>Obama has come and gone. The final polls are out. And the campaigns are focused on GOTV. Getting out the vote is what it is all about. And while there are reports that about 175,000 New Jerseyans requested mail-in ballots (and about 100,00 or so had been returned as of the end of last week), the vast majority of the 2.4 million or so voters (out of 5.2 million voters) will cast their votes tomorrow in voting booths around the state.
Who’s winning? Well it ALL depends on our assumptions about turnout.
Let me use the last two Quinnipiac Polls to illustrate this. They are very helpful since they are one week apart, but report very different results!
The latest Quinnipaic Poll (released today) suggests Corzine has dropped back. The last QPoll (10/28) had CORZINE up 5 points. This one has CHRISTIE up 2 points, 42%-40% with Daggett at 12%. The margin of error is +/-2.5%.
The key point I see in this poll is that Corzine and Christie remain at parity among their party voters. Corzine has 77% of Democrats and Christie has 78 percent of Republicans pretty much where they were a week ago. So without any real change in partisan support for the candidates, there has been a 7 point swing in the reported poll results. Where is it? It’s not in the independents, who are now 47%-32% Christie, compared to 45%-30% last week – making no swing at all here. And the Don’t Knows are at 6% this week, compared to 5% last week, so it can’t be them either. So what the heck is going on?
Let’s take a closer look at the numbers.
Given that there are MANY more Democrats in NJ than Republicans, this should benefit Corzine. For every 100 registered Democrats there are 60 Republicans.
So if we only cared about Democrats and Republicans, and if BOTH parties vote in the same proportion as they are registered, Corzine gets 77 votes from Democrats and Christie gets 47 votes from Republicans (from each 100 Dems and 60 Reps). That’s quite a margin.
Of course some Democrats plan to vote for Christie (Qpoll says 6% of Democrats will do this.) And some Republicans will be for Corzine (10% according to the poll).
So we have to add these folks to the totals: 6% of Dems X 100 Dems = 6 more for Christie (now at 53) and 10% of Reps X 60 Reps = 6 more for Corzine (now at 83).
Note I am ignoring Daggett. This is not a political statement in the least. I believe voters should vote sincerely, that is, for the candidate they support, even if that candidate is unlikely to win. But for this calculation he is not relevant.
But of course all the voters are NOT Democrats and Republicans. So what about the Independents? Turns out there are 136 registered independents for each 100 Democrats. And Qpoll says that they go strongly for Christie, 47% – 32%.
So let’s add them in. 47% of 136 Independents is 64 more votes for him, bringing Christie to 117 votes. Corzine gets 32% of the 136 I’s, giving him another 44 votes, bringing Corzine to 127 votes.
Wait a minute. When I do the calculation based on how many registered voters there are, Corzine is winning, 127 to 117, which translates to 42.9% to 39.5%, a 3.4% margin in favor of Corzine (again ignoring Daggett voters and Don’t Know voters, who make up the remaining 17.6%.) But Quinnipiac reported a 42% – 40% race in favor of Christie.
Where do the numbers come from?
So is Quinnipiac cheating? Of course not. In fact they do a great job with their polling. The problem in this example is my own basic assumption: Democrats, Republicans AND Independents will turn out tomorrow in the same proportions they exist as registered voters – that is, 46% Independent, 34% Democrats, and 20% Republicans.
But this will not happen. We know for certain that voters who are registered independents are MUCH less likely to vote in ANY election, especially a non-presidential one. And Republicans (as the out party) MIGHT be more motivated to vote than Democrats, who do not appear to be that enthusiastic about Corzine, even if they support him.
So the polling results you see reported are based on some some expectation about who will turn out. And those assumptions are embedded in the screening questions public polls use to determine “likely voters”. The screening questions are different for each pollster, and who becomes a “likely voter” may be different as well.
A quick back of the envelope calculation suggests that in the Quinnipiac Poll I just used for the example, Republicans must be significantly more likely to pass the “likely voter” screening questions than Democrats or Independents. The sample is probably 28-30% Republican (versus 20% in the registered voter population) and 35-38% Independent (much lower than their 46% share of registered voters.) Democrats are probably somewhere near their actual ratio of registered voters (33-34%). I don’t know for sure because I can’t find it reported in the QPoll.
But Corzine was up 5 points in the last Quinnipiac Poll a week ago!
But why was the Quinnipiac poll only a week ago showing Corzine up 5 points? Did he lose that much ground? Let’s see, that poll looked like this:
Corzine — Dems: 79%; Reps: 7%; Inds: 30%;
Christie – Dems: 8%; Reps: 79%; Inds: 45%;
It’s obvious these numbers are not much different from the new poll. Doing the same calculation I did earlier with the same (wrong) turnout assumption I started with, that’s:
100 D’s – 79 Corzine, 7 Christie
60 R’s – 4 Corzine, 48 Christie
136 I’s – 41 Corzine, 61 Christie
Total – 124 Corzine, 116 Christie or 41.9% Corzine, 39.2% Christie.
This is nearly the same as this week! In fact if anything, Corzine is doing BETTER this week (42.9 versus 41.9 last week).
It seems obvious now that it is all about turnout assumptions, doesn’t it?
It would be very helpful to all of us if ALL public polls reported their partisan breakdown of likely voters. We did that in our pre-election poll, and Monmouth routinely does it as well. Their last poll shows 40% Democrat, 34% Independent, and 26% Republican. They have Christie up 43%-42%.
In our Rutgers-Eagleton Poll (fielded 10/15-20) we came up with 39% Democrat, 35% Independent, and 26 Republicans. And we had Corzine up 43%-40%. Where did we get these numbers? It’s what came through our likely voter screen.
If we applied the new Monmouth partisan shares to the new QPoll numbers, we would get:
Corzine: .77 X .39 + .32 X .35 + .26 X .10 = 43.8%
Christie: .06 X .39 + .47 X .35 + .78 X .26 = 39.1%
So using Monmouth’s likely voter partisan shares and Quinnipiac’s partisan results, we get a 4.7 point lead for Corzine, almost exactly what they reported on October 28. (Note, Monmouth says Christie wins 81% of Republicans, which is why he is leading by 1 point in their poll.)
It MUST be the case that the Quinnipiac Poll of October 28 had a larger share of Democrats in it than the poll of November 2 does. And the November 2 poll MUST have fewer Democrats than Monmouth Or Rutgers-Eagleton) reports. So either it is the random effects of sampling populations, OR Democrats have become less likely to say they will turn out this week than they were last week. And of course this week is the week that matters!
So who cares?
In a close race, it is essentially useless to use any of these polls to say one candidate or the other is winning. It is a statistical tie, and the media should make that clear. Basic assumptions, likely voter screens, and the number of partisans in the samples all make a difference when the race is this close.
The campaigns all know the numbers too, of course, but they do a lot LESS guessing about turnout. At a minimum they run their polling using voters registration and history lists. They KNOW who has voted in the past over the years, and thus who is more likely to turn out and who will stay home. And while it is not an exact science for them either, if we could see the internal campaign polls and compare them to the public polls, we’d have a much better sense of what’s really going on.
Or we can just wait until tomorrow night!