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

A negative bounce for Romney?

September 4th, 2012, 7:30am by Sam Wang

Although I have not seen Tuesday’s data yet (and can’t for some time because of a speaking engagement), enough is known to allow some advance comments. Bottom line: not only is there no bounce so far, the data suggest the possibility of a negative bounce. Update, 12:35pm: yep, there it is. ~10 EV and counting…
History of Popular Meta-Margin for Obama
The story follows…

The Meta-margin above is interpreted just the same way as a regular margin. Calculated using all available state polls, it is defined as how much opinion would have to swing to create an electoral near-tie. As you can see, opinion this year has stayed within a narrow range. The Meta-margin standard deviation (MMSD) has been about 1.0 percentage point. To put it another way, the Democratic-Republican margin has spent two-thirds of its time in a 2.0-point range. If the range is an indication, as many as 98-99% of us are certain of who we want as President (with 51% for Obama). In this context, the 3-point Ryan VP bounce I reported was impressive — but was not enough to flip the race.

This is an extreme situation. In 2004 and 2008, the MMSD was about 2.5 percentage points, typical of a re-election race (which 2004 was). This year it’s less than half of that, reflecting a very decided and/or polarized electorate.

In the first post-convention change, on Monday the Meta-margin moved towards Obama by a fraction of a point. Is this a small fluctuation…or is it the start of an “inverse” bounce? We will know more by later today (updates occur at 8AM, noon, 5PM, and 8PM). An inverse bounce might reflect the public response to the GOP convention, which ranged from tepid to dismal. It’s not great for the nominee to be overshadowed by an empty chair.

What will happen this week as a result of the Democratic convention? Look again at both the Meta-margin (above) and this year’s EV Meta-analysis:
History of electoral votes for Obama

Given the narrow range so far, any post-Democratic convention bounce is likely to be small. I expect the Meta-margin and EV estimator to stay below a ceiling of 4.0% and 320 EV, respectively. A bounce that exceeds those limits would be a noteworthy event.

However, as I’ve written before, all of this is of no import to the Presidential race. Based on polls and likely future movement, a low-assumption calculation gives President Obama’s re-election probability as 88%. That could change, but not because of convention bounces, which don’t last.

Far more important are the Senate and the House, where party control is on a knife’s edge. A lasting shift in the race, whether from the conventions or another source, will mean the difference in 2013 between divided government and single-party control of the Presidency and Congress. Control of Congress is a big story of the campaign, and a source of real suspense. In the coming weeks, I hope more pundits will notice.

Tags: 2012 Election · President

30 Comments so far ↓

  • The Political Omnivore

    For what it’s worth, PoliticIT, an Internet-Influence tracker finds Obama getting more positives during the RNC than Romney which would be a “negative bounce” (I’ll note I can only imagine what PEC thinks of the predictive power of the Internet-based IT score!) … You can read our post on Politic IT here:

  • BillSct

    What i keep noticing is the three plateaus at around 300EV and the long plateau at 320 EV. To me that is a visible indicator of stable public opinion (i.e. people have pretty much made up their minds.)

  • Albert

    Although MMSD does not take into account national poll, Rasmussen seems to have pointed out that the bounce has receded. Romney was +4 yesterday, now only +2. We all know that Rasmussen has a republican leaning tendency.

    However, I would caution that it might be too early to draw that RNC produce no bounce at all to the MMSD – since the model is driven solely by state polls and so far we haven’t seen many swing state polls apart from the one conducted by PPP (Michigan and Florida) and they’re affiliated to the Democratic party. I would personally want to see how the state polls look like this week …

  • Olav Grinde

    Like you, Albert, I am eagerly awaiting new state polls. I am very surprised that there are so few. Especially from the swing/toss-up states, I would have expected a whole sleight of them were being release now.

    So: What’s up with the drought of state polling?

  • Sam Wang

    There should be more polls in the next 24h.

    I have to say from a purely analytical standpoint: a negative bounce is really interesting.

  • JamesInCA

    Re a possible negative bounce: It certainly seems possible if we have an electorate that mostly knows what it thinks of the incumbent, but with a few wavering voters thinking, “I want to know a little more about this Romney fellow.” The convention may have helped them form a firmer opinion. And if what they saw/heard was mostly about a fairly extreme party platform, a vice-presidential-nominee speech roundly slammed for factual liberties, and Clint Eastwood with the chair….

  • Matt

    It really depends how you measure it. Nate calculated a 2 to 4 point bounce based on comparing the same poll before and after the convention for each of the pollsters that have produced numbers. However, it appears that at least some of those pollsters didn’t produce numbers prior to the Ryan announcement so he isn’t just measuring the convention per say. His actual model based on current polling (not the one that assumed a bounce) showed no change either way.

