Princeton Election Consortium

Current national margin and corrected EV estimator

September 5th, 2008, 9:11pm by Sam Wang

We aren’t going to know where things stand until weekend polling comes out on Monday. In the meantime, a brief snapshot of where things may be headed. Based on six national polls spanning September 1-4, on those dates Obama led McCain by 3.5 +/- 0.9 %

This is nearly identical to where things stood before the Democratic convetion. The polls are the most recent available from Gallup Tracking, Diageo/Hotline, Democracy Corps (D), CBS, Economist/YouTube, and Rasmussen Tracking, and draw upon 9390 respondents. If one-fourth of the respondents on 9/3 were reached after the Palin speech that evening, the sample contains 68% pre-Palin, 32% post-Palin respondents.

Based on the correction to pre-convention state polls I provided yesterday, applying the national margin as a correction to the Meta-Analysis gives a median EV estimate of Obama 311 EV, McCain 227 EV. The 95% confidence interval is Obama [278-342] EV, McCain [196-260] EV.

Sources: Pollster.com, RealClearPolitics.

Tags: 2008 Election

6 Comments so far ↓

• Greg P.

Dr Wang,

I see you provide the 95% CI for EVs. I wonder if you could also start providing the one-sided level of confidence corresponding to 270 EVs. Thus Obama’s confidence level would apparently be something like 96% and McCain 4%. It would be helpful in providing the a simple probability of who will win the election.

• Sorry, I won’t be doing that. To a non-technical person, an EV estimate with a confidence interval gives the right impression, more or less. A one-tailed probability does not. The snapshot does not give you a probability of an actual election win. It only tells where things stand today. Therefore I will be leaving the information in its current form.

With a bit of number crunching you can get what you want by looking at the histogram data. Simply paste it into a spreadsheet and sum rows 270 through 538. Also see the geek resource, which offers a lot (everything, actually).

Prediction requires modeling future movements, which are highly uncertain. If you really want this, go to FiveThirtyEight.com.

• Ragout

This is a system of N equations for N+2 unknowns, which cannot be solved algebraically. Therefore we need more information or assumptions.

Fortunately, Gallup provides more information, in the form of weekly averages.

With two weeks of data, we have N equations and N unknowns, which can be easily solved, except for rounding error. With three weeks of data, we have N+1 equations and N unknowns, which is overdetermined and requires a regression approach or something similar. With enough weeks of data, the rounding error problem should go away.

Here’s my solution to the last few weeks. Notice that if you take 3-day averages of my figures, you get Gallup’s numbers exactly.

Unreported Daily Polls
Obama McCain
9/6 42 48
9/5 45 51
9/4 48 45
9/3 48 39
9/2 48 48
9/1 51 39
8/31 48 42
8/30 51 45
8/29 48 42
8/28 45 39
8/27 54 42
8/26 48 42
8/25 42 42
8/24 45 48

• Vincent Ferrera

Sam,

1. Awesome site!

2. I’ve been wondering if/how the polls weigh expectations about voter turnout. For example, the Nonprofit Voter Engagement Network (www.nonprofitvote.org) has published the primary voter turnouts for the 2000, 2004, and 2008 primaries (America Goes to the Polls). The numbers show a huge increase in overall turnout, and especially among democrats. How is that factored into the polling data?

2. The same report shows a marked shift in voter demographics, with large increases in young (18-29) and Latino voters. Which, if any, polls are addressing the possibility of sampling bias, for example the possibility of systematically undersampling young voters?

• Ragout, I didn’t know about those averages. With two weeks’ worth of averages, in principle one has a system that allows simple matrix inversion to back out an answer.

However, you have neglected the rounding problem. Gallup reports only whole-number percentages. There isn’t enough precision to do it the way you describe. For instance, your percentages are all multiples of 3, clearly wrong. So I think your method is not suitable.

• Vincent, thank you.

In regard to high primary turnout, to be frank I think it means little. Primary turnout is usually low. This year the Democratic nomination race was interesting, either because of the quality of the candidates or because Democrats sensed the possibility that the White House was well within reach this time. So there is no guarantee of a high fall turnout on the Democratic side. Also, pollsters have methods for measuring voter enthusiasm. So this factor should already be taken into account in polls.

It’s hard to gauge the extent to which young voters are undersampled. Prof. Charles Franklin has an excellent analysis of this question here. He thinks that the advantage, if any, won’t be that large.

One group worth thinking about is people who have cell phones but no landlines. This population is not captured in most phone surveys. The Pew Center has analyzed the phenomenon in an excellent report. The upshot seems to be that Obama supporters are systematically underestimated, perhaps by several percentage points. But again, we don’t know if pollsters have found tricks to compensate for the deficit.