Research question/goal:

In the run-up to the federal election in 2005, the Christian sister parties of CDU and CSU started into the campaign with a huge lead in the polls, but then in September came in with their second worst results since 1949. This unexpected drop in public support is exemplary for the recent trend of increasing volatility in party support time-series. At the same time, and seemingly contradictory, a look at the strength of the two major political blocks in Germany shows them to be almost stable over several federal elections. The objective of this project was to account for the high volatility in party preferences between elections on the one hand, and stability of election outcomes on the other.
In the first stage of this project, we conducted data analysis using existing public opinion time-series. Our findings were in support of an existing theory, according to which undecided voters play a major part in inducing volatility. Undecided voters, from this perspective, are mostly disappointed government supporters who routinely return to "their" parties when Election Day draws near. It follows that volatility is partially induced artificially, since the predictable pro-government vote of undecided voters is left unaccounted for in midterm surveys.
In the second stage, we conducted an experimental study designed to look for further evidence of this theory. In the context of the federal election in 2009, we tested whether the future vote of undecided voters could be predicted by unobtrusive attitude measures (the Implicit Association Test). We found conditional support for this hypothesis: Undecided voters with high political interest had strong attitudes that predisposed them toward a political camp, while less interested undecideds turned out to be unpredictable. Overall, our findings should be useful for improving current forecasting practice by helping to eliminate artificial volatility.