Econometric Estimation of Rank Ordering Data
from Stated Preference Surveys
    
John E. Calfee and Clifford Winston
   
    
Daniel McFadden's seminal work in qualitative choice has facilitated
the measurement of consumer valuations of the attributes of a variety
of goods and services.  Most of these measurements use market data
derived from revealed preferences. There are many contexts, however,
where researchers are interested in ascertaining consumer preferences,
but are unable to derive them from market outcomes.  In these
situations, researchers have developed surveys that directly ask
consumers for their valuations or that ask them to state their
preferences given a hypothetical set of alternatives.  The first
approach, commonly referred to as contingent valuation, has been
attacked by many analysts and can be fatally flawed.  (The most severe
problem is termed embedding where consumer valuations are invariant to
quantities consumed.)  The second approach called stated preference
analysis is not free of methodological concerns, but does overcome
most of the problems of contingent valuation.
    
The central empirical issues that arise when conducting a stated
preference analysis are designing a survey that generates reliable
orderings and obtaining consistent and efficient parameter
estimates.  This paper explores these issues in the context of
estimating travelers' value of automobile travel time---a critical
input to determining optimal highway congestion tolls.  Although
there is a long line of value of time estimates, they have usually
been derived from mode choice models and thus incorporate
travelers' disutility for alternative modes such as transit.  This
disutility is irrelevant for our purposes; that is, we are simply
interested in how much automobile travelers are willing to pay to
save time.  This is difficult to infer from market transactions;
thus we conducted a stated preference analysis to estimate this
figure.  In the process of carrying out this analysis, we indicate
how reliable preference orderings can be obtained by comparing
estimates obtained from "ratings" data and "rankings" data, and we
explore the sensitivity of parameter estimates to alternative
estimation procedures such as ordered probit and the Beggs,
Cardell, Hausman multinomial logit approach.