Laura Felone, Khasan Redjabov and Eli August (University of Wisconsin)
Abstract: We replicate Teele et al.’s (2018) substantively important question about political representation with a recent methodological development on conjoint survey experiments by Abramson et al. (2019, working paper). We apply bounds proposed by Abramson et al. to Teele et al.'s results to demonstrate potential issues with making causal claims based off average marginal component effects (AMCEs). AMCEs have potential skew caused by ranking feature preferences among respondents, which results in AMCEs being unable to provide generalizable insights about absolute levels of preferences between groups. Abramson et al.’s bounds provide the reasonable proportion of population support for a given feature over another within a single category. When applied to Teele et al.’s data, we find that none of the features on their conjoint survey have a lower bound of 0.51, indicating that we cannot accurately guarantee a reporting of a majority preference for these features if we rely on AMCEs.