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 possible House flip enters pundit imaginations

September 28th, 2012, 8:00am by Sam Wang

Eight days (or five weeks, take your pick) after I pointed it out, the idea of a Democratic takeover of the House has crept into the national conversation. Yesterday, Rachel Maddow (citing us as evidence) discussed the possibility with Nancy Pelosi, who thinks the probability is about 60%. Clearly Pelosi’s duty out on the basketball court is to engage in trash talk. But this week, conservative stalwart Bill Kristol said:

an Obama +3 victory on Election Day [could] drag the Democrats to an edge in the congressional vote—and control of the House. In any case, based on current polling, I don’t think one can say that it’s now out of the question that we could wake up on the morning of November 7 to the prospect of … Speaker Nancy Pelosi.

There’s been continuing pushback on my analysis. The math I used is fairly simple – and it needs to be covered before we get into any of the fine details, which have attracted so much notice.

First, some basic ideas.

Basic idea 1: Whoever wins the popular Congressional vote will control Congress. Here is a plot of House seat margin, plotted against national House popular-vote margin, for 1946-2010.

I have shaded in the regions where one party both won the popular vote and took (or kept) control of the House. Note that 32 out of 33 elections fall inside the shaded regions. Considering that the House is determined by 435 separate elections, many in convoluted districts, how can this be?

Most of you know that in the Presidential race, the winner of the popular vote usually wins the election – but not always. When it doesn’t happen, it’s because the Electoral College is “lumpy”: most states’ electoral votes will go entirely to Obama or to Romney, whether the win is by 1% or 30%. House elections are like that too, except that there are 435 of them, making the “lumps” really small.

It is only when the national-vote margin gets to less than 1% that Basic Idea 1 breaks down. Even then, it failed only once out of four elections – in 1996, when the margin was 0.3%.

Basic idea 2: Generic Congressional preference polls can be used to predict the November popular vote outcome. I won’t revisit the whole argument. I outlined it here, here, and here. It consists of two parts: (a) In September and October of on-years (e.g. 2004, 2008), these polls tend to move slightly towards the Presidential winner. (b) By Election Eve, an average of polls ends up very close to the actual vote.

Based on these two ideas, the Democrats’ current narrow lead suggests that this November, they are favored to re-take the House. To turn this to a probability, we have to consider the exact Democratic-Republican margin today and the range of possibilities for movement. This is not a complicated calculation. It leads to the estimate I gave last week.


Here are some objections I have encountered, along with my replies.

Objection 1: Incumbents have an advantage. This might be true, but the question is: how much? I made a rough estimate that incumbency is worth an equivalent of about 1.3% of popular vote for the party in power. Even that might be an overestimate, considering Basic Idea 1 above.

Another issue with the incumbent argument is that it is in the same category as “fundamentals”-based arguments that political scientists put into their predictive models. In other words, it is a cause that drives voting preference – which is measured by the generic Congressional preference poll. And if it is already baked into the measurement we have in hand, it does not need to be further accounted for.

Objection 2: Redistricting has tilted the playing field. In many statehouses, Republicans have controlled the post-Census redistricting process. In Slate, Dave Weigel has claimed that gerrymandering makes the House safe for Republicans.

I do not have much patience for this kind of qualitative argument. Also, Slate‘s editorial policy is to be contrary – even if it means being wrong. Again, the real question is: exactly how large is the “structural” advantage? Most of the data in the graph above came during a period of prolonged Democratic dominance. If you look at it, any structural advantage for them was no more than 1% of national popular vote – and it’s likely to overlap with the incumbency advantage. After only one cycle of redistricting, it is doubtful that Republicans can match even this small number.

How can redistricting accomplish so little? One answer is that from year to year, elections are uncertain. You can pack a lot of your opponents into a few districts, but if your own districts are only 55% for your own party, a 10-point swing can knock you out of office. And the swing from 2010 to 2012 is currently about 9 points. Another answer is that redistricting is sometimes done to protect specific incumbents – which results in packing one’s own party members into a district. On average, the whole thing could well be a wash.

As an exercise, I suggest that you go to’s House summary page. Mouse over the “leaning Republican” and “leaning Democratic” seats. You’ll find a lot of small margins. Even a 5-point swing in either direction would make a big difference.

