I often caution students contemplating using self-reported data and urge them to consider observational data where possible (and appropriate to the research question and design). Why Bother Asking? The Limited Value of Self-Reported Vote Intention, by Rogers, Todd Rogers (Harvard--Kennedy School) and Masa Aida (Greenberg Quinlan Rosner), contribute empirical support to my intuition, certainly as it relates to predicted voting behavior. The findings' implications for an array of sub-fields are important. An excerpted abstract follows.
"Phone surveys from three elections, including one survey experiment, are analyzed to compare respondents’ pre-election vote intention with their actual voting behavior using administrative records (N=29,403). Unsurprisingly, many who predict that they will vote actually do not vote. More surprisingly, many who predict that they will not vote actually do vote (29% to 56%). Records of past voting behavior predicts turnout substantially better than self-prediction. Self-prediction inaccuracy is not caused by lack of cognitive salience of past voting, or by inability to recall past voting. Moreover, self-reported recall of turnout in one past election predicts future turnout just as well as self-prediction. We discuss implications for political science research, behavioral prediction, election administration policy, and public opinion."