Princeton Election Consortium

A first draft of electoral history. Since 2004

Want to see individual state probabilities?

July 31st, 2008, 1:58pm by Sam Wang


Check the right sidebar for some cool interactive maps. The state-by-state probabilities are in this file; the first number on each line gives the percentage probability that Obama is ahead. The bars indicate electoral votes for which win probabilities are currently greater than 95% for either candidate.

Tags: Site News

9 Comments so far ↓

  • Jane Mohler

    Sam:
    I’m an epidemiologist- thrilled to follow your site stats first thing every morning. I send all my friends to you for the most accurate udates-
    I’ve thanked you with for the senate races, and now I an thanking you in person!
    Jane

  • Sam Wang

    Sebastian and RF, thank you so much for writing. It’s good to hear from you, especially after that rollercoaster ride four years ago.

    I am still playing around with how to manage this site. The subject of polling analysis has boomed considerably compared with four years ago, and so have my other responsibilities! If commenting gets turned off, don’t take it personally – but do feel free to write me.

    - Sam

  • Sam Wang

    Allen, it’s in fact correct – but needs to be better labeled. The bars indicate only states with win probabilities of more than 95% for either candidate. Look at the shading closely – some states are intermediate in their probabilities. We’ll address the labeling issue soon.

    The map is currently interactive. It’s currently configured to assign all states after you click once. At that point the total will be 538.

  • RF

    Need to clarify and revise my last comment — I’m guilty of some very tangled prose.

    The beginning of my last comment seemed to say I want a more precise prediction of the November result. I don’t.

    What I want is to dispel the false sense of certainty readers may get about the final outcome when they look at early polling numbers.

    I want to give readers a reality check.

    For example, if the following “fact” I just made up were true, it would be interesting to know: “in elections since 1972, the leader in August 1 polls has lost twice.” Also interesting would be some characterization of how many percentage points (in whichever direction) polls have typically drifted between August and November. Of course, I don’t have the polling data I’d need to figure any of that out.

    Revising my other comment, I no longer think the right approach is to estimate how likely the current leader is to lose in November — that’s as much art as science. Rather, we need just enough evidence to viscerally persuade readers that early polls truly haven’t been perfect predictors of the Fall result in the past. The goal is to qualitatively change someone’s approach to the data.

    Finally, fanmail: I saw the Google talk you and your book co-author did, and recently when some coworkers were talking about rebutting anti-Obama smears, I cited your NYT editorial. And if you need a hand: for a living, I hack Perl and manage data, have even done some statistical analysis (regression in GNU R, implementing the u-test in Perl), and could prob’ly sign an NDA and implement what you need implemented (and explore your yummy datasets :) ) as a volunteer.

    (More clarifications: from work, I know how relevant something like your power #s can be for targeting; I certainly wasn’t trying to put them down by calling them “cool.” And it is noteworthy that your analysis focuses on low noise; even if you *did* somehow know just how uncertain the Fall result is, you wouldn’t want to lose track of precisely where the electoral vote count would stand if the election were held today.)

  • Sebastian Jano

    Prof. Wang, I’ve been waiting to see if you would return. I am so happy that I will have this site to turn to when I want the strongest empirical evidence available.

    Your site was a delight last time around. I believe your analysis was essentially correct, but there were some voting irregularities that you couldn’t possibly have accounted for in your calculations.

    Welcome Back! and Thank You!

  • Sam Wang

    Cindy, hi. On each line, from left to right, we have
    - current probability of an Obama win in % (McCain would be 100 minus that), based on the last 3 polls
    - current median of the 3 polls
    - probability, if the true margin is 2 points toward Obama (or if opinion shifts by that much)
    - probability, if the true margin is 2 points toward McCain (or if opinion shifts by that much)
    - two-letter state postal abbreviation.

    RF, I think your question is fascinating. But I also think that making a prediction about November is intrinsically uncertain. There is interesting work on this topic by Prof. Abramowitz at Emory, as well as Ray Fair, which you can look up.

    The great value of what I present here is the fact that the EV estimator I give is lower-noise than any other metric you will find anywhere, including at fivethirtyeight.com. As a consequence, you can see in real time (i.e. over days) whether some shift has happened in the race. Conversely, if it doesn’t move – not much happened. If you ever get concerned about an individual poll, come here to see what’s really happening.

    Secondly, the voter valuation in the right sidebar tells you how much your vote is worth – and by extension, where your efforts will have the greatest effect on the race.

  • RF

    Most readers here want to figure out who will win and not just who’s currently ahead. (Though the other info like power indices is awesome, too.)

    This analysis of whether/how much to discount early leads, which I’m sure you’ve seen, gets at my worry:
    http://www.fivethirtyeight.com/2008/06/should-we-be-discounting-obamas-lead.html

    But instead of estimating the *size* of the winner’s final lead based on early polling, I’d like to see the *probability* of an electoral vote win for the candidate with an X% lead in the national polling X days ahead of the election.

    With relatively few elections’ worth of polling data, it takes some creativity (or some dipping into House/Senate data) to come up with a good model. And each election’s different from the last. But not trying to estimate the uncertainty around these numbers has got to be worse — the typical reader will see the histogram at right as representing the chances Obama *will* win in November and not that he’d win if the election were today.

  • Cindy Ware

    Yes, I alerted husband Jim Ware to this site this morning and will send it out soon to many more Obama supporters.

    But first, for the non-techies among us, please explain what each of the three additional numbers for each state represents and how we should interpret changes in these middle figures.

    We look forward to regular updates!

  • Jim Ware

    Thanks for this terrific site. It’s so important for us all to get past the shallow/partisan MSM reporting to understand what is really going on.

    I’m looking forward to following this site carefully in the coming months – and citing you regularly on the other blogs where I post.