimates of the site demand models will be
estimates of the remaining twenty-one site biased (see Heckman 1976). However, the magnitude
demands models, where the inclusion of of the bias will depend upon the distribution of the
visits to the site. If the average number of trips to the
the costs of on-site time was judged to im- site is large and the variance fairly small, we can ex-
prove the models.15These results are based pect the truncation at one to have little impact on the
on the use of the wage rate as the opportu- estimates.
nity cost of time. The estimated coefficient To gauge the importance of this problem, we used
for travel costs (TC + MC)is quite compa- Olsen's (1980) scaling coefficient for converting OLS
estimates to approximate maximum likelihood esti-
rableto the OLSestimates of a model with- mates. Several of the sites where truncation effects
out the variable reflecting on-site time were judged to be serious also appeared to be better
costs. As indicated in Table 5, the mea- modeled with the simultaneous framework recogniz-
sured demand responsiveness to changes ing the role of the costs of on-site time. For the bal-
ance, the mix of support for the full cost and Cesario
in travel costs is quite stable. The coef- proposals was approximately equal. Thus, this quali-
ficient for the costs of on-site time may fication would not affect the overall character of our
reflect two opposing effects. On-site time conclusions. For more details on the nature of these
serves as a substitute for visits to a site in findings see Desvousges, Smith, and McGivney
producingrecreationalservices. Thus, this (1982), chapter 7.
"The first stage instruments were composed of the
effect would imply a positive influence, included predetermined variables along with sex and
since increases in the costs of on-site time a qualitative variable to reflect whether the recrea-
should lead, ceteris paribus,to the substitu- tional activities included camping. Several sets of
tion of more visits for on-site time per visit. first stage regressors were considered (see chapter 7 of
However, our model also maintains that Desvousges, Smith, and McGivney (1982) for further
details). The results reported here are for the form that
on-site time is constant for all visits and provided the best results in terms of the sign and sta-
thus is a component of the price of each ad- tistical significance of the estimated parameters.

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
274 Land Economics

TABLE 5
TRAVELCOSTDEMANDMODELS:Two STAGELEASTSQUARES
EndogenousVariablesare On-SiteTime and Ln (Visits)a

