Eurekalert points to up and coming research on how to forecast the votes of the undecided:
Using subjects in Vicenza, Italy, where article co-authors Silvia Galdi and Luciano Arcuri reside, the researchers interviewed 129 residents about the impending enlargement of a U.S. military base in their community. The plans were controversial, and media reports showed strong polarization among residents.
The researchers interviewed each subject twice, one week apart. Each time the participants were first asked if they were ‘pro,’ ‘con’ or ‘undecided’ about the expansion. They then were asked to answer questions about their beliefs on environmental, political, economic and other consequences of the enlargement of the base. Finally, they were given a computer-based latency test of automatic mental associations, in which they were asked to categorize pictures of the base, and positive and negative words as quickly as possible. The full questioning and testing was performed a second time a week later. Automatic associations that undecided participants revealed in the first round significantly predicted their conscious beliefs and preferences as expressed in the second round.
In other words, the researchers could predict future choices of participants who were still undecided in the first session.
Gawronski says, “This kind of testing has many applications, but certainly political polling at election time would be one. It can’t give answers to all questions, but it could certainly help pollsters to get more information than people now share.”