New 2010 projections: Dems lose House by 12

crossposted at StochasticDemocracy and DailyKos

These are the Labor Day election forecasts of Stochastic Democracy, in collaboration with Professor Wang of the Princeton Election Consortium. We  have launched a new Election Forecasting System for House, Senate and Gubernatorial Elections and for tracking the Generic Ballot.

But before we discuss the forecasts in more detail, let me first get out of the way who we are and why you should listen to what we’re predicting:

Who we are (quoted from the FAQ at our site):

The team consists of David Shor, Harry Enten, and Rasmus Pianowski. David is a Math student currently visiting Princeton University as a Visiting Graduate Student. Harry is an undergraduate at Dartmouth and an intern at Rasmus (that’s me) is a freshman at University of Hamburg, he has done political consulting and Media Outreach work for Montana congressional candidate Tyler Gernant.

The site is closely affiliated to Professor Wang’s Princeton Election Consortium.

Why you should listen to us

We have an outstanding track record.

In 2008, we correctly predicted the results of 49 of the 50 states in the Presidential Election, missing only Indiana (where we predicted that Obama had a 48% chance to win). We correctly predicted every single Senate and Gubernatorial election, and were off on the national popular vote for President by only 0.08%..

We also predicted that Obama would get 364 Electoral Votes, he ended up with 365 Electoral Votes.

In 2009, we correctly predicted the outcome of the 2009 Israeli Knesset election as well as 4 of the 5 notoriously hard to predict 2009 off-year elections- got only 2 correct, didn’t even put up predictions.

And we have a solid forecasting methodology that combines advanced statistical techniques with a huge polling database- and we do account for House Effects, so that you can be sure that our ratings aren’t swayed by Republican-leaning Rasmussen Reports polls too much.

Now on to the Forecasts.

Let me get out the bad news, and there is a lot of bad news, quickly:

The GOP is favored to take over the House, several Senate seats and Governorships.


For a complete list of election results, please visit Stochastic Democracy, here you’ll only find an analysis of several important races, maps and tables with results for select races.




As you can see, things look rather badly for the Democrats on the Senate front. While Republicans seem like relatively heavy underdogs to get a majority in the Senate, they at least will pick up several seats.

Arkansas, North Dakota, Indiana and Delaware are gone. With deficits of more than 15% in even the closest of these four races, it doesn’t make much sense to keep fighting except for helping down-ballot races.

In Pennsylvania, Joe Sestak has a fighting chance to mount yet another comeback, but so far his campaign hasn’t really taken off and with less than two months to go until the election he’s down against Pat Toomey by a bit less than 6%.

Colorado is close right now, even though the Republican candidate Ken Buck is ahead by a bit more than three points right now. The infighting among Colorado Conservatives in the Gubernatorial race might help Bennet to catch up.

In Florida the race is all but officially between incumbent Governor Charlie Crist, the Republican-turned Independent, and Tea Party favorite, Marco Rubio. The Democratic candidate Kendrick Meek has just about a 1 in 200 chance to win, while Rubio maintains a 4.5 point-advantage over Crist. The race is so unstable though that Crist can easily come back- or collapse.

In all other races, the incumbent party is currently favored (more or less) to retain their seats. For the Republicans, that means that Rob Portman (OH), Roy Blunt (MO), Rand Paul (KY), Richard Burr (NC) and the winner of the New Hampshire GOP primary (probably Kelly Ayotte) will more likely than not win.

The likely Democratic winners include Harry Reid (NV), Alexi Gianoullias (IL), Barbara Boxer (CA), Russ Feingold (WI)and Joe Manchin (WV), who is so heavily favored to retain the late Senator Byrd’s seat that this race doesn’t show up in the ‘most likely pickups’ table. On the other hand, Gianoullias, Reid and Feingold are all in races that could still go either way, even though they’re favored over their respective opponents right now.

Gubernatorial races



In the gubernatorial races, there are a few more highlights for Democrats, even though the bottom line looks rather bleak for Democrats here as well.

We are almost sure to lose six Governorships to the Republicans: In Kansas, Michigan, Iowa, Tennessee, Pennsylvania and Illinois. In every of those races except for Illinois, the Republican candidate is ahead by double digits, and most likely less than two months won’t be enough to close that gap.