Site
Site Name Intercept T+ M Cost Costb Income Age R2 DF

Beaver Lake, AR 1.71 -0.006 -0.0003 -3.0x10-6 -0.010 0.42 221


(11.45) {-8.12) (-2.21) (-0.61) (-0.27)
Benbrook Lake, TX 2.00 -0.005 -0.0001 3.0x 10-6 -0.002 0.31 41
(6.05) (-3.84) (-0.55) (0.34) (-0.29)
Blakely Mt. Dam, 1.72 -0.008 -0.0002 -9.0X10-6 0.005 0.21 86
Lake Ouachita, AR (3.08) (-4.57) (-0.40) (-0.81) (0.78)
Canton Lake, OK 1.79 -0.017 -0.0008 5.0 x 10-6 0.004 0.26 69
(5.83) (-2.83) (-0.80) (0.47) (0.58)
Cordell Hull Dam &
Reservoir, TX 1.60 -0.014 -0.0002 3.0x10-6 0.007 0.35 99
(7.58) (-4.53) (-0.48) (0.35) (1.71)
DeGray Lake, KY 1.59 -0.008 -0.0004 -2.0x10-6 0.010 0.21 44
(4.51) (-3.24) (-1.38) (-0.21) (1.44)
Ft. Randall, Lake
Francis Case, SD 1.78 -0.004 -0.0021 9.0X10-6 -0.009 0.38 45
(4.74) (-2.43) (-2.28) (0.79) (-0.94)
Grapevine Lake, TX 2.15 -0.005 -0.0001 9.0X10-6 -0.013 0.50 87
(12.73) (-5.06) (-2.31) (1.72) (-2.91)
Greers Ferry Lake, AR 1.61 -0.007 2.0X10-5 9.0x 10-6 -0.005 0.28 212
(10.60) (-7.70) (0.08) (1.48) (-1.09)
Granada Lake, MS 1.55 -0.007 -0.0016 1.0x10o-5 0.010 0.26 70
(5.06) (-2.86) (-2.64) (0.71) (1.99)
Hords Creek Lake, TX 1.94 -0.005 -0.0001 -2.0x10-5 -0.003 0.20 49
(5.81) (-2.06) (-0.19) (-1.77) (-0.36)
Lake Washington
Ship Canal, WA 0.51 -0.004 0.0465 2.0x10-5 -0.002) 0.21 32
(0.66) (-2.40) (1.07) (0.78) (-0.15)
Leech Lake, MN .29 -0.003 0.0004 1.1x 10-5 0.007 0.17 43
(0.79) (-2.33) (1.26) (0.95) (0.93)
Millwood Lake, AR 0.83 -0.009 9.0X 10-5 2.0x10-5 0.013 0.30 48
(2.33) (-3.88) (0.02) (2.54) (1.88)
Mississippi River Pool,
No. 6, MN 1.20 -0.006 -0.0006 2.0 x10-5 0.007 0.24 65
(3.81) (-3.24) (-1.66) (1.95) (0.97)
Norfork Lake, AR 0.67 -0.006 -0.0008 1.0x10-5 0.015 0.20 37
(1.65) (-2.85) (-0.28) (0.91) (1.55)
Philpott Lake, MN 2.21 -0.007 -0.0007 2.0X10-6 -0.008 0.47 33
(7.21) (-3.75) (-1.49) (0.18) (-1.07)
Pokegama Lake, MN 1.34 -0.003 -0.0004 -9.0x10-6 0.002 0.24 63
(3.62) (-3.54) (-1.32) (-0.86) (0.34)
Proctor Lake, TX 1.78 -0.015 0.0003 2.0x 10-6 0.005 0.56 47
(6.47) (-5.99) (0.84) (0.33) (1.01)
Sam Rayburn Dam &
Reservoir, TX 1.16 -0.010 -0.0001 2.0x10-6 0.010 0.17 62
(3.38) (-2.92) (-0.29) (0.19) (2.00)
Whitney Lake, TX 1.53 -0.003 -0.0009 6.0X10-6 0.007 0.10 196
(8.30) (-1.44) (-4.15) (1.11) (1.88)

aThe numbers in parentheses below the estimated coefficients are asymptotic t-ratios for the null hypothesis
of no association.
bSite cost is the product of time spent on site and the hedonic wage rate.

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
Smith, Desvousges, and McGivney:Travel Time 275

squares interpretation of two-stage least that the time constraints facing individual
squares for linear simultaneous equation recreationistsare complex and that simple
models.16 Since we are applying it for the approximations to relate the opportunity
case of nonlinear two-stage least squares, cost of time to the wage ratewill not be able
we have conjectured that the statistic is also to accommodate all applications. While
valid in this case. As a consequence, we our theoretical analysis provides formal
have used generous significance levels in supportfor Cesario'sconcerns for the valu-
interpreting the test results to allow for dis- ation of travel time, the empirical findings
crepancies between the small sample and indicate that where the costs of on-site time
asymptotic distributions of the test statis- can be omitted fromsite demand functions,
tic. Nonetheless, the results in Table 6 are Cesario's proposal is not unambiguously
quite clearcut at either the 10% or the more superior to using the wage rate as the op-
restrictive 5% levels. The majority of the portunity cost. Indeed, until better infor-
sites' results imply that both hypotheses mation on the natureof the time constraints
should be rejected. facing individuals in their recreation deci-
Taken as a whole, these results imply sions is available,the wage rateprovides an