Objection 3: Professional prognosticators say it won’t happen. It has been pointed out that the Cook Report, Stu Rothenberg, and others disagree with me – and that they are often right. Well, let’s look at that. Here is how they (and I) did in 2010:

There I am in the middle of the pack, not far from Larry Sabato. Collectively, forecasters underpredicted the Republican victory, with an average error of about 12 seats. And that was on Election Eve. We got away with it because the midpoint of the likely range was far from a tie. Nearly all of us correctly said the Republicans would take over the House, so we got little grief.

This year, the midpoint is close to a tie. And every prediction has some uncertainty, which prognosticators usually do not report. Misunderstanding uncertainty is a conceptual error that I discussed a few days ago. The bottom line is that the pundits’ margin of error allows the possiblity of a Democratic or a Republican House.

Objection 4: 74% is such a large probability. You are crazy! Actually, it is not large (and I am not crazy, mostly). The probability is a logical consequence of my estimate that the national election margin midrange will be between D+0% (tie) and D+8%. People don’t mind that. Turn it into a probability, and they freak out.

“74% probability” and “D+0% to D8%” are mathematically equivalent. And 74% is not that significant. Many responders have an unclear understanding of the relationship between a confidence interval in vote margins and a win probability.

Generally, any probability in the the 20-80% range is a knife’s edge situation. Under current conditions,  the Democrats are favored to take the House…but there is a 26% chance that they will not. Why do you think Bill Kristol is sounding an alarm about the House, but not the Senate? The Senate has slipped away from the Republicans – but not the House. The probability I have given can change – but in which direction? This uncertainty is contained in my calculation.

For example, a recent single Bloomberg poll says R+2%, hopeful for the GOP. However, generic polls are noisy data. Even the median jumps around a bit. Like you, I wait for more information.

Objection 5: District-level polls will tell us what’s going on with more precision. In fact, this is true. In the last Presidential year, 2008, district polls predicted a likely outcome of 254-260 D. The actual outcome was 257 D, right in the middle of the range. As this year’s district polls come, we’ll learn much better what’s going on.

Tags: 2012 Election · House

58 Comments so far ↓

  • Barry

    Hi Sam…the rule of thumb out there has been that Dems need an average margin of 3.5% to 5% or greater to win the House, b/c Dems actually get to the polls less reliably than Repubs.

    If I understand your thinking, you’re indicating this margin is not needed, and that even a 2.5% Dem margin in the average (or median) generic poll is sufficient for Dems to gain control. Is this your assessment?

    • Sam Wang

      The graph suggests that such a rule of thumb would be a real outlier. If you like, post a link to the argument. Another estimate is D+3% to D+4%. But I dissected that in the original analysis and in follow-up. Please read that. At the time I assumed that D+1.3% was necessary.

      Other readers: do not re-ask questions that I have already addressed. Most points are contained in the uncertainties that I originally assumed. Rather than create a long, repetitive thread, please read the other posts linked from this essay.

  • RadicalCentrist

    I wouldn’t say it’s out of the question, but 70% really is out of line. The current RCP average is D+1. But that includes a Reuters poll of registered voters that was D+6. Include only likely voters and you have a tie.

    Then, it may be anecdotal, but I live adjacent to NY-19. This district has swung twice in the last few cycles, having been won by Sen Gillibrand in 2006, then very narrowly by a Dem in a 2008 special election and then by a Republican in 2010. A recent Siena poll has the R incumebnt up 16. There is no way that jibes with a 70% chance of a D takeover.

    I would put the odds at 30%, perhaps, at current polling.

    • Tim

      Sully just posted on this, and cited a skeptical Charlie Cook. Though it doesn’t look like Cook was responding to you.

    • Sam Wang

      Well, I do respect Charlie Cook, and he does this full-time. However, his estimates are snapshots, and change over time. In September 2010 he forecast GOP gains of over 40 seats, which grew to 50-60 seats by Election Eve. The outcome was a gain of 63 seats. That was a wild year, not his fault to be off. The point is that his September forecast was about 20 seats off base. In 2010, it didn’t matter in the sense that the GOP would have taken control with the smaller gain. This year, such an error would flip the sign of the outcome. So I think it is early to reduce House control into a categorical yes/no statement at this point.

  • Amitabh Lath

    Yes! 74% CL is not that significant. In the sciences you don’t even get out of bed for anything less than a 95% CL and to claim even a hint of a signal takes better than 99% CL (3 sigma).

