Princeton Election Consortium

A first draft of electoral history. Since 2004

Meta-Margins for control: House D+2.8% Senate R+4.8% Find key elections near you!

Obama’s red ceiling

October 3rd, 2008, 3:32pm by Sam Wang


I just gave a talk over in the Politics department/Woodrow Wilson School. A lively group! It was edifying and fun, and I was sorry to rush off. Here are my slides as PDF and PowerPoint. Unfortunately, they don’t capture the interplay or the chalkboard work, which I’ll write about later. But first, some thoughts on where the race stands.

The EV estimator has dropped slightly from yesterday’s high of Obama 349 EV, McCain 189 EV, to Obama 344 EV, McCain 193 EV. Looking back, these small downticks often indicate that we’ve reached a maximum or minimum. The 95% CI is [308, 365] EV. (By the way, that information is always available at this link. The variables are: Obama/McCain medians, modes, safe and toss-up, Obama 83% CI, Obama 95% CI, date, and Meta-Margin. More geekery here.)

Even if yesterday was not a maximum, Obama is certainly approaching a plateau. We can see this by assuming that the margins were to swing in all states by the same percentage. In these scenarios, the EV estimator would change like this:

bias plot for 3 October 2008 3:00PM

bias plot for 3 October 2008 3:00PM

In this plot, a value of shift=0% corresponds to today’s polls (Obama 344 EV). If McCain picks up 2% of margin across the board, the median shifts by 35 EV. But if the shift is 2% toward Obama, the median only shifts by 20 EV. This asymmetry results from a core of strongly Republican states that are totally out of reach for Obama. After Missouri and Indiana, it would be hard to capture much more.

To give you an idea of the difficulties, the states where McCain currently has his smallest lead are MO (1%), IN (2%), WV (4%), GA (8%), TX (10%), MT (11%), and KY/AR (12%). Another indicator is the probability histogram, which is spiky because of the relatively small number of states in play based on polling data. I give Obama a “soft” ceiling of 380 EV. Soft because there is always the outside chance of a larger swing than what we have seen.

Purely as a matter of systematic polling error, a 2% shift toward Obama is not out of the question due to the cell-phone effect. A 2% shift toward McCain is less plausible due to the disappearance of the Bradley effect. Actual swings in opinion over the coming month an interesting story, and a subject for a future post.

And now, I am off to work on an exciting research manuscript that we are about to send…

Update: by request. As you can see, the relationship is roughly linear between -10 and 0, which corresponds to McCain ahead by 6% to Obama ahead by 4%. Beyond that, variation in support for the candidates starts to get “lumpy,” especially in red states. Note that this is only a crude approximation since for swings this large, demographic factors are likely to cause swings themselves to be unevenly distributed.

Tags: 2008 Election

17 Comments so far ↓

  • gprimos1

    Very informative. It would be fantastic if this could be one of your regularly generated charts.

  • Vijay

    Well, since the election has now started getting a little boring, how about doing some Meta-Analysis to calculate the chance of an *Obama-Palin* or a *McCain-Biden* victory?

    No, you didn’t read that wrong. There is some chance that the country could have Obama as Prez and Palin as Veep. And both campaigns, humorless spoilsports that they are, are apparently devoting some effort to preventing this occurrence. For details, read this article:
    http://www.realclearpolitics.com/horseraceblog/2008/10/nebraska_and_maine.html

    The short version is this:

    For the Obama-Palin team to win, the first condition is that there be a tie in the electoral votes. (The EV histogram currently gives this about a 5 in a million chance.)

    Once there is a tie, the House picks the Prez, with each state getting one vote; so the second condition is that the House picks Obama. Since the Dems control the state caucuses in the House, this has a pretty good chance of occurring.

    And in case of a tie, the *Senate* elects the VP; so the third condition is that they pick Palin. Again (if I understand the rules correctly) each state gets *one* vote, and so it suffices for the Republicans to have more states than the Dems even if they don’t have an absolute majority in the Senate. (For example, since EV tie scenarios most likely give more states to McCain, those states could insist that their senators vote for Palin.)