"6Thetest statistic used for these results is derived


TABLE 6 from the GLS interpretation of two-stage least
TESTRESULTSFORALTERNATIVE
TREATMENTS squares.For example, if an equation from a linear si-
multaneousmodel is given as:
OF OPPORTUNITY
COST OF TIME: N2SLS ESTIMATESa y= YA+ZiB+e
with y a Txl vector and Y,A, Zl, B, and e conformably
Test SignificanceLevels dimensionedmatricesand vectors of included endog-
enous variablesand their parameters,included exog-
Full-Cost Cesario enous variables and their parametersand stochastic
Site error.
Hypothesis Hypothesis
Premultiplyingby the matrixof all predetermined
BeaverLake,AR .90 .01 variables,Z leads to
BenbrookLake,TX .01 .01 ZTy = ZTyA + ZTZiB + ZTe
BlakelyMt. Dam, .01 .01
LakeOuachita,AR The covariance matrix for ZTewill be r2(ZTZ)(if the
CantonLake,OK .02 .01 covariance matrix for e is r2I)and two-stage least
CordellHull Dam .01 .85 squaresis the Aitken GLSestimatorof this model.
&Reservoir,TX Using this formatthe test statistic is given as fol-
De GrayLake,AR .06 .04 lows:
Ft. Randall,Lake .71 .01
FrancisCase,SD df eZTZ (_ - TZ ZTZ) Z
lZ
GrapevineLake,TX .01 .01
GreersFerryLake,AR .01 .01 d\ e TZ(ZTZ) IZTU
GranadaLake,MS .01 .01
HordsCreekLake,TX .01 .01 where
LakeWashington .01 .01 e, = vector of 2SLS residuals estimated sub-
Ship Canal,WA ject to restriction;
LeechLake,MN .01 .01 &e= vector of 2SLS residuals estimated with-
Millwood Lake,AR .08 out the restriction;
.79
df, = degreesof freedom parameter(takento be
Mississippi RiverPool the number of observations less the esti-
No. 6, MN .01 .01
NorforkLake,AR .01 mated parametersin the second stage);
.01
PhilpottLake,VA .10 .01 df2 = numberof parameterrestrictions.
ProctorLake,TN .01 .01 The test statisticand degreesof freedomselection (es-
SamRayburnDam .01 .65 pecially df,) were selected based on Basmann's(1960)
&Reservoir,TX proposals for alternatives to the identifiability test
WhitneyLake,TX .01 .01 statistic. His findings suggested that an F-statistic
similarto this test provided a more appropriatetest of
aThesesignificance levels are for the asymptotic over-identifyingparameterrestrictions than the con-
distribution. ventional asymptotic Chi-squarestatistic.