    That being said, any number within shouting distance of 50% says it is in play. Hope is in the House.

    The innumeracy on display in some of the pushback posts is painful. The Weigel piece in Slate was a good example. Lots of anecdotes. No numbers.

    In general, I trust numbers over narratives.

    Next time you are on with some of these talking heads, maybe teach them about simple single-variable integrals. Sheesh!

    My thesis advisor (who founded the Union of Concerned Scientists) used to warn us to stay within the inner circle of certainty. Say exactly what you did, and don’t apologize for the data.

    • Sam Wang

      Amitabh Lath is exactly correct. 74% is not that significant. RadicalCentrist, think about the relationship between a confidence interval in vote margins vs. a win probability. I have made two mathematically equivalent statements. One of them seems surprising to you. My view is that intuitions should come into line with the math.

    • Joel

      I’m innumerate, and I thought you were crazy about the 74% thing, but it actually makes sense when you explain it.

      I mean, in some ways (in every concrete way in my life, say), 74% is a toss-up. If Romney offered to bet me $10,000 based those odds, I’d say no. I can’t afford to lose $10,000, and I probably (not _your_ kind of probably, _my_ kind of probably) would.

      If they said 99%, I’d want to bet $100,000.

      But 74%? No way.

    • Sam Wang

      Exactly! Joel, you’ve made my morning. An innumerate person gets the argument! However, I am not so sure how innumerate you really are…

      Everyone else, my repeated use of the phrase “knife-edge” might be a clue as to where the House stands.

  • Amitabh Lath

    Sorry to disagree with you RadicalCentrist, but I disagree with 30%. The correlation with the popular vote margin is unmistakable. The polling this last few days is also pretty solid +D in several states.

    If you accept those two things, then the numbers are what they are. No gut feelings allowed, sorry.

    • RadicalCentrist

      Amitabh Lath: Sorry to disagree with you. The polling in several states is solid D+ for President. That doesn’t necessarily correlate to a House vote. If it did, the Dems would have re-taken the House in 1996 when Clinton was re-elected with a comfortable margin. Note that in the district near me, NY-19, where the R incumbent is up by 16, Obama leads by 4. So the correlation isn’t good.

      The reality is that the generic Congressional ballot on RCP is effectively tied. Now look at the graph above. There were 4 elections (not sure which years) where the vote was very close to even. Yes, in 3 of those, the individual results followed the popular vote, but, I’m sorry, n=4 simply lacks any statistical weight. Do a chi-squared test on 3 out of 4 if you doubt me and tell me the p value.

    • Sam Wang

      RadicalCentrist, your email address you gave me is fake.

      I am giving you a timeout for a little bit. You are starting to give wrong facts, for instance failing to notice that the poll median is calculated from data. It is hard enough to respond to legitimate concerns.

    • JGabriel

      RadicalCentrist: “The polling in several states is solid D+ for President. That doesn’t necessarily correlate to a House vote. If it did, the Dems would have re-taken the House in 1996 when Clinton was re-elected with a comfortable margin.”

      Sam Wang: “I have shaded in the regions where one party both won the popular vote and took (or kept) control of the House. Note that 32 out of 33 elections fall inside the shaded regions. … It is only when the national-vote margin gets to less than 1% that Basic Idea 1 breaks down. Even then, it failed only once out of four elections – in 1996, when the margin was 0.3%.”

      So there’s no correlation whatsoever, RadicalCentrist? The percentage of people who show up to vote for a Democratic President tells us nothing about what percentage are showing up to vote for a Democratic Congress?

      Interesting hypothesis. The data don’t seem to agree. Outside of the 1996 election, do you have any other evidence to back up it up? After all, one data point is anecdote, not a pattern.


    • Sam Wang

      People do love to talk about that 1996 data point, don’t they.

    • Sam Wang

      One problem is that these generic polls are noise-filled beasts. The recent Bloomberg and Rasmussen polls are suggestive of a possible swing back toward the Republicans. This kind of thing has happened all summer, but the overall picture is a Democratic lead.

      I am afraid we have to keep watching those bits of noise come in. As I wrote in my original prediction, there may be movement. Obviously I will revisit this question.

