Using Individual Effects from Conjoint Experiments to Measure Preferences

• In Event: Conjoint Design in Focus: Practical and Theoretical Advances

Abstract

Due to their flexibility and power, conjoint experiments have quickly become a popular tool in political science. One recent methodological advance was the development of a procedure to estimate individual marginal component effects (IMCEs) from conjoint experiments. As opposed to average marginal component effects (AMCEs), IMCEs measure individual-level rather than aggregate-level preferences. The estimation of IMCEs allows researchers to explore effect heterogeneity and to use revealed individual-level preferences as predictors in inferential analyses. At the same time, the method to estimate IMCEs -- respondent-specific OLS regressions -- comes with recommended adjustments to the standard conjoint design. These include the use of interval outcomes and maximizing the number of rated profiles in a conjoint task. In this paper, we explore whether these restrictions are truly necessary. Using data from a candidate choice conjoint experiment implemented in Germany in 2017, we show that IMCEs can be successfully estimated even for forced-choice conjoint designs with only five profile pairs. Then, we use multinomial logistic regression to show that revealed preferences measured via conjoint IMCEs predict real-world party choice. In other words, estimated IMCEs from relatively short conjoint tasks are reliable enough to be useful as predictors in inferential analysis. Finally, since the OLS method has known limitations when applied to binary outcomes, we explore how it compares to logit and probit for the estimation of IMCEs in forced-choice conjoint designs.

Authors

• Kirill Zhirkov, University of Virginia

• Fabian Guy Neuner, Arizona State University