Thursday, October 22, 2009


The obvious answer based on recent polling is that if the election were held today, it would be a tie. At least that’s what it appears to be, when all polls are now showing the leader (whether Chris Christie or Jon Corzine) within the margin of error. Even our new Rutgers-Eagleton Poll, while showing one of the largest leads for Corzine (+3), is still within its +/- 4.1 percent margin of error. (see the poll press release at

But the real answer is that it depends greatly on the assumptions being made. Most importantly it depends on the assumptions made about who will turn out. All horserace polls have to model “likely voters” and this is hard.

As the New York Times reminds us recently, turnout in the last two NJ Governor elections has been about 49% of registered voters. Random digit dialing (RDD) as most public polls do reaches a lot of people who are just not going to vote. But it is really hard to figure out who they are. Consider, for example, that those who are actually willing to stay on the line and talk to pollsters are already different from other people. They are more interested in politics and thus more likely to show up on Election Day.

But how much more interested are they? Instead of 49% of registered voters, should we find that 60% of those polled will vote? 70%? 55%? Who knows?

Well internal campaign polls often have the edge on this because they use voter lists to do their polling. Thos lists come with vote history information – they know if you have voted in past elections and in what kind of elections. Are you a consistent primary voter – you’ll almost certainly be a general election voter even in a gubernatorial year. Do you show up only when there is a presidential election – then they won’t count on you for 2009, even if you claim that you are likely to vote when asked. Simply put, one of the best predictors of voting this year is having done it in the past, in elections large and small.

Since RDD pollsters – including Rutgers-Eagleton – don’t have your voting history at hand, they have to ask questions. We base our likely voter modeling on several things. First, you have to be a registered voter. Second, we do ask if you expect to turn out. But people lie about this – a lot. So we ask how likely you are to turn out and generally only count those who claim to be very likely or almost certain to vote. We also ask if you voted last year. If you didn’t vote in 2008, you almost certainly won’t vote this year – unless you were too young in 2008. But this still isn’t enough to convince us you are a likely voter. So we ask some questions about interest in and attention to the campaign. Those who show significant interest and are paying some attention – in addition to registered voter, self identified likely turnout and 2008 voting – become our likely voters. Others may approach this process differently and have different screens. Your mileage may vary – and so may the reported horserace!

One interesting result of differing screens for likely voters is that various New Jersey polls – such as Rutgers-Eagleton, Monmouth, and Fairleigh Dickinson often end up with slightly different mixes of partisans in their likely voter samples. Monmouth recently reported the race tied at 39% for Christie, 39% for Corzine and 14% for Chris Daggett. Their weighted sample shows turnout at 40% Democratic, 26% Republican, and 34% Independent. Democrats are much more likely to vote for Corzine, Republicans for Christie, and Independents in the poll are split, but favoring Christie. Obviously the more Democrats who are assumed to be showing up the better Corzine will poll all else equal; the more Republicans, the better for Christie.

We reported Corzine 39%, Christie 36%, Daggett 20%. Our breakdown by party is 39% Democrat, 35% Independent, and 26% Republican, very close to Monmouth. But what if we reported 40% of the voters would be Democrats and 34% would be Independents, leaving Republicans at 26%? This one point different in turnout assumptions would increase Corzine’s lead from 39%-36% to 41%-36% (rounding), putting Corzine up 5 points and outside the margin of error.

But that’s NOT what we reported because our data suggests the 39-26-35 turnout breakdown. I just want to point out that even small changes in assumptions lead to interesting changes in interpretations. For what it is worth, the actual voter registration as of October 5, 2009 in New Jersey is 34% Democrat, 20% Republican and 46% Independent. Notice we all assume Independents are much less likely than partisans to show up in 2009.

So while it is trite to say it, this election IS all about turnout.

And who will win? Well, remember polls ask “if the election were today” so they are not predictive of the future. And who they say is winning heavily depends on turnout assumptions. 40% Democrat turnout share and Corzine may win it. 35% Democratic share and he loses it. At least if the election was today!

No comments:

Post a Comment