Princeton Election Consortium

Innovations in democracy since 2004

Outcome: Biden 306 EV (D+1.2% from toss-up), Senate 50 D (D+1.0%)
Nov 3 polls: Biden 342 EV (D+5.3%), Senate 50-55 D (D+3.9%), House control D+4.6%
Moneyball states: President AZ NE-2 NV, Senate MT ME AK, Legislatures KS TX NC

Daily News – hold the fizz?

November 3rd, 2008, 8:04am by Sam Wang

The Daily News asked me for a piece written in their style. So… “[The Bradley effect] could wind up stealing the fizz from Obama’s Election Night champagne.”

Hey, it was either that or write about topless muggers, or whatever else makes the front page on normal days. By the way, the counterpoint piece is good too.

Also accompanying the article is a revealing photograph of another possible cause of Tom Bradley’s narrow loss…

A somewhat frightening tie.

Tags: 2008 Election

42 Comments so far ↓

  • Michael

    It’s really terribly unfair of you to judge politicians of the ’80s by modern sartorial standards, Sam.

  • Sam Wang

    Exactly my point. Times change. Bradley effect’s gone – and teeny-tiny patterns are out. Nothing personal – I have ties like that in my closet.

    By the way, that photo’s from 1992, messing up the joke a bit.

    We now return to our regularly scheduled, logic-based analysis.

  • Fred

    I can’t find the methodology for chosing state polls. Is it the polls of the last week? The last three polls?
    AS per your prior posting, I’d like a cheat sheet–your thoughts on what to watch for.
    Thansk for a great site!

  • Matthew Ward

    Awwww … that gets us to November 5th. Then what do we do? I say push the tie analysis and make the transition to a full fledged fashion site. It is that or the statistical modeling of Top Chef.

  • Sam Wang

    Fred – All methods are here. We use last 3 polls plus last 7 days of polling. Everything but the squeal, where squeal=Zogby Interactive.

  • WG

    Hi, why is the meta-margin dropping? Can you give an explanation? Are the polls wrong?

  • dave kliman

    Did you at least discuss potential reverse bradley effect?

  • MB

    The meta-margin has plummeted in the last three days or so. Wasn’t in the mid 8’s last Friday? Seems to be a great deal of volatility recently.

  • Richard Vance

    Sam, please after the election we need meta-analysis of leading economic indicators.

  • Sam Wang

    WG and MB – Previously I think the volatility arose from increased frequency of polling. But in the last few days, undecideds begin to be able to report their preference, and they do…

    The latest move appears to be driven by a genuine narrowing over the last few days. Go to and look at FL / MO / OH / PA / VA. Turn the smoothing tool to “more sensitive.” You’ll see upticks for McCain, some of which are sharp.

    I’ll write more after the morning’s polls are all in.

  • Mary

    Sam, I lived in LA in 1982; I was an LA City and County reporter. I had every intention of voting for Bradley for governor. At the time of the election I was confined to bed because of pregnancy-related medical problems. Still, I phoned my doctor and told him I wanted to vote; he said fine, just as long as somebody took me to the polls, which I did arrange. However, that night, in our mountainous area of the city, there was a horrendous thunderstorm. I never made it to the polls, thinking that since Bradley was so far ahead in the polls my vote wouldn’t make any difference. The weather that night is never mentioned, but it was the primary factor in my decision not to go out.

  • Fred

    Merci beaucoup.

  • Michael

    It is that or the statistical modeling of Top Chef.

    This one gets my teen-aged daughter’s vote, although she’d probably prefer America’s Next Top Model.

  • icy

    Thanks for your site.
    Last election the Meta analysis over estimated and gave the blue ahead, the results were tight. Maybe the percentage of optimistic assumptions should be accounted for.

    Who ever gets elected will carry blames, crashes and slides, consequences of the last 7 years decisions…
    Good luck America!

  • WG

    Thanks for the information.