On the bright side, we’re also almost certain to pick up the Governorships of Hawaii, Minnesota and Connecticut.

Maine is a race that doesn’t qualify as a ‘sure loss’ yet, but it doesn’t look good for Democrats, as Republican Paul LePage is leading Democrat Libby Mitchell by more than 8 points.

In the close battleground races it currently looks like Republican John Kasich is going to unseat incumbent Governor Ted Strickland of Ohio, Kasich is currently ahead by 4 points. The same could be said of New Mexico‘s gubernatorial race, where Republican Susana Martinez is favored to beat the Democratic Lieutenant Governor Diane Denish. Both of these races are close enough for Strickland and Denish to mount a comeback though.

Meanwhile, Democrats look like slight favorites in the gubernatorial elections of Florida, where Democrat Alex Sink leads the scandal-ridden Republican nominee Rick Scott, and in Rhode Island, where Democrat Frank Caprio is a slight favorite to win the governorship, edging former independent U.S. Senator Lincoln Chaffee. The Republican candidate is far behind.

Meanwhile, the true Toss-ups right now are in California (Brown vs. Whitman), Wisconsin (Barrett vs. Walker), and Oregon (Kitzhaber vs. Dudley). None of these races have a clear favorite right now, even though Barrett, Brown and Dudley would be slight favorites if the election was held today.

Democratic Governor Martin O’Malley (MD) and Republican Governor Rick Perry (TX)are both moderately favored to win their re-election bids against strong challengers, respectively former Governor Bob Ehrlich and Houston Mayor Bill White.

Incumbent party-candidates in New Hampshire (Lynch), Georgia (Deal), Massachusetts (Patrick) and in all races that aren’t listed here are favored to win their elections.

U.S. House



I don’t have enough space here to talk about all the interesting races as in the Senate and Governor sections, so let me just highlight the results of a couple of races that I know are dear to the netroots and of some especially close races:

In Florida-08 (Grayson-D), Alan Grayson is projected to lose to his challenger, Dan Webster, by 7.5%. He still has about a 25% chance to win reelection. Keep in mind that our model does not directly look at fundraising (it does look at Cook ratings though, and Cook does include fundraising in his ratings)- and Grayson has a gigantic warchest.

In Illinois-10 (open-R) Dan Seals is a slight favorite to finally become a U.S. Representative after twice unsuccessfully running against Mark Kirk.

In New York-20 (Murphy-D), Scott Murphy, who was elected with a strong Kossack phonebanking drive in the 2009 special election to replace Senator Gillibrand, looks like a slight favorite to win reelection. Bill Owens in NY-23 is a slight underdog though.

In Virginia-5 (Periello-D), red-district Progressive Tom Periello will almost certainly lose reelection.

In Idaho-1 (Minnick-D), the probably most conservative House Democrat Walt Minnick is projected to be a slight underdog in his reelection bid. You shouldn’t take this number too seriously though: This race is extreme. Minnick was, and the model can’t incorporate this, endorsed by the Tea Party Express and the Chamber of Commerce and praised by the Club for Growth, and the NRCC has already pulled resources from this district in the light of polls that show Minnick far ahead of his Republican opponent. Expect this number to shift as more polls come in, but right now our model isn’t convinced of Minnick’s staying power, as McCain won the district by a 61-35 margin in the 2008 Presidential Election while Minnick just barely beat a disliked Republican incumbent who won only 60% of his primary vote after winning a six-way GOP primary with 26% of the vote in 2006.

Races on knife’s edge: Here you can really make a difference

Julie Lassa (D) vs. Sean Duffy (R) (WI-07, Obey retiring)

Rep. Gerry Connoly (D) vs. Keith Firmian (R) (VA-11)

Rep. Phil Hare (D) vs. Robby Schilling (R) (IL-17)

Colleen Hanabusa (D) (likely) vs. Rep. Charles Djou (R) (HI-01)

Rep. Gabrielle Giffords (D) vs. Jesse Kelly (R) (AZ-08)

All of these 5 races are so close that every additional volunteer or donor might tip the race.


Again, the bottom line isn’t pretty: The Democrats are likely to lose the House, several Senate seats and more Governorships than they will pick up.

Talking about continuing the gains that Democrats made in 2006 and 2008 is irrational now- what we can all do is pick a few campaigns and invest a lot of our time and put our best efforts into limiting our losses.

