Princeton Election Consortium

A first draft of electoral history. Since 2004

Assessing the effect of the bounce [currently Obama 324, McCain 214]

September 4th, 2008, 12:19pm by Sam Wang


We are at the end of several very eventful weeks. The finale is, of course, tonight: McCain’s acceptance speech at the Republican convention (and a pre-rebuttal by Obama on The O’Reilly Factor). National polls will come in over the weekend. Here’s a tool to help you interpret them.

It will be at least a week before state polls will show us where the post-convention race stands. In fact, it’s been a polling desert for at least a week, as you can see from the flat part of the history graph. Until then, the best indicator will be national polls. How to convert them to an EV estimate? The answer is in the following graph.

EV adjustment curve with pre-convention polls except AK

EV adjustment curve with pre-convention polls except AK

(Click the figure to get a higher-resolution version for reading off values)

The graph was generated starting from the most recent pre-convention state polls, which are basically a pre-Democratic convention snapshot. Because some of them are pre-8/21, they do not even fully reflect the post-”Seven” bounce. The exception is Alaska, which is now assumed to be McCain +20. At that time, in national polls Obama led McCain by 3%. Then a range of offsets was added, and the Meta-Analysis was redone. Use the current national polling margin, which you can get from Pollster.com or RealClearPolitics, to get a corrected median EV estimator. As of the evening of 9/4, national polls taken immediately before the Palin speech give a median of Obama +5%, corresponding to Obama 324 EV, McCain 214 EV.

The gray band indicates the 95% confidence band, assuming an uncertainty of +/-1% in national margins, close to what I calculated before.

This graph can also be used to calculate the effect of systematic polling error. An example of systematic error that favors McCain is the possibility that polling margins overstate the relative support for Obama because of racial bias (the Bradley/Wilder effect). An example that favors Obama is the fact that people who have a cell phone but no landline, and are therefore missed in conventional surveys, skew strongly toward Obama. You can add in your own assumptions and read the result. I’ll provide an updated version of this graph as the election approaches.

Tags: 2006 Elections · Uncategorized

2 Comments so far ↓

  • Sam Wang

    Aaron, those three polls are not included in the graph generated for this post. Including them only changes Minnesota’s win probability. Obama already leads McCain by a substantial amount in Iowa and the median in Ohio would not change. Their total effect on the median estimate is 1 EV in Obama’s favor (it just updated in the banner).

  • Aaron

    Prof Wang-

    Did the new polls from CNN in IA, MN, and OH change anything?