    • wheelers cat

      RadicalCentrist:”The polling in several states is solid D+ for President. That doesn’t necessarily correlate to a House vote. If it did, the Dems would have re-taken the House in 1996 when Clinton was re-elected with a comfortable margin.”
      The 1996 data point ignores the past 16 years of demographic electoral evolution. In 1996 non-hispanic caucasians made up 83% of the electorate. In 2008 74%. The percent white voter is projected to be 70 to 72% in 2012.

  • Dave

    Sam, two questions: first, I note that you didn’t include a red dot for yourself in your graph for a September 2010 forecast. Why not?

    And second, in 2010, most prognosticators were off by a bit (as would be expected), but you were roughly in the middle of the pack. This year, my sense is that you are more toward one of the poles. Could you provide a similar chart comparing current forecasts for the November 2012 election?

    • Sam Wang

      That’s a good point. Unfortunately, it’s not my graph. Click on it. I believe the Monkey Cage people have mixed Election-Eve snapshots with long-term predictions. Note that in 2010, an off-year, movement was likely to go toward the out-party. It may be that even late snapshots did not capture the last-minute moves. This year, those moves are likely to be toward the winner of the Presidential election.

      You are correct that September prediction has even more uncertainties. I included these in my original analysis. Please read that for the gory details. So many gory details!

  • Olav Grinde

    Thanks for a crystal clear summary of the reasoning behind your prediction/calculation –and especially succinct objections to other pundits’ objections.

    Dave, I don’t think Sam is necessarily “more toward one of the poles”. Most likely he is early. :) Let’s see what the other pundits predict a week before the election; I suspect more of them will belatedly offer predictions in line with Dr Wang.

    • wheelers cat

      I suspect other predictions will fall in line with Dr. Wang. But will this be mathematical convergence based on pure analytical statistics, or flocking behavior based on Social Network Theory influence?
      Or alternatively Rasmussen data massage so he can claim credibility post-2012?
      Hard to say.

      I do not think 2010 is predictive of the 2012 electorate. The average difference in turnout in election years over the past five election cycles is +16.1% for the presidential. The midterm is always a referendum on the preceding presidential. The extreme animosity exhibited by the republican base for President Obama was obviously a factor.
      I think it was unable to be modelled by a prior.

      “Note that in 2010, an off-year, movement was likely to go toward the out-party. It may be that even late snapshots did not capture the last-minute moves.”
      –what he said.

    • Dave

      I have a tremendous amount of respect for what Sam does, but I do think his current prediction is significantly more optimistic for the democrats than just about any other pollster’s current prediction. Now I hope he is right, but still, if he is citing past performance as an argument that his current predictions are credible, I think it is worth pointing out that whatever else may happen between now and November, his current prediction is not be “middle-of-the-pack.”

      I also note that in an apples-to-apples comparison, Sam’s 2010 prediction is not really middle of the pack. Shouldn’t we judge his November prediction against other November predictions?

      Again, I’d love to see his predictions come true, and I actually think we may: I think Romney is likely to turn out to be a drag on the entire ticket, and the Republican part is a bit tone-deaf. One more scandal from a candidate on either side could go a long way to determining how this all ends up. But I don’t think this outlook would be justified if the election were held today, for example.

      Of course, it isn’t held today….

    • Sam Wang

      Reasonable skepticism, though I would be happier if you had a clear statement of which numerical assumption was incorrect.

      Funny, I really think you are misinterpreting the number. You really doubt it could go either way? Hmmm. See nonmath post.

      Note I did not attempt September prediction in 2010. The point is not that I am accurate – the point is that the “authorities” cited are not better.

    • wheelers cat

      “I do think his current prediction is significantly more optimistic for the democrats than just about any other pollster’s current prediction”
      And Dr. Wang’s current prediction is based on pure statistical analysis and not “feelings”. If you “think” Dr. Wangs elegant and transparent methodology is flawed and leading to unfounded “optimism”, you should be able to provide supporting data for your hypothesis. If you “feel” Dr. Wangs prediction is an outlier from other poll aggregators, then you should be able to defend that position with specifics.
      I absolutely love that PEC is data-based.
      Haruspicy was for the Ancients.

  • Jacob Hartog

    It seems like you could do a couple of things.

    a) Model each house race individually, using a multi-level/hierarchical framework, in which your predictors are both national-level variables (predicted 2012 presidential margin, generic congressional ballot) and district-level variables (incumbency, 2008 and 2010 house margin, 2008 presidential margin, recent polling). If you compute empirical bayes point estimates for the individual races, it will tend to “shrink” the margin away from the extremes and towards the national median. I assume this is what 538 does, but I don’t know.

    b) Use the generic congressional ballot, estimate variance of vote margin in all 435 house races using the last 4 presidential-year results, and assume that the margins in individual contested elections will take a t distribution centered around the generic congressional ballot and with the variance you calculated. From that you should be able to quickly simulate a predicted range of outcomes, and generate a confidence interval, without worrying at all about what any individual race will do. This seems the more PECish approach…

  • Ebenezer Scrooge

    Sam may have enough data to test the Weigel hypothesis, without looking at individual districts. If gerrymandering is important, its effects will show most strongly in election years ending with 2 and 4, and most weakly in election years 8 and 0. If he appropriately colors the dots in his first scatterplot, something interesting might pop out. Or not.

  • The Tragically Flip

    It’s amazing to me that this is even a controversial prediction. The trend towards increasing nationalization of Congressional and even state level elections has been long apparent. The last 3 Congressional elections (at minimum) have been almost entirely won or lost based on national political forces.

    Most voters don’t know their House member (the death of local media seems germane here) and the increasing polarization of the parties (driven primarily by increasing conservativism of Republicans) means there are fewer and fewer House members who are distinct from the norm for their party anyway. The iconoclasts get crushed by the Club For Growth and the Tea Party anyway so the chances that your Republican incumbent actually is a down the line generic conservative Republican are pretty good, if you like that you’ll vote for him/her, if not, you’ll vote Democratic.

    Going forward, the winner of the Presidential election is the safest bet for who controls the House. Those arguing otherwise should face the steeper climb.

  • Todd Horowitz

    The gerrymandering effect should be the strongest in the election immediately after redistricting, i.e. those ending in ‘2’, and then decrease in years ‘4’, ‘8’, and ‘0’ as voters move around.

  • Dr. Grant Denn

    Hi, heard the interview on POTUS Sirius/XM 104 yesterday and I just wanted to say I was thinking the same thing (about a possible democratic takeover of the house.) Great interview- thanks- Grant Denn

  • cafl

    Isn’t Ebenezer pointing to the census year as the prequel to redistricting?

  • William Ockham

    I want to suggest an alternative method of predicting the House outcome. In recent election years (since the end of the Southern Democrats as a semi-official third party) there has been a strong correlation between a party’s Presidential vote share and that party’s House vote share. In 2008, House Dems received a larger share of the two party vote (SotTPV) than Obama did (by about 2.5%). In 2004, House Dem SotTPV was about the same a half a percent higher than Kerry). In 2000, the House Dem SotTPV was a 3/10 of one percent less than Gore’s.

    You can argue that the 2.5% difference in 2008 was due to racial animus towards Obama (possible, it tracks fairly closely with other measures) or you can argue that because Obama is an incumbent this time, House Dems will do slightly worse than him. I would just say that the most accurate predictor of the House outcome is the Presidential polling average.

  • njorl

    “People do love to talk about that 1996 data point, don’t they.”

    While I generally agree with you, I think the 1996 results are more important than you acknowledge. You note that 32 of the last 33 elections follow the trend that control of the house is predicted by the congressional popular vote, but electoral dynamics have changed significantly over that period. The deep south routinely sent a disproportionately high percentage of Democrats to congress given its aggregate congressional voting up until the institution of the Republican “Southern Strategy”. There were non-electoral forces at work which are, hopefully, not applicable today.

    Just as it might be instructive to analyze the data with respect to “years-after-redistricting”, I think it might be instructive to break it down into “before-Southern-realignment” (1948-1962), “during Southern realignment” (1964-1984), and post realignment.

    Granted, that only makes the 1996 data a 1/13 fluke rather than a 1/33 fluke.

    • Todd Horowitz

      @njorl: I think you’re making too much of the 1996 point. Sam didn’t plot the regression line, but the 1996 outcome does not look like an outlier to me. The popular vote predicts the number of seats, but not perfectly, so if your edge in the popular vote is narrow, you may not win control of the chamber. If the Democrats had won the popular vote by a large margin, but were still stuck as the minority in the House, then we could talk about “non-electoral forces at work”. But that’s not what happened.