What the DCCC does– cutting incumbents that can’t win loose, might be a good strategy for private volunteers and donors as well.

97 thoughts on “New 2010 projections: Dems lose House by 12”

  1. What kind of formula did you use?

    Why is it 100% likely we lose AR-02 and only 98% likely we lose LA-03?

    Something tells me you don’t take into account the personal popularity of certain candidates like Minnick. But I’m certainly glad to see that Bobby Bright isn’t here.

  2. I got 48 states right, nailed the Senate and got the number of Democratic gains almost spot on. I see a net loss of 32 House seats right now. Senate and gubernatorials in the diary below.

  3. The only published polls were by Republican pollsters. The most recent one was way off on the demographics of the likely voter pool and was in the field before a string of bad news for the Republican candidate, Brad Zaun. Local news radio and every local tv station carried the story about his ex-girlfriend having to call the police in 2001 to get him to back off. There are also stories about Zaun ignoring a court order to pay off a debt for years until just before he announced his run for Congress.

    Again, all of the polls showing Zaun ahead were taken before any of this news broke, and all made very favorable (for Republicans) assumptions about the makeup of the electorate.

    Meanwhile, there’s no sign yet that the NRCC will spend serious money for Zaun. They reserved air time in at least 40 districts, but not IA-03. Zaun is at a major financial disadvantage compared to Boswell.

    The race is a tossup. There is no way Zaun has an 84 percent chance of beating Boswell, as per your analysis.

  4. AL-05 can’t switch from D to R since it is already in Republican hands.

    -While I expect (hope) Mark Dayton to win the Minnesota Governor’s race I think a 98% chance is waaaaaay to optimistic.


  5. Where is the polling on KS-03? I have not seen any polls of the race. How do you justify predicting an 87% likely Republican win in this very moderate district? 87% suggests a high degree of certainty. What do you base it on?

    KS-03 includes the two most Democratic counties in Kansas (Wyandotte and Douglas) as well as a very moderate suburban county (Johnson County) that has been strongly trending away from the Republicans, especially since 2006. Obama won KS-03 in 2008. Democrat Dennis Moore has won re-election here very easily many times, proving that the district will elect and re-elect a Democrat that they are comfortable with.

  6. But I really have trouble giving your results that much credence when you give so many races 100% chances for one candidate.

  7. Are you saying there’s 64% chance that district will flip from D to R ???

    If so, I would find that somewhat hard to believe.  Critz had 52.6% of the vote in the special, while Burns had 45.1%.  Therefore, Burns % would have to go up by 3.8 or higher for him to win in Nov. [(52.6-45.1)/2] …

    looking back at historical special elections, there have been extremely few circumstances in contested/high-profile races where the same two candidates were running, where the loser gained 3.8 from the special to the general election … I looked back over 20 years and couldn’t find a single real race …

    a good example of this may be WI-1, where there was a special in May 1993.  Peter Barca (D) won over Mark Neumann (R) by 50-49 (675 votes).  In Nov. 1994 Neumann came back to beat Barca by 49.4-48.8 (1120 votes), so even though there was a party switch, the percentages didn’t really change that much …

    so, unless 2010 has a much bigger GOP wave than 1994, I would find it hard to fathom a 64% chance that Burns wins this …  

  8. This race definitely isn’t a 100% guaranteed picked up for Republicans.  The general election has barely started.  It will be a much closer election than people are predicting.    

  9. The Senate picks and Governor picks are about where I would put them right now, but the House I think is a bit pessimistic.  I think the major flaw of this analysis is twofold:

    1.There is an over-reliance on the PVI, especially in districts where there has been little polling thus far. (ie: PA-10, GA-8)  Tom Marino a 91% favorite over Chris Carney?  That doesn’t jive with any piece of objective knowledge I’ve seen this cycle outside of PA-10’s PVI.

    2.There doesn’t appear to be a large premium put on candidate quality. (ie: AZ-8, TX-17, NH-1).  I mean really, does anybody believe that there is a 98% chance of Chet Edwards losing?  I don’t.  Nor do I think TN-8 is a guaranteed pickup for Team Red, Roy Herron’s strength as a candidate is underrated here.  

    sidebar: Those that have disagreed over my take about the Dems losing lots of seats in the Southeast have to be pulling their hair out over these rankings.  Even I’ve never had Jim Marshall or Suzanne Kosmas losing.