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
276 Land Economics

equallyplausible approximationforthe op- for our analysis, simply do not provide this
portunitycost of travel time. information. Indeed, the Federal Estate
component of the 1977 Nationwide survey
VI. IMPLICATIONS is amongthe most detailed data sets availa-
ble for estimatingtravel cost recreationde-
This paperhas considered both the theo- mand models.
retical basis and empirical support for the The empirical findings of our analysis of
principal methods proposed for measuring forty-three site demand models suggests
the opportunitycost of travel time in mod- that:(1) the treatmentof the costs of on-site
els of the demand for recreation sites. Our time is importantto the empirical analysis
results indicate that a redirection in the re- of the opportunity cost of travel time; (2)
search on modeling the demand for out- the differences between the full-cost and
door recreation sites is warranted. More Cesarioproposals for measuring these op-
specifically, the theoretical analysis im- portunity costs are not sufficient to prefer
plies that the definition of the natureof and the latterto the former;and (3) the estima-
the interrelationshipbetween an individu- tion of the multiple relatingthese opportu-
al's time constraints plays a central role in nity costs to the wage is not a trivial task.
the determination of the individual's cor- Ourfindings suggest that moving beyond a
respondingopportunitycost of travel time. judgmentalguess for this multiple will be
This result was clear from our simple difficult within the constraints of existing
model, which distinguished two types of data.
time constraints in order to reflect the im- Based on our findings, two areas of re-
plications of flexibility in the availability search offer promise. The firstentails addi-
(by time-of-day or whatever the relevant tional empirical appraisal of these ap-
time horizon) of that time. In making their proaches for estimating the opportunity
actual recreation decisions, different indi- cost of time with different types of recrea-
viduals will experience differentdegreesof tional sites, especially those appealing to
flexibility in the use of their free time. The national markets.The second involves im-
relevant time constraints for recreation proving the projections of wage rates as-
choices may well depend upon the type of signed to individuals to reflect explicitly
site used and the recreationalactivities un- the amount of flexibility and discretion
dertaken.For "local" sites, serving the rec- availablein the use of nonworktime and its
reationdemands of individuals within a re- effectas a compensatingdifferentialto mar-
gion surrounding the site, the time ket wage rates.This approachmay offerthe
constraints and associated opportunity most direct means of improving the availa-
cost of time may well be differentfrom that ble measures of the opportunity cost of
of sites serving national markets. Often at time.
greatdistances from a substantial subset of
theirvisitors, these "national"sites require
the allocation of different "types of time"
References
(i.e., annual vacation or leaves). The rele-
vant measure of the opportunity costs of Basmann, R. L. 1960. "On Finite Sample Distri-
time can be expected to be different for butions of GeneralizedClassicalLinearIden-
these sites. tifiability Test Statistics." Journal of the
To deal adequately with the problems American Statistical Association 55 (Dec.)
650-59.
posed by a theoreticallyappropriatedefini- Becker,G. S. 1965. "A Theory of the Allocation
tion of the opportunity cost of time-both of Time." Economic Journal75 (Sept.):493-
traveland on-site-will requiredetailed in- 517.
formation on the nature of the time con- . 1974. "A Theoryof Social Interactions."
straints facing each individual recreation- Journalof Political Economy 82: 1063-93.
ist. Present surveys, such as the one used Belsley, DavidA.; Kuh,Edwin;and Welsch, Roy

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
Smith, Desvousges, and McGivney:Travel Time 277