  • Peter Norvig

    If the attention is shifting to the house, why does your “Act Blue” page still list only senators?

    • Sam Wang

      No, quite the converse! It emphasizes the DCCC at the very top, and lists Senators as an added option. Trying to be clear there, perhaps failing…

  • Michael S

    “The innumeracy on display in some of the pushback posts is painful.”

    This is the fight here. And pretty well everywhere.

  • Peter D


    Would you please be so kind as to make the data behind the scatterplot available? That would be wonderful.

    I would just like to test a few hypotheses of mine as to what drives the margin without deluging you with questions. I promise to share anything worth sharing.


  • JaredL

    Great post, Sam.

    As a poker player, at least before online poker got shut down, it’s easy for me to see that 74% is no lock. It’s about the odds of winning with AK when your opponent has KQ. Put in non-gambling terms, 26% is just a bit more likely than tossing heads twice in a row. If you regularly toss coins, you wouldn’t be at all surprised if that came up.

    One very small point:
    “In other words, it is a cause that drives voting preference – which is measured by the generic Congressional preference poll. And if it is already baked into the measurement we have in hand, it does not need to be further accounted for.”

    I definitely like your argument against Silver’s model on a similar line. Nate is putting 30% weight on economic and other indicators, but it’s pretty clear from polling that people know what’s going on in the economy and like Obama just fine. It certainly seems to me that adding it in, at least this late in the game with so few undecideds, is mostly noise.

    In the case of generic-congressional ballots, I’m not so sure. It’s a cliché that people hate congress but love their congressman. To the extent that it’s true, it could apply here. It could especially be true with the tea party and more traditional Republicans – people could prefer generic Democrats to Cantor but not Boehner.

    This is a minor point, but I’m not as quick to dismiss fundamentals-based arguments when the polling and voting responses are different.

  • Patrick McL

    Re: 74%

    Yes, that number feels to the human gut like a high probability.

    But what it’s saying is that based on the *math* (guts don’t do math), the odds of the GOP retaining the House is equal to flipping a coin twice and getting heads both times.

    That’s really not all that unlikely an outcome.

    This is, broadly speaking, just a careful examination of what electoral coattails mean (at least on the Democratic side. I’m curious if the same’s true if numbers swing the other way. Is this symmetrical?).

    And it makes sense when you think of it that way. The larger Obama’s coattails (or the larger the cratering of the GOP ticket at the top…), the more marginal GOP voters will just stay away. Why vote, when they know Obama will pound Mitt in their state? The slightly less marginal voters–who follow senatorial candidates–who see the GOP candidate in their state in the same light (either a clear loser anyway, or perhaps so toxic…), the more they’ll just stay away.

    Widen the margin enough, and in the extreme case, where everyone KNOWS that Obama will win their state, how many will stay home and not vote? Human psychology will bring out more who want to vote for their winning candidate. The down-ticket impact only grows. So we’re at–what Sam’s math tells us, not our guts–the point where any tails result in two coin tosses gives the Democrats the House. That may or may not make you happy, but it is the math.

    Critique the math, and any assumptions, not the results.

    (Me, I’d like to see things move to the at least three coin flips level for the GOP to retain the House. But that’s my gut, not the math…)

  • Sam Wang

    JaredL, I agree that there is some incumbency advantage – what you refer to as loving your Congressman but hating Congress. Last week I did a simple fit to estimate the size of the effect. It appeared to be equivalent to 1.3% for the Republicans. But with a big error bar. I am fairly sure it is not larger overall, because of the graph.

    Ideas like yours ought to be convertible to an equivalent fraction of popular vote. In some sense, I use the graph to show the extremes that can be reached by these means.

    I would be delighted to see redistricting treated the same way. I have a feeling someone has addressed it. But I don’t think Weigel did. That issue is still nagging at me a little.

    • William Ockham

      Not all incumbency advantages are created equal. Andrew Gellman has done some work on how much incumbency is worth in individual races. Don’t have link at the moment, but it can be as much 5-8%. But, and this is my analysis, it is much weaker after redistricting, for first term representatives and the ideologically extreme (for their district). Republicans, due to the 201o wave, have a triple whammy working against their incumbency advantage.

