There Go the Undecided Voters

by Larry Bartels on September 21, 2012 · 8 comments

in Campaigns and elections,Data

Lynn Vavreck has an informative piece on the New York Times Campaign Stops blog today tracing shifts in presidential voting intentions from late 2011 through early September. The data are from the Cooperative Campaign Analysis Project, which interviewed nearly 44,000 people last December and has subsequently been reinterviewing 1,000 per week. (Top monkey John Sides is a collaborator in the CCAP study, and I received access to some of these data for an earlier Campaign Stops post that Vavreck and I wrote together.)

Through most of the spring and early summer, more than half of the survey respondents who were undecided last December were still declining to choose a candidate, with the rest breaking slightly for Mitt Romney over Barack Obama. Since around mid-June, more of these previously undecided voters have begun to commit, with Obama gaining and, in the last few weeks, surpassing Romney among those who were originally undecided. According to Vavreck, “These decisions seem largely to have been motivated by party identification.”

Meanwhile, both candidates have managed to retain the vast majority of prospective voters who supported them last December. Over the course of 2012, Obama has held 96% of those who supported him in 2011 and added 3% of those who originally said they would vote Republican. For his part, Romney has held 94% of those who intended to vote Republican and added 2% of those who intended to vote for Obama. (Vavreck notes that the 2008 CCAP study found almost as much stability in candidate preferences, with Obama holding 90% of his early supporters and John McCain holding 92% of his.)

To readers versed in election studies, these findings will seem very reminiscent of those from the first scholarly analysis of campaign effects: “conversion is, by far, the least frequent result and activation the second most frequent manifest effect of the campaign.” However, whereas Lazarsfeld and his colleagues in 1940 studied 600 prospective voters in Erie County, Ohio, Vavreck and her colleagues in 2012 have 44,000 nationwide. That’s real scientific progress.

{ 8 comments }

Scott Monje September 21, 2012 at 11:07 am

Is it the case that they make up their minds only once? Once people have switched from undecided to decided in favor of either Obama or Romney, then they probably won’t change their minds again?

Larry Bartels September 21, 2012 at 12:56 pm

Vavreck only observes them once between the baseline survey and Election Day, so her data do not speak directly to this question. (Her post-election followup will.) I would guess that current vote intentions would be considerably weaker among people who were undecided in December than among those who expressed a preference at that point and have stuck with it. More generally, it seems sensible to think of the intercepts in her graph as reflecting measurement error and random movement in and out of the “undecided” category–even in the first reinterviews shortly after the baseline survey, about 40% of her original “undecided” voters indicated a Democratic or Republican preference. The slopes in the graph indicate net movement of voters out of the “undecided” category over the course of the election year.

Larry Bartels September 21, 2012 at 12:58 pm

I should have added that _The People’s Choice_ was published four years after the 1940 election. Vavreck is working in close-to-real time, and her analysis is being disseminated by the _New York Times_. That’s real progress, too.

Alan T September 21, 2012 at 1:39 pm

What is the most frequent manifest effect of the campaign? I must be overlooking something so obvious that you do not feel the need to state it.

Also, some of the details in Lynn Vavreck’s article confuse me. 94% of 44,000 people had made up their minds, and 1,543 were initially undecided. 1,543 is not 6% of 44,000. Furthermore, “half of the 1,543 initially undecided voters … are still unsure,” yet ” the share of still uncommitted voters (from that initial group) had dwindled to 25 percent.”

Larry Bartels September 21, 2012 at 3:17 pm

Poor quotation on my part. The most common effect of campaigns, both in _The People’s Choice_ and in subsequent scholarship, is to reinforce the predispositions (largely, partisan loyalties) prospective voters bring to the election season.

Vavreck has boiled down a lot of data and analysis into 1500 words, which unavoidably introduces some ambiguities. I believe 1,543 and 94% both exclude people who said they were undecided but then said in response to a followup question that they leaned toward one party or the other (and in both cases the relevant denominator is not 44,000, because some people said they wouldn’t vote). And “half . . . still unsure” refers to the average over the entire year, while “dwindled to 25 percent” refers to the fraction still uncommitted in early September.

LV September 21, 2012 at 3:48 pm

Thanks for all of the interest in our work! On the N, let me add a little bit more intuition: We will end up with 44,000 impaneled cases, but right now, 37 weeks in to the year we only have 37,000 re-interviews. We interview 1,000 of the initial 44,000 each week. OK, second thing, lots of people don’t vote. So, questions about decisions in voting booth eliminate those people who say they definitely will not vote on Election Day. That reduces the 37,000 by roughly 20%. I also set aside people who reported they would vote for someone other than one of the two major party candidates (roughly 4%). So … that’s going to get you down to a total N where the percentages add up!

Alan T September 21, 2012 at 9:47 pm

Thanks!

Tommy Bennett September 21, 2012 at 5:35 pm

Is that chart plotting a LOESS curve? If so, how much faith can you put in the line’s accuracy over its most recent weeks?

Comments on this entry are closed.

Previous post:

Next post: