I sometimes wonder how undecided undecided voters really are. Specifically, I wonder whether implicit attitudes that are at least partially independent from self-reported attitudes can predict the behavior of undecided voters. Since the late 70s, psychologists have produced more and more evidence that our introspective experience of our own higher-order cognitive processes and attitudes is highly inaccurate. We don't have direct access to these processes and attitudes, but instead produce theories about them from our behavior, and if our conscious beliefs about them are accurate, it is only because we've come up with a good theory. For this reason, psychologists have produced a wealth of indirect measures of cognitive processes and attitudes that do not rely on subjects' self-reports. One species of indirect tests is designed to test for "implicit attitudes," or attitudes that the subject may not be aware of, and may not accord with their self-reported, or even experienced attitudes. Some of these implicit attitude tests have been shown to be able to predict behavior (e.g., consumer behavior) quite well. I wonder, then, if they might also be able to predict voting behavior, particularly the voting behavior of undecideds.
To see how this would work, I'll briefly describe one implicit attitude test. This test, called the Evaluative Movement Assessment, or EMA1, is built around the idea that people are motivated to approach positively evaluated stimuli, and avoid negatively evaluted stimuli. In addition, for some strange reason (don't ask me why; it's voodoo), these approach/avoidance motivations are active when people are asked to move something toward or away from their names2. In EMA, the participant's name is placed in the center of a computer screen, while words appear on either side of her name. The participant is told to move positive words towards her name, and negative words away from it. In addition, she is told to learn a set of target words, and in each EMA session, she is told to move all of the positive word in one direction (either toward or away from her name). The idea is that response latencies will be greater when participants are told to move positively evaluated target words away from their name, and negatively evaluated words towards their name. After several sessions, the average toward and away scores can be obtained for each target word. The toward scores are then subtracted from the away scores to produce a "valence score," with positively evaluated words having strongly positive valences, and negatively evaluated words having strongly negative valences.
The relationship between implicit attitudes and actual behavior is still somewhat controversial, but implicit attitude tests have been shown to accord with several different types of behavior, including consumer behavior, which is roughly analogous to voting behavior. The idea, then, would be to have undecided voters move the names of candidates and/or political parties towards and away from their names, compute a valence for each candidate, and then use that to predict for whom they will vote. My suspicion is that most undecided voters do have implicit attitudes toward the candidates that are not reflected in their self-reports, and perhaps even in their own beliefs about their attitudes toward the candidates. Furthermore, I suspect that the more positive the implicit attitude toward a candidate is, the more likely an undecided voter is to vote for that candidate, while the more negative the implicit attitude, the less likely a person will be to vote for the candidate.
1 Brendl, C.M., Markman, A.B., & Messner, C. (in press). Indirectly measuring evaluations of several attitude objects in relation to a neutral reference point. Journal of Experimental Social Psychology.
2 Chen, M., & Bargh, J. A. (1999). Consequences of automatic evaluation: Immediate behavioral predispositions to approach or avoid the stimulus. Personality and Social Psychology Bulletin, 25, 215-224.