  10. Interesting analysis and thanks for the work. While it sucks to see the results you have done your best to be unbiased and take any emotion out of the equation so it is what is. A few comments that I have.

    1) Yes it sucks but Tom Perriello is toast. I live in Northern Virginia and did canvassing for him this weekend and the amount of people that said “I voted for him in 2008 but no way this year” is hard to figure how he wins having only won by 800 votes in 2008.

    2) I have to agree with the poster above regarding IA-3. I don’t see how Leonard Boswell only has a 16% chance of retaining his seat. His opponent has had some pretty bad publicity lately and is very much underfunded. Boswell is 78 and will likely retire if he wins. The district could be chopped in redistricting.

    3) I realize that it is a nasty climate but Chet Edwards has only a 2% chance of surviving? This guy has been targeted time after time and yet still survives. Tom Delay targeted 5 people with his 2004 redistricting plan and all but he have been defeated. His opponent is not a top tier recruit either.

    4) I really don’t see Jesse Kelly as a viable candidate against Gabby Giffords. District is pretty moderate as it is primarily Tucson and Giffords fits the mold. Kelly is inexperienced and very much underfunded. Paton would have been a very tough challenger but thankfully Kelly helped out team blue.

    5) I agree with you about supporting Julie Lassa as that is a swing district that typically goes Dem that we will need to hold but I strongly disagree about Gerry Connoly and VA-11. I live nearby the district and this is one area that is very quickly moving from team red to team blue. Connoly is also well known in the district as the former chairman of the Fairfax County Supervisors and it is VERY expensive to get your message out given that it covers Washington DC’s TV market. Dems are having trouble in blue collar districts not in white collar districts like this. I really don’t see this one as anything less than a team blue 7 point win.

    6) Hard to do this work and everyone can complain about it but it is a baseline right now that can and will change (hopefully towards team blue).

  11. This isnt quite a done deal. Well it is for Quinn but not for Brady. Mostly because Quinn is so clueless that he makes the dems look awful. Illinoians wont vote for a Republicans because the last rep governor is currently incarcerated. I think its time for a change and that illinois voters will cast thier ballots for independent Scott Lee Cohen .

  12. You open your post by bragging about your “outstanding track record.”

    But you then proceed to confuse me entirely by posting two contradictory statements about your 2008 predictions.

    First, you say that  “In 2008, we correctly predicted the results of 49 of the 50 states in the Presidential Election, missing only Indiana (where we predicted that Obama had a 48% chance to win).”

    Then you go on to say “We also predicted that Obama would get 364 Electoral Votes, he ended up with 365 Electoral Votes.”

    Okay, that is a neat math trick, but which one did you actually predict? How did you get to 364 electoral votes if you got Indiana wrong and every other state right. Where did those extra ten electoral votes come from?

    But beyond that nit-picking (but annoying) point, my larger question goes to this: what were you predicting at the beginning of September 2008, not what were you predicting on election eve.

    If we want to judge your forecasting model and the relevance of the forecast you are making now, the fairest thing is to compare it the forecasts you were making at a comparable point in time.

    So, how exactly did your September 6, 2008 predictions stack up against the November 2008 election results? Was your track record so “outstanding” then?

    Not asking this to be combative, but simply to make sure we’re being upfront about the value and limitations of forecasts made fully 2 months before any election.  

  13. Did you adjust for the how recently it went Republican (this Summer) and why (3 way race with two Dems, so 40% was enough)?  Basically, you’re projecting a 10% swing to the Republicans in a matter of months, with candidates they already know, and just voted on.

  14. The American Future Fund, a right-wing 501(c)4 organization, just released a batch of Iowa polls (pdf file). In IA-03, they found the generic Congressional ballot at 40 R/39 D. They found Boswell leading Zaun 48 percent to 39 percent. The “certain to vote” head to head toplines were Boswell 47.3 percent, Zaun 41.3 percent. Their party breakdown (I assume this was for the likely voters was 42%D, 36%R, 20% independent or third party).

    Now, this is just a snapshot, and some key figures in the American Future Fund backed a different Republican (Jim Gibbons) in the primary against Zaun. So, I’m not saying Boswell is certainly ahead. But I strongly disagree with your contention that Zaun has an 84% chance of winning IA-03.