E. 1980. Regression Diagnostics. New York: Dummy Variablesin SemilogarithmicEqua-


JohnWiley and Sons. tions: Unbiased Estimation."Economic Let-
Bishop, RichardC., and Herberlein,Thomas A. ters 10:77-9.
1979. "Measuring Values of Extramarket Goldberger,A. S. 1968. "TheInterpretationand
Goods: Are Indirect Measures Biased?" Estimation of Cobb-Douglas Functions."
AmericanJournalof AgriculturalEconomics Econometrica36 (July-Oct.):464-72.
61 (Dec.):926-30. Haspel, Abraham E., and Johnson, F. Reed.
Bockstael,Nancy E., and McConnell,K. E. 1981. 1982. "MultipleDestinationTripBias in Rec-
"Theory and Estimation of the Household reationBenefitEstimation."LandEconomics
Production Function for Wildlife Recrea- 58 (Aug.):364-72.
tion." Journal of Environmental Economics Heckman,JamesJ. 1976. "The Common Struc-
and Management8 (Sept.):199-214. ture of Statistical Models of Truncation,
Burt,OscarR., and Brewer,Durward.1971. "Es- Sample Selection, and Limited Dependent
timationof Net Social BenefitsFromOutdoor Variablesand a Simple Estimatorfor Such
Recreation."Econometrica 39 (Sept.): 813- Models." Annals of Economic and Social
28. Measurement5 (Fall):475-93.
Cesario,FrankJ. 1976. "Valueof Time in Recre- Kelejian, Harry H. 1971. "Two Stage Least
ation Benefit Studies." Land Economics 52 Squaresand EconometricSystems Linearin
(Feb.):32-41. Parameters but Nonlinear in Endogenous
, and Knetsch,JackL. 1970. "TimeBias in Variables."Journalof the American Statisti-
Recreation Benefit Estimates." Water Re- cal Association 66 (June):373-4.
sources Research6 (June):700-4. McConnell,K. E. 1975. "SomeProblemsin Esti-
Cicchetti, Charles J.; Fisher, Anthony C.; and matingthe Demandfor OutdoorRecreation."
Smith, V. Kerry.1976. "An Economic Evalu- AmericanJournalof AgriculturalEconomics
ation of a Generalized Consumer Surplus 57 (May):330-34.
Measure: The Mineral King Controversy." .1976. "SomeProblemsin Estimatingthe
Econometrica44 (Nov.):1259-76. Demand for Outdoor Recreation: Reply."
; Seneca, Joseph J.; and Davidson, Paul. AmericanJournalof AgriculturalEconomics
1969. The Demand and Supply of Outdoor 58 (Aug.):598-9.
Recreation.New Brunswick, N.J.:Bureau of , and Strand,Ivar. 1981. "Measuringthe
EconomicResearch,RutgersUniversity. Costof Time in RecreationDemandAnalysis:
De Serpa, A. C. 1971. "A Theory of the Eco- An Application to SportFishing."American
nomics of Time." Economic Journal 81 Journalof AgriculturalEconomics 63 (Feb.):
(Dec.):828-46. 153-6.
Desvousges, William H.; Smith, V. Kerry;and Nichols, Len M.; Bowes, Marianne;and Dwyer,
McGivney,Matthew P. 1982. A Comparison JohnF. 1978. "ReflectingTravelTime in Tra-
of Alternative Approaches for Estimating vel-Cost-BasedEstimates of Recreation Use
Recreation and Related Benefits of Water and Value."ForestryResearchReportNo. 78-
Quality Improvements.Draft report to U.S. 12, AgriculturalExperimentStation, Univer-
EnvironmentalProtection Agency, Research sity of Illinois-Champaign(Dec.).
TriangleInstitute,July. Olsen, RandallJ. 1980. "Approximatinga Trun-
Deyak,Timothy A., and Smith, V. Kerry.1978. cated NormalRegressionwith the Method of
"Congestion and Participation in Outdoor Moment." Econometrica 48 (July): 1099-
Recreation: A Household Production Ap- 106.
proach." Journal of Environmental Eco- Smith, V. Kerry. 1975. "Travel Cost Demand
nomics and Management5 (Mar.):63-80. Models for Wilderness Recreation:A Prob-
Dwyer,J. F.;Kelly, J. R.;and Bowes, M. D. 1977. lem of Non-Nested Hypotheses." Land Eco-
Improved Procedures for Valuation of the nomics 51 (May):103-11.
Contributionof Recreation to National Eco- . 1983. "The Role of Site and JobCharac-
nomic Development. Urbana-Champaign: teristics in Hedonic Wage Models." Journal
Universityof Illinois. of UrbanEconomics (May)(forthcoming).
Freeman,A. Myrick III. 1979. The Benefits of , and Kopp, RaymondJ. 1980. "The Spa-
Environmental Improvement: Theory and tial Limits of the Travel Cost RecreationDe-
Practice.Baltimore:The Johns Hopkins Uni- mand Model." Land Economics 56 (Feb.):
versity Press. 64-72.
Giles, D. E. A. 1982. "The Interpretation of Vaughan, William J.; Russell, Clifford S.; and

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions
278 Land Economics

Hazilla,Michael. 1982. "A Note on the Use of Ziemer,R.;Musser W. N.; and Hill, R. C. 1980.
Travel Cost Models With Unequal Zonal "Recreational Demand Equations: Func-
Populations:Comment."LandEconomics 58 tional Formand ConsumerSurplus."Ameri-
(Aug.):400-7. can Journal of Agricultural Economics 62
Wilman,ElizabethA. 1980. "TheValue of Time (Feb.):136-41.
in RecreationBenefitStudies."Journalof En-
vironmental Economics and Management7
(Sept.):272-86.

This content downloaded from 195.194.14.7 on Thu, 10 Jul 2014 08:39:23 AM


All use subject to JSTOR Terms and Conditions

You might also like