    • Sam Wang

      Note that I was referring to effective advantage combined across all 435 districts. The question is bit different in that context. Not how much it’s worth to the individual’s November vote share expectations, but how much it affects the intercept on my graph. If I understand you, Gelman addressed the first question.

      For redistricting, same story: effect on the 435’s seat share for a given vote share, not on the 1’s future prospects.

      These concepts are clearly related. If you have light to shed, I’d be quite interested.

  • John O

    I just wanted to express my thanks for this website and the work that goes into it.

    I am not educated in statistics (or indeed math as a whole) and I read this site for the same reasons I read statistics-oriented baseball and football sites – I enjoy the challenge of deciphering what the hell you are talking about.

    It is one of the few sites where I feel I am being educated – not in facts and information, but in something more important: how to think.

    I also appreciate the comments section, which I often search for clues to the puzzle that is (to me) the main post.

    Though I don’t always (understatement) succeed in understanding, I really, really love trying.



    • Sam Wang

      John, I am so glad that you wrote in. It is inspiring to know that the site is appreciated by a non-mathematical person.

      My view is that these problems – and statistics in general – are not all that hard to think about, if one can cut through the small issues to see the big issues. I like simpler rather than more complex explanations. Sometimes I go down a wrong path. The goal is to make it all clear to myself – and along the way, you. Thank you for reading.

  • Amitabh Lath

    Hi Sam, I saw on Andrew Sullivan’s blog that Cook has joined Weigel in the doubting the House numbers.

    This is turning into classic moneyball. The numbers guys vs. the go-with-your-gut guys.

    By the way, I am amazed that you have a numerical estimate of the love my congressman effect.

    “…I did a simple fit to estimate the size of the effect. It appeared to be equivalent to 1.3% for the Republicans. ”

    Got plots? Need errors.

    • Sam Wang

      I did a multiple regression of

      seatmargin = a0 + a1*votemargin + a2*previousseatmargin. The parameter a2 is positive, then calculate the advantage of incumbency as the margin for which there is a near-tie in seats. That margin is -(a0+a2*previousseatmargin_2010)/a1.

      For 1996-2010, the advantage is D+1.3 +/- 2.4% (not +/-1.0%, as I reported before; oops).
      For 1946-2010, the advantage is D+0.6 +/- 1.0%.

      These are moving suspiciously towards zero, aren’t they? See my discussion with William Ockham. One cannot rule out the possibility that in the aggregate, there is no measurable incumbency advantage.

      Note that this is a simpler form of the “four-factor model” run by Nate Silver. I tried that too but the errors exploded. After some tinkering, I decided it was inherent to the whole dirty business.

      The worksheet is here. Please excuse they kludgey error calculation – as well as the use of Excel. I may have to turn in my physics merit badge.

  • William Ockham


    I can point you to this paper

    with two important caveats. First, the model used won’t apply this year because it depends on using the results of the last election in each district and you can’t do that in the first election after a redistricting (years ending in 2 plus other random years if you live in Texas like I do, yes I’m still upset over Tom DeLay’s rogue redistricting).

    Also, they look at the average vote share per district, rather than the national vote totals. They state (just prior to the 2006 election) that “the Democrats have averaged over 50% of the vote in Congressional races in every year except 2002”, which obviously doesn’t match up with your graph. Later on the paper that gets explained.

    Now that I’ve looked over that paper again, it has a really interesting result. It says that even if you figure incumbency is worth 8% in an individual race (which is the estimate they use in their formula), overall it does the party no good at all. Because if you when 50% of the national vote (the way you and normal non-poli sci people define it) you have 50% chance of winning control of the house. If you get 53% of the national vote, you are almost certain to gain control.

    • Sam Wang

      Thank you. I’ll enjoy that.

      The macro effects of incumbency ought to be baked into the seats vs. margin plot. That’s how I got the 1.3% estimate.

      But redistricting…this is starting to bother me. Short of detailed simulation (really not my cup of mead), feels like some formula must be derivable. The deviations on that graph put upper bounds on the effect, but do not help in determining how asymmetric it can be.

    • Sam Wang

      Regarding redistricting, before 2010, Democrats had a 2:1 advantage in seat-redistricting over four cycles. Yet the seats v. margin graph still goes through the origin, more or less. This paper by Klarner discusses it in the context of 2010 redistricting. So how great could it be?