    Apparently your model assumes that the identity of the pollsters doesn’t matter too much and that any new information will be relatively unbiased, if I understand it correctly. From that standpoint you get a negative bounce. If in fact the pollster profile isn’t really a factor, perhaps that’s just the Ryan bounce, which you did register, continuing to recede.

    I think the most accurate thing you could probably say is that we haven’t seen much change at all in the polling that can be contributed to the convention.

  • Pat

    Indeed, Nate find a 2-3 point bounce comparing each of the pollster just before and after convention. But also by comparing the results after the convention with their average values in the last months.

    Also, the running average of national polls as calculated by RealClearPolitics shows a pretty sharp increase in Romney’s numbers (since mid-August) which shows no sign of slowing down.

    So it is indeed pretty surprising that your method shows this pretty large negative bounce of about 15 EV, as there doesn’t seem to have been new polls showing especially better results for Obama than before.

    • Sam Wang

      Pat – Perhaps Nate and the commenters here should learn how to calculate a p-value, confidence bands, and limits of a given statistical approach. Or at a minimum, look at the dates of polls to see what time period is spanned. In the case you cite, it’s over a month, which includes Ryan-VP, Akin, and the GOP convention. That bounce came immediately after the first of these three events.

    • Sam Wang

      Pat – The reason 15 EV looks like a lot is that is that here, the noise is quite small. In the other measures you cite, it is a struggle to make out a signal. On average, 15 EV works out to be equivalent to a little more than 1 percentage point of public opinion. The accompanying tick in Meta-margin is even smaller. National poll averaging can’t do that — though at the moment, there are enough national polls for them to be competitive.

      To re-emphasize, this calculation has a higher signal/noise than national polls. And it is in the units that matter, electoral votes.

  • Albert

    Pat –

    Comparing RCP Nat Average with MMSD is like comparing apple and orange. It’s two different measurements. One is driven by national poll, the other driven by state polls.

    You’re correct in pointing out Nate’s finding that there is 2-3 points bounce, however, if you look at his model there is practically no bounce at all. Obama is now at 71% chance to win according to Nate. Therefore, a finding / change in raw number doesn’t necessarily change the model.

    So, I would advise you to look into the data more carefully and ensure what you’re comparing is comparable. ..

    • Sam Wang

      Actually, MMSD is the standard deviation of the Metamargin. So Metamargin and avg national margin are the comparables here.

  • Olav Grinde

    @Pat: Unless I am misreading their website, RealClearPolitics shows Obama 332 EV and Romney 206 EV (with no states left tossups). has identical numbers, also rating the race at 332–206.

    In other words, both those aggregators actually a higher EV score than Dr Sam Wang has on this site!

  • Pat

    Sure, but that still doesn’t explain what causes this ‘negative’ bounce. Among the rare new polls that have been made, these didn’t seem to show better results for Obama than previously. If almost all new polls showed an improvement for Romney compared to polls of the same pollsters prior to the convention, isn’t it strange to have this negative bounce? Isn’t that a bit of an artefact due to the fact that we may be comparing polls from different pollsters? (i.e. going from a Romney+1 poll by Rasmussen to an Obama+1 poll by Quinnipiac is not a very strong indication that the race moved to Obama)

    PS: I have an extra practical question: how long do you keep each poll for your median calculations? I mean, after what time do you consider a poll obsolete and don’t consider it anymore into your calculations? You certainly explained it in much detail but I can’t quite remember. Thanks

  • Matt

    I’m curious how much of the negative bounce you calculate has to do with Michigan moving toward Obama in your model. Michigan has had some weird and divergent polling this election and seems to bounce up and down in various aggregations depending on who polled it last.

    I tend to think, given the demographics and what not, that Michigan should be as favorable if not more to Obama than Ohio, so I’m less skeptical of PPP’s numbers than some other pollsters.

  • DaveM

    Perhaps rather than representing a “negative bounce,” the upturn in Obama’s margin could be thought of as reflecting a return to normalcy following what we might think of as an “anticipatory bounce.”

    That is, the coalescence of whatever factors have historically produced a post-convention bounce occurred a week or so ahead of the convention: the early selection of Ryan removed that factor from the convention’s environs, the Akin “gaffe” focused the no-exceptions crowd on the no-exceptions platform well in advance of the convention, etc. The base got its red meat early, so to speak.

  • Olav Grinde

    @DaveM: “The base got its red meat early, so to speak.”

    I find the imagery of giving political followers “its red meat” immensely troubling, and telling. I know this expression has become standard, but I am puzzled as to where it comes from. Can anyone enlighten me?

    Well, the right wing of the GOP may have gotten what it yearned for, but I sense that old-style or moderate Republicans — and a plurality of the American electorate — are indeed very ill at ease at the sight and smell of the “red meat” they are being served.

