Walsh, Johnson, and McKean
Issues in Nonmarket Valuation ɪ 81
majority of estimates clustered near the bot-
tom of the range in values and a relatively few
extremely high estimates. This substantially
increases the sample mean, and thus it is ques-
tionable whether the mean truly reflects the
sample as a whole. The median would be a
more appropriate measure to use if the purpose
of the analysis is to determine a representative
estimate.
Second, the approach does not reveal what
is causing the extreme range in values, whether
variation in characteristics of users, quality of
sites, or research methods. A potentially useful
approach to the data transfer problem would
be to pool the data from existing studies and
apply multiple regression analysis. If the basic
model specification is complete, that is, if it
includes all the relevant explanatory variables
in the correct functional form, then it could
explain the variation in benefits embodied in
differences among the explanatory variables.
The net benefit estimated for a site lacking data
would then be predicted by inserting appro-
priate values of explanatory variables into the
model fitted to data from the other study sites.
Theoretical Basis for an
Empirical Model
The empirical model used to explain the vari-
ation in benefit estimates should be based pri-
marily on applied microeconomic theory
(McKean and Walsh; U.S. Department of the
Interior). In an ordinary demand function for
a recreation site, the dependent variable to be
explained is the quantity demanded. The list
of independent variables that influence de-
mand includes a proxy for direct cost or price
and such factors as travel distance or the value
of time, the price and availability of substi-
tutes, consumer income, other socioeconomic
variables such as age, quality or attractiveness
of the site, population of the consuming group,
individual taste or preference, and expecta-
tions or experience with respect to crowding.
Other variables related to research method may
include: recreation activity; sample size and
coverage; CVM, TCM, or other method; sta-
tistical model; econometric estimators; type of
CVM question; and site administration.
The possible effect of the specification of
each of these variables should be carefully
evaluated. For example, measurement of
quantity demanded in different units may af-
fect the benefit estimate, whether trips, hours,
visitor days per person or per capita. Choice
of travel cost measurement as distance mul-
tiplied by variable travel cost per mile from
the U.S. Department of Transportation or re-
ported by respondents may also affect benefit
estimates (Duffield). The effect of travel time
cost on benefit estimates has been shown to
vary with the percent of wage rate used
(McCollum, Bishop, and Welsh). Shaw con-
siders the effect of sample truncation and re-
lated problems of on-site surveys. Smith and
Kaoru make an important contribution to un-
derstanding the effects of alternative methods
of estimating travel time cost, presence of a
substitute price term, use of a regional model,
type of site studied, functional form (linear,
log-linear, or semilog), and estimators (ordi-
nary least squares, generalized least squares,
or maximum likelihood-logit-tobit) used in
TCM studies. They conclude that these meth-
odological variations significantly affect ben-
efit estimates. The question remains whether
method would have the same effect in a regres-
sion model holding constant the effects of other
potentially important variables.
In the future, it seems likely that an ever
larger number of studies will be accumulated
on the demand for outdoor recreation. In this
event, each subsequent work in the growing
science of reviewing research can examine
many possible variables that might be impor-
tant and provide a basis for eliminating some
of them as serious candidates for new research.
Using prior reviews to reduce the number of
experimental variables should improve the
statistical analysis and allocation of resources
to new studies. Thus, each succeeding litera-
ture review should build upon previous ones.
In the early stages of this evolving process,
the critical problem will be to correctly specify
the variables that are expected to influence the
benefit estimates. For if important determi-
nants are omitted, the statistical equation will
not predict effects accurately, as illustrated by
Allen, Stevens, and Barrett. Thus, the early
review efforts should be treated with caution,
since by leaving important variables out of the
regression analysis, more or less of the varia-
tion may be attributed to those that are in-
cluded than would be the case with a more
complete specification, as illustrated by Smith
and Kaoru.