  • Rob Manis

    Just one note about your donation recommendations. Actual candidates buy at a discount but groups like the DCCC pay market rates. It would be worthwhile to generate a list of toss up districts to maximize donation dollars.

  • Ragout

    The biggest problem with Wang’s 74% forecast is that it isn’t based on polls of likely voters. Instead, Wang estimates a D+4 generic ballot based on a mixture of polls of likely voters, registered voters, and even polls of all adults! Polls of registered voters tend to be several points more Democratic than polls of likely voters, and polls of all adults are even more biased.

    If the generic ballot estimate were limited to polls of likely voters, it would be about Dem+0.

    • Sam Wang

      This is false. In LV polls since the DNC, the margin is D+1.5+/-1.4% (median +/- estimated SEM, n=10). All polls give D+3.0+/-1.5% (n=13).

      Focusing on LV is not that strong a refuge. In the Presidential race, after Labor Day the EV estimator does not hiccup, as it would if the RV/LV thing mattered that much. Also, the RealClearPolitics generic Congressional average lands on the final outcome even when all polls are accepted.

      While we are on the subject, I also omitted the fact that generic polls tend to drift toward the President’s party in September and October. This would have favored the Democrats-win hypothesis, but I left it out. I thought that would have been gilding the lily.

    • Ragout

      You say the Generic Congressional average lands on the final outcome even when all polls are accepted. But I think this is because pollsters switch to LV polls close to election day, not because RV and Adult polls are unbiased. I clicked on the two most recent links to final generic ballot polls in your article, and they both showed 100% LV polls.

  • Patrick Draut

    The aspect that I appreciate most about this site is Dr. Wang’s involvement in the message boards. It improves the user experience and shows Dr. Wang’s ability to listen to valid POV’s.

  • Seth


    I am fascinated by your analysis, but I think there are analytical problems comparing the electoral college and its lumpiness to control of the house, and its lumpiness.

    Unlike states, which may favor one party or the other but are not artificially constructed to do so, Congressional districts are gerrymandered specifically to maximize the electoral benefit to one party. Thus, the likelihood that such lumpiness would affect control of the House must be higher than the likelihood that random lumpiness would affect the outcome of the Presidential race.

    I know you dismiss the anecdotal critique, but I think you underestimate the possible effects of gerrymandering. For example (these numbers are totally hypothetical), if a state has 8 congressional districts and is evenly divided, 2 million Democrats to 2 million Republicans, it would be possible to create 2 districts that 100% Democrat, and 6 districts that were 66% Republican and 33% Democrat. In such a circumstance, even though the state would be evenly divided, there would e almost no way for a Democrat to win in 3/4 of its districts, even with massive voter swings. Though the Republicans are not that efficient, I don’t think you can just dismiss gerrymandering out of hand on the theory that 55% majority districts might be in play.

    • Sam Wang

      The question is how to render this point quantitatively. You are the Nth commenter to make this qualitiative point!

    • seth

      Your rejoinder to gerrymandering relies on a series of assumptions that may or may not be true, but for which you provide no evidence.

      1) You argue that because Democrats previously enjoyed a structural advantage, but did not reap a significant outperforming of their popular vote, gerrymandering must not have been effective. However this assumes that democrats gerrymandered so as to maximize their house numbers. Is this true historically? It may be, but my impression was always that Southern Democrats at least, gerrymandered so as to prevent minority representation as much as possible, which would have had the opposite effect.

      2) You suggest that all Congressional races with 5% could flip with a Democratic surge of 9%. But isn’t a surge of 9% already priced into the lean Red or lean Blue determinations?

      I fully admit that I could be wrong, but where is the clear historical example where a party controlled state houses, explicitly tried to gerrymander so as to maximize Congressional representation, and failed to significantly outperform its national poll numbers.

  • Steve l

    I think if you look district by district there are many races that shouldn’t be tight, ie.Mary Bono, Steve King, even Michelle Bachman. I think we need to look at the tight races on the R side that shouldn’t be so tight.This may back up Sam’s theory. We are also dealing with an unusual dynamic in Mitt Romney. We’ve never had a candidate at the top of the ticket that had a disdain for a large section of the voters. We’ve also never had a major party candidate at the top of the ticket who has been as undisciplined as a politician. I think this is having an effect that may may this house control closer than we expect. Mitt Romney has created variables that haven’t been in play before.

Leave a Reply to Michael S (Cancel)