    But the most evil scenario would be if the Senate vote were also to be tied and the tie-breaking vote came to the man who says he is part of the Legislature and not the Executive … Dick Cheney!

  • RKC

    Many thanks for your great information. Your proportional EV map finally got the concept through my 12 year old’s head when I was ready to give up!

  • Pete D

    You’ve made a very good point, Sam. Besides keeping all of Kerry’s states, it looks like Obama would pick up a significant number of red states if the vote were held today – NM, CO, NV, IA, OH, VA, NC and FL. It looks as if IN and MO were almost dead heats and if Obama won those states, he’d have 375 EV.

    Interestingly, the only states Obama would lose in which he had any ambitions would be AK, MT, ND and GA. It’s quite possible that Palin’s spot on the ticket was responsible for keeping the first three states red. Alaska’s her home state, of course, and MT and ND are two Interior West states without a significant Latino presence. Now, Georgia? SUSA’s latest poll shows only 23% of Georgia’s white voters are planning to vote for Obama – a figure much closer to AL and MS as opposed to VA and NC.

    From my perspective, there are really two surprise elements contributing to Obama doing as well as he is so far. First, the Latino vote. Obama definitely is doing better with that electorate than Kerry or Gore. In 2006, I warned fellow Republicans that nativism didn’t go well with the non-White Anglo crowd (as if I should have had to warn them!). Second, Obama’s performance among white voters. I came across a magazine article from last spring that pointed out Obama would need a whopping 40% of the white (or, more correctly, white Anglo) vote in order to win the presidency. Well, just in VA, FL, IN and MO, Obama’s picking up between 40-43% of that vote.

    But, you’re quite right, Sam. I think Obama’s hit his peak. At best, he can top Clinton’s performance in 1992 (370 EV) but probably not his performance in 1996 (379 EV).

  • Eddie

    That’s such a smart and useful plot with this post. This is not a request, but it would be interesting to see it extended more percentage points out– I suppose there are a few plateaus– then again, I’m not sure if that works meaningfully.

  • Michael

    “And in case of a tie, the *Senate* elects the VP; so the third condition is that they pick Palin. Again (if I understand the rules correctly) each state gets *one* vote, and so it suffices for the Republicans to have more states than the Dems even if they don’t have an absolute majority in the Senate. (For example, since EV tie scenarios most likely give more states to McCain, those states could insist that their senators vote for Palin.)”

    This is incorrect. Each Senator gets a vote, and the Senate will be decidedly Democratic. There is zero change of the Senate electing Palin VP. There is a very small chance of a McCain/Biden administration, but not Obama/Palin.

  • Andy F.

    It is interesting that Obama is ahead in the national tracking polls by about 6.5 points but a 4 point drop brings him to parity in EV. Looking back at the national polls vs. the EV analysis here, one sees that even when McCain was ahead in the polls by about 2 point for a week, Obama’s expected EV total never dropped below 270. Sam, are the battle ground states somehow less volatile than the country as a whole? If so, do you think it is the barrage of advertising that keeps them somewhat insulated from events that take over the news cycles?

  • Sam Wang

    Andy F., that’s an excellent point. The disparities between the national polls and the Meta-Analysis is pretty interesting indeed. In general, national swings seem to be “attentuated” in the Electoral College.

    A few weeks ago the post-RNC bounce in national polls seemed to be disproportionately concentrated in red states. I wrote about it at the time, and it’s the phenomenon you have noticed.

    Today, if you look at the Pollster.com data, big battlegrounds such as OH, VA, and FL do seem to have swung less than national opinion. It could indeed be because of advertising, of which I don’t see any in low-voter-power NJ.

