Although consumer researchers often assume that survey responses reflect true beliefs, attitudes, and intentions, most recent research in social judgment and behavioral decision making suggests that people often construct answers to survey questions on the basis of their responses to earlier items. A framework proposed by Feldman and Lynch suggests that answers are most likely to be constructed when the respondent does not have other, more diagnostic inputs accessible in memory. Two field experiments conducted during the 1988 and 1992 presidential elections support this model. In experiment 1, when a respondent had voted for one of the candidates in the primary, voting intention was not based on earlier survey answers; when a respondent had not voted for these candidates in the primary, voting intention appeared to be constructed. Regardless of primary voting behavior, all respondents appeared to construct issue opinions. The effect of earlier answers on issue opinions decreased as the election neared, consistent with the idea that knowledge about the issues increased with time. In experiment 2, placing questions tapping primary voting behavior first affected responses to questions about specific issue opinions, with resulting carryover to a later measure of voting intention. We discuss the implications of these results for consumer judgment processes and for measurement of consumers' attitudes and intentions.