    It is nothing short of unpalatable

  • Brian

    @ Olav Grinde

    I’ve done research on various political phrases over the years, and never been able to find the true origin of that phrase. In a legislative sense, it was used to describe the specifics or actual numbers outlined in a bill or law (the meat and potatoes if you will), but it has evolved to mean rhetoric from a politician to his base. The first time I can recall it being used by a political commentator was after Pat Buchanan’s speech at the 1992 GOP Convention….don’t recall who said he had “tossed red meat to the crowd like feeding hungry lions”.

  • pechmerle

    The Urban Dictionary says that Jon Stewart popularized the phrase in 2011, in describing the way Republican primary candidates tossed ‘what they wanted to hear’ to the audiences.

    As Brian points out, the usage is much older than that. I would bet it goes back long before the 1992 GOP convention too.

  • Amitabh Lath

    What is the mechanism behind this “bounce” that everyone is discussing? The default picture I have of how poll numbers move is that undecided voters make up their minds. They do so as they encounter information, and presumably conventions are information-rich.

    So how do we get a bounce? Do people look at a convention and all the balloons and buy in, and then a few days later come to their senses and revert back to undecided? Or the other guy?

    Or maybe there are two groups of voters, one reacting quickly to the positive info (Ryan, my fiscal superhero!), and another reacting somewhat slower to the negative info (timing yourself on a marathon ain’t rocket science).

    This mechanism only works if you suppose that every bit of news has a positive impact fast, and a corresponding negative impact a little later.

  • Michael K

    @Amitabh: At least in theory, I imagine bounces and other poll movement could occur if one side’s enthusiasm changes such that they are more (or less) likely to pass “likely voter” screens.

    Also, a few “undecideds” could be wavering between one of the major candidates and a thirty-party candidate (rather than between the major candidates).

    I wonder if there is enough data from polls that release both registered and likely voter numbers to break down how much of the meta-margin movement can be attributed to each of these effects?

  • Froggy

    This use of “red meat” in a political context goes back way before Buchanan. Reagan served up “red meat” to his audiences, as did Gerald Ford, Spiro Agnew, and even Adlai Stevenson(!). The oldest example I could come up was from 1950, when Harry Truman had this to say to Rep. Emmanuel Celler of New York, concerning Celler’s exposure of monopolies:
    “Keep it up,” urged Truman. “You’re doing a great job for the country and, incidentally, providing me with red meat for campaign speeches.”

  • Matt McIrvin

    Interesting that the motion in the Meta-Margin is so much smaller, comparatively, than the motion in the median EV count. These electoral votes must be, in some sense, soft… mostly from Florida, maybe?

  • Pat

    Sam – Of course, I totally understand that noise is smaller. But that doesn’t answer my question as to how come this occurs if almost all new polls showed an improvement for Romney compared to polls of the same pollsters prior to the convention. Isn’t that a bit of an artefact due to the fact that we may be comparing polls from different pollsters? (i.e. going from a Romney+1 poll by Rasmussen to an Obama+1 poll by Quinnipiac is not a very strong indication that the race moved to Obama).
    As Matt suggest, it may have to do with which polls happen to be included in the median while other polls (from different pollsters) are phased out. Again, what is the time frame over which you calculate the median? (all polls in the last xx days?)

    • Sam Wang

      Pat – The Meta-analysis is able to combine the effect of many small changes in polls. Your rough argument misses this quantitative point. For example, if three polls for one state show a lead of 3% and an SEM of 2%, the leader’s win-today probability is 88%. But if the lead slips to 2%, then the win-today probability slips to 80%. Combining many of these probabilities gives considerable statistical power.

      As for comparing different pollsters, I have written about this extensively in past comment threads. My experience is that it doesn’t matter as much as you think it does. The whole poll-sniffing game, so popular with fans of the genre, becomes less important with sufficient statistical power. Anyway, wait a day for more clarity.

      Your other questions are answered elsewhere on this site.

  • Olav Grinde

    @Brian, Pechmerle & Froggy: Thanks for casting light on the history of the “red meat” expression. Most interesting!

  • xian

    the meta margin just started tailing back toward romney. is this the ebbing of a negatve bump, or … what exactly?

  • Josie

    I’m curious how Virgil Goode and his newly approved place on the ballot will work in Virginia. Since it’s a dead heat there and he is a local from the rural part of the state, the Republicans would take a hit. They are currently challenging his petition signatures.

  • Obama gets a lift, Romney flatlines in post convention polling | Harry J Enten | WorldNews2GO

    […] supposed to gain about four points thanks to the convention, but he probably gained no ground, or even lost some in the wake of the convention. Compare this outcome to a Democratic national convention that has, […]

  • Obama gets a lift, Romney flatlines in post convention polling | Harry J Enten | Old News

    […] supposed to gain about four points thanks to the convention, but he probably gained no ground, or even lost some in the wake of the convention. Compare this outcome to a Democratic national convention that has, […]

Leave a Comment