  • EL

    I’m basing my vote on whose ties I like better. It”s called the Brooks Bros. effect. Stick that in your Meta whatever.

  • Snowball


    The method was sound and predicted the exact result. However, Sam made one wrong assumption which won’t be repeated this year. Check out the FAQ:

    “Why should I believe the Meta-Analysis? In 2004, didn’t it predict a narrow Kerry victory?

    Actually, the method was fine, but its inventor, Prof. Sam Wang, made an error. In the closing weeks of the campaign, he assumed that undecided voters would vote against the incumbent, a tendency that had been noticed in previous pre-election polls. Compensating for the “incumbent rule” had the effect of putting a thumb on the scales, lightly – but unmistakably – biasing the outcome.

    Leaving out this assumption, the prediction in 2004 was exactly correct: Bush 286 EV, Kerry 252 EV. “

  • Johnathan F. Taco III, Esq.

    Can we get an analysis of ties’ impacts on voter turnout and their decision-making process? :)

  • egc52556

    When I see Bradley’s tie I’m reminded of the old gospel, “Blessed be the ties that blind.”

  • David Palmer

    Well, I was going to comment on Icy’s question, but Snowball beat me to it.

    I will add, however, that I have a great deal of faith in the meta-analysis given that it accurately predicted the results (including the Ohio/MI/MN/PA bunch) last cycle. Impressive.

  • ICDogg

    I think it is a mistake to accept that the Bradley effect ever existed in the first place.

    I would submit that it is just as likely that pollsters invented the Bradley effect as a cover story for their incompetence at creating a ridiculous likely voter model for that race.

  • Sam Wang

    Icy, Snowball, David Palmer – This year my assumption is that polls do not need to be corrected. Under that assumption in 2004, the poll averages got correct every state except Wisconsin, which was within 1 percentage point. The validation is here. In addition, the Median EV Estimator was exactly equal to the final outcome.

    ICDogg – see my posting on this site, and an analysis by Dan Hopkins of many races.

    Everyone else – maybe read the article before commenting?

  • Rachel Findley

    One of the Daily News commenters offered anecdotal evidence of a reverse Bradley effect:
    “I know many older white women in the south (Mom’s still there). Her and many of her close church WOMEN friends are talking quietly among themselves that they are voting for Obama, but Lord knows, don’t tell their husbands. But for some reason they have started talking to each other in their kitchen groups. This is happening in a very solid red state. ”

    Of course, we don’t know what those women told the pollsters.

    It is a time of historic transition. It’s great to have a steady statistical analysis to go along with all the stories swirling around.

  • Mr. Earl

    It’s a tie that b(l)inds, isn’t it?

    As for me, Paris Hilton said it all.

    “All my BFFs were Jerry Garcia ties, or they’re not my best friends.” (okay she didn’t really say that, but one day maybe she will)

  • piktor

    A majority of white males have not voted for a Democratic candidate since 1964. Clinton in ’92 and ’96 had less than the 44% white male support Sen. Obama has today.

    The Bradley effect has a lot to do with white males favoring GOP candidates. At the time Willie Brown was Speaker of the California Assembly, also an African American Democrat.

    So, there were strong reasons, political or otherwise (Bradley was BORING)for his defeat. Willie Brown was Speaker until 1995.

  • Paul

    The Bradley effect is so last month’s worry. I want to know how alien mind rays might sway the race!

    Seriously, though, I’m in utter (and inconsequential) suspense about ND, MO and IN. My gut says McCain takes all three. But what will the meta-analysis say? Edge of my seat here!

  • Snowball

    “Everything but the squeal, where squeal=Zogby Interactive.”

    Sam, what did you mean by this (excuse my ignorance on the pollsters)?

  • luis

    Snowball, he means that he uses every major available poll except Zogby Interactive (presumably because they poll online and that methodology is extremely suspect.)

  • luis

    (Or to be more specific, that ‘last three plus seven days’ is drawn from a pool that includes every major available poll except ZI.)

