Inattentive respondents cause attenuation bias in conjoint experiments

In forced choice conjoint experiments, in expectation, the respondents should be picking between two answer options at the same rate thanks to randomization of profile attributes. Using replication data, I demonstrate that this empirical regularity does not always hold across various political science experiments: respondents pick the first option more often. I employ data from seven original conjoint experiments to show that it is inattentive respondents who are driving this result. I derive that this primacy bias leads to attenuation of estimates. Finally, I present a research design that pins down the mechanism behind primacy bias and elucidates whether it is the product of the language of instruction or survey layout.