  15. http://fivethirtyeight.blogs.n

    Republicans have a two-in-three chance of claiming a majority of House seats in November, the FiveThirtyEight forecasting model estimates. And their gains could potentially rival or exceed those made in 1994, when they took a net of 54 seats from the Democrats.

    Nate’s a lot smarter than I am but I just don’t think things are that bad.

  16. There have been polls, absolutely dismal polls, in AR-2, while there haven’t been any in LA-03.

    As for personal popularity – We try to account for it by incorporating Cook Ratings into our regression. But in the past, popular candidates in very conservative districts still tend to go down. Look at Chet Edwards in this cycle. Still, there might be some room to explore interaction effects that could help the Democrats in uber-red districts and their ability to weather the storm so well.

    I do suspect that Minnick is probably ahead. But the problem is that the polls involved that show a Minnick lead have had absolutely terrible sample-sizes and have been by crappy relatively unknown pollsters. So the model gives more weight to the regression model, which sees a candidate that barely won in 2008 during a wave election.

    Even then, the model thinks that the race is pretty competitive, and as more polls come in, it should adjust.  

  17. As for formulas – These forecasts were the result of some fairly sophisticated hierarchical Bayesian modeling that incorporates district-level polls, regression analysis, and the generic ballot. I’d refer you to the FAQ on our site.  

  18. I’m sorry, but a SUSA or Ras poll isn’t a “real poll” at this point. What I’m noticing is that a lot of people on campaigns who I talk with are telling me their internals, and while they’re not great for Democrats, pretty consistently they’re better than the public robopolls.

    I think a lot of people basing their fancy projections on public polling aren’t giving much consideration to possible GIGO effects.

    BTW, I chuckled that you declared Delaware Senate “gone” and NH hopeless.  You know, there hasn’t been a primary yet in either state, and the RSCC has had to jump in to try to save Castle’s ass, and the last public poll I saw–which of course isn’t much to base things on, but probably has some curiosity value–actually showed LaMontagne moving on Ayotte.  

    Predicting the outcomes of elections without knowhing who the candidates will be, I’m sorry to be so blunt in saying, is silly.  

  19. between Kelly and Giffords, which is along the lines of what we’re projecting. Our regression also thinks that the race should be within a point.

    As for Minnick, yeah, that’s the race where we think that our projections are seriously off, for various reasons, like the really unusual ideology of Minnick (who could pass for a Republican even now without changing parties, and who got endorsed by the Tea Party and CoC) and the weak challenger he got in this Labrador guy.  

  20. and while I’d agree on some of the points you make (mainly concerning the lack of young voters and the question order, the other points don’t seem valid to me), none of them would likely affect the bottom line very much. I’m pretty sure that if you adjusted the crosstabs of the poll to match your ‘demands’ it would still point to a Zaun +5 or 6 outcome.

    Also, don’t forget that we include House effects adjustments in our ratings, and Ayres&McHenry is fairly prolific, allowing us to adjust for their bias pretty well.

    There have been two other polls as well, by another GOP pollster, ‘Victory’, showing Zaun up by 9 in June and by 7 in August.

    I don’t think that scandals concerning congressional challengers have much of an impact on races, because people don’t care much about him and aren’t likely to read or remember such stories. And even if they do, they might not remember his name until the next time it pops up in the news.  

    So…we’ll see, but I don’t think that a chance of 19% to retain his seat his selling Boswell short by a lot. I certainly wouldn’t characterize the race as a Toss-Up.  

  21. Thank you, it’s a bit hard to keep track of every number in every race. Every suggestion helps. We’ll fix it in our update tomorrow.  

  22. If you are going to brag on your super sophisticated computer model you had better get the basic inputs right.

  23. was 37%D, 37%R and 26% no-party, it seems fanciful to predict a 2010 electorate of 25%D, 31%R and 41% no-party. Iowa Democrats have an extensive early voting program, and several campaigns will be focusing on Polk County (two-thirds of the district). I’m not saying Republican voters couldn’t outnumber us in this kind of a year, but there is no way independents will turn out at a much higher rate than Dems in an off-year election. They also undersampled women in the most recent poll, and I know that several of the Iowa Democratic campaigns will have a particular focus on turning out women.

