Changing the lens: The contingency of results from conjoint experiments on the outcome variable and the estimand

Conjoint experiments have become popular in political science for studying opinions, attitudes, and preferences on various issues. While the methodological literature discusses two dependent variables—forced-choice and rating outcomes—many studies continue to use (or report) only the former. Additionally, many studies primarily focus on analyzing causal quantities—Average Marginal Component Effects (AMCEs) and do not report the descriptive estimates—Marginal Means (MMs). This article highlights the contingency of results from conjoint experiments on the outcome variable and the estimand used. It calls for the inclusion of the rating outcome and for reporting the MMs alongside AMCEs. As the two outcome variables elicit distinct preferences, it explains how relying solely on one may obscure important findings and limit the insights gained from the experiment. This is particularly consequential for the analysis of MMs. These arguments are demonstrated by replicating and reanalyzing recently published conjoint studies. The article concludes with practical recommendations for applied researchers.