A few people have asked that we include the crosstabs for likelihood of voting in the January Marquette Law School Poll. Those are now included in the .zip file for all crosstabs that can be downloaded from the “Results and Data” page here. The results for likelihood of voting are in the file q2.txt. There are 573 respondents who say they are “absolutely certain” to vote in November and an additional 78 who say they are “very likely” to vote. The margin of error for the 573 absolutely certain group is +/-4.2 percentage points. Combining the two groups gives 651 respondents with a margin of error of +/-3.9 percentage points.
Different pollsters use different criteria to define likely voters, including using additional variables such as attention to the campaign so this single question is not the only way likely voters can be defined. Be especially aware that we did not ask how certain the respondent was to vote in a recall election so those results might differ. See the end of this post for some links to further issues in analyzing likely voter samples.
The results for the governor’s race vary by one or two points, well within the margins of error. If we compare the full registered voter sample with the “absolutely certain to vote” and with the “absolutely + very likely to vote” group, here are the results:
For the presidential race the “absolutely certain” group and the “very likely” group are quite different and that produces a much bigger difference depending on how likely voters are grouped:
Obama support drops and Romney increases among those absolutely certain to vote, transforming an eight point Obama lead into a tie among those certain they will vote. But add in those very likely to vote and the results move back to a six point Obama margin. Interestingly, this is more movement than we see in the governor’s race, even though we are comparing the same set of people. Motivation to vote plays only a small role in the trial heat for governor but motivation seems to matter more in the presidential race. One reason may be that voters in Wisconsin have focused a lot of attention on the potential recall election but have not been similarly focused on a presidential race that is still 10 months away.
While likely voter samples are commonly used by pollsters close to election day, there has been considerable doubt raised as to their value when elections are still distant. Much of the change in candidate support appears to be from changes in how sure voters are to turn out rather than actual shifting of preferences. And voters shift their enthusiasm for turning out quite a bit with the ebb and flow of the campaign. This has caused some analysts to conclude that registered voter samples are actually more reliable through most of the campaign, with a focus on likely voters more meaningful only close to election day when potential voters are more sure of whether they will vote or not. One of the best analyses of this issue is by Erikson and colleagues who conclude
When polling on the eve of an election, estimating which respondents are likely to vote is an essential aspect of the art. This article has pointed to dangers of relying on samples of likely voters when polling well before Election Day. Our evidence suggests that shifts in voter classification as likely or unlikely account for more observed change in the preferences of likely voters than do actual changes in voters’ candidate preferences. (Erikson, Panagopoulos and Wlezien, Public Opinion Quarterly, 2004. Emphasis added.)
Mark Blumenthal of Pollster.com also has a nice discussion of this issue at this link.