Bayesian Approach for Identifying Efficient Stated-Choice Survey Designs with Reduced Prior Information

Abstract

This paper presents the results of a stated-choice survey design in which reduced information is available about the values of the parameters. The main objective of the survey is to develop discrete choice models to analyze the carrier selection process by shippers. The only information available is the expected sign of the parameter in the utility function. Therefore, the use of a Bayesian approach is necessary in analyzing the efficiency of potential survey designs. The measure of efficiency adopted is the 95th percentile value of the D-error, the determinant of the asymptotic variance–covariance matrix, obtained via Monte Carlo simulation. This measure would maximize the expected gain of information with the experiment. In the survey, the respondents (shippers) will be asked to provide the main carrier's attributes used in the selection process. On the basis of the answers, the survey will be customized to each respondent (situation). In this paper, only the results of the design for situations in which shippers select continuous attributes are presented. An algorithm is developed in a statistical package to search and evaluate the efficiency of 1,000 randomly selected potential survey designs. The algorithm maintains the 10 most efficient designs to be evaluated and uses box plots to identify the most suitable design. The results show that the approach adopted improves the efficiency of the design substantially, which would result in more accurate models from the survey data.