Problem of Identical Attribute Assignment in Conjoint Experiment

• In Event: Improving Survey Experiments and Analysis

Abstract

The identical attribute assignment problem in conjoint experiment refers to the case where a nonignorable portion of tasks contain identical attribute levels in both profiles. In that case, AMCE, the popular causal estimand in conjoint experiment, is equivalent to AMCE of heterogeneous attribute assignments weighted by the proportion of heterogeneous attribute assignments. AMCE of identical attribute assignments is either zero by construction (in the unit-wise computation) or not identifiable (in the pair-wise computation). We report this puzzling problem that has not been discussed in the literature and discuss a feasible methodological approach that resolves this problem in causal inference using conjoint experiment data.

Authors

• Jong Hee Park, Seoul National University

• Joonseok Yang, Sungkyunkwan University

• Sung Eun Kim, Korea University

• Inbok Rhee, KDI School of Public Policy and Management