  • Michael S. Cullen

    Gotta question. Real Clear Politics today gives a spread in Minnesota of 18 points to Obama, two days ago McCain led by ONE point. 19 points in two days? Can anybody explain?
    Michael S. Cullen
    Berlin, Germany

  • Sam Wang

    Michael S. Cullen, you are quoting individual polls, which vary for three reasons: (1) true movement, (2) methodological differences among pollsters, and (3) random variation.

    In this case the extreme poll was conducted by the Minneapolis Star-Tribute. Local polling organizations often do not use standard practices to make their results more accurate. Newspapers seem to be particularly bad offenders. You should do what we do here, which is take a median of recent polls. The median of the last three polls in MN is currently Obama +11%.

  • Steko

    “At best, he can top Clinton’s performance in 1992 (370 EV) but probably not his performance in 1996 (379 EV).”

    From a polling standpoint perhaps. But we’ve seen what one sided advertising did for McCain in MN.

    If the current margin is maintained in this campaign, McCain will throw all his money at FL/OH/VA while Obama (who has more money to begin with) will have the WV/TX/MT airwaves all to himself.

  • Dave

    “This is incorrect. Each Senator gets a vote, and the Senate will be decidedly Democratic. There is zero change of the Senate electing Palin VP. There is a very small chance of a McCain/Biden administration, but not Obama/Palin.”

    I think from a political standpoint, no senator would choose Palin over Biden, not to mention the 99.9% chance of a democratic senate. Even Lieberman isn’t going to vote for Palin if McCain has no shot.

    McCain/Biden will only happen if multiple one-vote-majority democratic delegations, from states that voted for McCain, have a representative turn red on them.

    They should know this by the time the electoral college meets. If that’s going to be the case, Obama can instruct his electors to put McCain forward for VP, so that the house can vote for Obama and the senate for McCain. That’s fairly unlikely, but may be Obama’s only play if multiple democratic congressmen can’t face rebuking their constituents’ presidential choice.

  • Fokko van de Bult

    I am confused as to why the upper and lower bounds of the 95% percentage interval would not be an increasing function of the percentage of swing Obama gets. You can clearly see in both graphs these bounds move generally up, but sometimes they dip a little. Can anyone explain this?

  • R Clayton

    Interesting analysis — of course. Weakness is that it presumes national swings would be uniform across the nation. However, it is possible that Obama’s lower-performing states are those in which he has more support-growth potential than the higher-performing. Further, of course, one would expect differentials in support growth based on how much the campaign invested in advertising, organizing, candidate and surrogate time, etc. So a five-or-more point swing in a given state seems at least theoretically possible even with only a minor national swing, or none at all. Of course, it may be that in the states where Obama is now performing poorly, there are virtually no persuadable voters due to the state’s particular demographics and political culture. That suggests a state-by-state analysis would be necessary to truly evaluate Obama’s EV ceiling.

  • Sam Wang

    Fokko, I noticed that. I agree that the CI bounds should be monotonically increasing. I think there might be a rounding error lurking in there. If you know MATLAB (you gave an email address that suggests this possibility) you are welcome to pore over the data. I’ll look into it.

    R Clayton, agreed.

  • Matt McIrvin

    Elsewhere, people are starting to speak of landslides. This annoys me because it clearly isn’t going to be a landslide in the traditional, Reagan sense. Electorally, if current patterns hold Obama will have a solid win somewhere near the Bill Clinton level, made somewhat more impressive because of the lack of the confounding effect of Ross Perot. But we’ve become so used to knife-edge squeakers that this looks immense.

  • Mark

    I’d really like to see some fresh polling data for Louisiana by someone like CNN with a sample of 8-900. The ARG poll 9-12 September (ie. before the crash) had McCain +7; the Rasmussen margins have been much greater but they look skewed. It could be that Louisiana is much closer than people assume; Georgia has also narrowed by about 5% since Sept 15. On that basis, if ARG were right, Louisiana would be within the margin of error.

Leave a Comment