  • gprimos1

    Dr Wang,

    I just saw the history of the meta-margin chart you added. Why does the latest data point show a value of around +8% when today’s meta-margin is 6.68%?

  • San Fran Sam


    Bradley had to win a statewide race.

    Brown only had to win in San Francisco.

    ‘Nuff said.

  • piktor

    San Fran Sam:

    Brown was Speaker of the California State Assembly 1980-1995. He had been Speaker for two years when Bradley ran for governor. I am saying race did play a role in 1982 but Willie Brown and Bradley’s lack of TV charisma also were factors as several other factors, Willie Brown and his nation-wide celebrity and outsize charisma playing into the mix. Some rabid leftist Democrats at the time called Bradley an Uncle Tom.

  • DFS

    OK, what are the chances a 6.54 meta-margin gets reversed in 24 hours?

  • Paul

    @gprimos1: I’ll let Sam correct me, but here are three likely explanations:

    (1) Random noise. (Note that there is a lot in that graph sam posted, and -1.32% is certainly within the range of other apparently meaningless fluctuations we see on that graph.) This is the most likely.

    (2) A genuine contraction of the race in several battlegrounds, perhaps reflecting McCain’s push in PA, OH and recent slight movement back away from Obama in ND, MO, IN.

    (3) A general contraction of the race in the final day, which Nate Silver’s Magic 8 Ball thinks we should expect. This is belied by the national polls for the last two days, which are generally same as ever, or have Obama slightly up.

  • Walter

    @ Paul:

    It’s of course too late to do this for this election cycle, but it should be possible to do error propagation from the polls through to the Meta-Margin in order to tell whether the fluctuations are statistically significant.

  • John Lapeyre

    The Wilder paper’s evidence that there was a Bradley effect is more suggestive than convincing. There were only about 18 races in the data set with about 2.6 polls per race. The exact number of races showing the effect on average is not clear, but figure 4 shows only 10 polls in the period that the Bradley effect was supposedly strong. (There is a lot of other data in the paper, but it is not directly relevant) The mean poll size was 700. The mean discrepancy in outcome compared to the polls was 1%.

  • abe fisher

    On volatility…

    I’d love to see some kind of statistical measure of volatility of the median EV and the meta-margin numbers. If nothing else, it might give us a way of estimating how likely it is that they are moving at high enough (i.e. fickle enough) to change the outcome.

  • Bob Yoder

    So, what the EV History graph says is that there is a 95% chance that Obama will get between 330 and 380 EV?

  • Sam Wang

    gprimos, Paul, Walter – The Meta-Margin graph is not re-calculated dynamically. We’ll update it later, but in the meantime you’ll have to imagine that last point.

    Walter – you will get your wish of some kind of error propagation. I call it a bias analysis. It’s buried in the mists of the 2004 analysis. You can see a more recent example when I wrote about Obama’s red ceiling:

    This graph was plotted assuming that the margins were to swing in all states by the same percentage. The EV estimator would then change like this:

    bias plot for 3 October 2008 3:00PM

    Needless to say this will be recalculated later.

    Everyone else, including Bob Yoder, please read the FAQ!

  • Walter

    Ah, that’s exactly the sort of thing I was looking for — or rather you can extract it from there, by looking at the x-intercepts of the error region.

    From there it looks like the meta-margin is -3.7 +/- 1.6 (using the 95% CI as the definition of error bars). If the size of the error bars hasn’t changed much since then, then it’s plausible that the recent swing toward McCain is just the effect of statistical fluctuations.

  • Sam Wang

    Yes, you are reading that correctly. On that day the Meta-Margin was 3.7%. I agree that you can get a CI that way.

  • Paul

    Sam: That graph confused me momentarily, until I realized it is dated Oct 3!

    Looks like the current version of it would be more or less identical, but shifted ~1.8 points left?

Leave a Reply to Bob Yoder (Cancel)