The graph above gives the percentages of survey respondents in a 2004 Annenberg pre- and post-debates panel (N=1140) who switched towards or away from John Kerry and George W. Bush. About 2% of respondents said before the debates that they would vote for John Kerry but they changed their minds at the end of this period; stating that they would either vote for Bush, not vote at all or were unsure. About 4% of respondents moved in the other direction. By contrast, Bush had more losses than gains.
Now this is just a graph I created over lunch (oh academic life!) but it shows that small aggregate changes can be the consequence of a modestly large number of voters changing their minds. In this sample, about 11% of respondents changed their vote intentions (including switches to and from Nader). That is not very much but it is much more than the percentage of voters who said “they don’t know” who to vote for at the beginning of the panel (about 3% in this sample).
In this example, many individual changes canceled each other out. Could this have been different if all new information had pointed in the same direction? I am not sure about the answer to this question. Some of this switching probably reflects randomness in survey responses. On the other hand, the 11% who changed their minds didn’t appear so different from the others in terms of the factual political knowledge questions Annenberg asked; so I doubt ignorance is the full explanation.
I suspect it is pretty common that people change their minds for all sorts of reasons. Some may be impressed by Romney’s debate performance while others fear for Big Bird. Still, I would caution against analyses that argue that potential movement in the electorate is limited based purely on reports of aggregate vote intentions.