In my earlier post, I modeled the choices of decided voters and then looked to see what that model would predict about the remaining undecided voters. The apparent answer was that these voters will break pretty evenly for Obama and Romney.
My co-author Lynn Vavreck suggested a different approach. In our data for The Gamble, we have interviews from a December 2011 YouGov survey with approximately 44,000 respondents. Then each week from January until Election Day, YouGov reinterviewed a different subset of 1,000 respondents from among these 44,000. In other words, we have a two-wave panel, where the date of the second wave varies.
Lynn’s suggested approach was: take the people who were undecided in December 2011 but decided in a later survey, model their choices, and then see what this predicts for the remaining undecided voters (as in my first post, these are the 195 undecided respondents in the 3 YouGov polls since the first debate). The assumption here is that the remaining undecideds will make up their minds the way the earlier undecideds have — which is more plausible than assuming, as I was, that undecideds make up their minds like the decideds do. (Of course, it’s still an assumption.)
In this case, the answer is much the same: the remaining undecided voters are predicted to split fairly evenly — about 55% for Obama and 45% for Romney. Given the uncertainty in these estimates, I don’t think this implies a definitive Obama “win” among these voters. Ultimately, these models do not suggest either candidate will get a significant advantage from late-breaking undecided voters. The same caveats noted in my earlier post apply here as well.