    The other issue is money. Zaun doesn’t have it now. I realize that some GOP challengers will win despite being outspent 5 or 6 to 1, but giving him an 80 percent chance to win in that case seems like a lot to me.

  24. The issue was with the auto-table generating code. There are a couple programs and a lot of files involved, so some errors bubble through.  

  25. I know 98% produces a bit of a sticker-shock. But realistically, say you have 35 candidates, each who have large robust leads in every poll by multiple pollsters. How many of them  do you expect to lose? Upset victories happen, but by definition not very frequently.

    There’s probably a bit of an over-estimate, since these sort of models probably slightly underestimate the occurrence of outliers. But still, we’re looking at a neighborhood of 95-99% here.  

  26. The general consensus has been that it’s going to be Boswell vs. Latham in Central Iowa in 2012. What do you think will happen if Zaun beats Boswell this year? Will they have a R vs. R battle in Central Iowa anyway, or will they throw Zaun in with Loebsack in a D-leaning district?

  27. bit more detail.

    Party registration: A lot of Republicans/Tea Partiers are identifying themselves as Independents, and so are a lot of progressive Democrats. That wasn’t the case yet in 2006. Politics have, at least nominally, moved away a bit from organized party politics to things like the Tea Party, which is nominally (and only nominally, I know) independent.

    Gender: You’re talking about a 50-50 vs. 52-48 split. That’s no big deal at all. The gender gap isn’t that big. Reweight their numbers to a 52-48 or even a 55-45 pro-Women split and Boswell doesn’t gain more than 1 or at most 2 points. (Party ID is a bit more relevant, but I don’t think that Boswell would gain more than 4-5 points out of the 10 point deficit with your numbers either, though I’m too busy to do the exact math right now.)

  28. how the districts will shake out. Three possible maps were analyzed in this Bleeding Heartland post.

    If Boswell beats Zaun I think he will have to retire in 2012. I don’t know any central Iowa Democrat who likes the idea of a Boswell/Latham matchup.

    If Zaun wins, I don’t think he could be thrown in with Loebsack. Hard to see Polk County ending up in the same district as Linn County (where Loebsack lives).

    I think that Latham will move to the redrawn IA-03 if Story County doesn’t end up there. Latham would much rather run against Boswell or a new Democrat, or in a GOP primary against Zaun. Most of the GOP establishment would support Latham against Zaun. (Latham not only has way more seniority, he is an old smoking buddy of Boehner.) Latham could get thrown into King’s district (the redrawn IA-04), but I doubt he would fancy a GOP primary against King.

  29. Iowa has a non-partisan redistricting process, and one of the criteria is compactness. You might be able to draw four equal population districts with Polk and Linn in the same district, but those certainly wouldn’t be compact districts. The Iowa City/Cedar Rapids corridor (Linn/Johnson) would have to be split up, and very few of the other counties around the Des Moines area would be in the same district as Polk.

  30. I think the only way you realistically have a chance to protect both Latham and Zaun in 2012 is to throw Latham in with King, and hope the new district’s GOP rank-and-file are more mainstream than the wingnuts who dominate western Iowa.  There are plenty of Iowa Republicans who would love to erase King in favor of Latham and Zaun because King is such an embarrassment, IF they have to settle for only 2 GOPers.  It’s going to be a nonpartisan map, so Iowa GOPers won’t have much if any influence, but they won’t make any noise if Zaun upsets Boswell and a first map gives them a good chance to get rid of King.

    Eastern Iowa has too much population for a nonpartisan map to throw Latham or Zaun in with Braley or Loebsack.

  31. …can do the same.

    The truth is this diarist’s and his colleague’s model could end up being exactly right on the number of seats and House control flipping, but way off on a couple dozen individual races–and that’s a LOT of races to misfire.

    But it’s a normal number at THIS stage, still a couple months out.  House races develop late, by late October we’ll be breathing sighs of relief over a few endangered incumbents, but simultaneously shitting bricks over some people we never thought were seriously endangered.

  32. and was able to come very close on the 2008 predictions.


    President: Only off on Indiana and NE-02. I practically nailed North Carolina!

    Senate: Only barely missed Minnesota.

    House: My projected 22-seat Democratic gain was close, though I wrongly predicted some seats would flip (AK-AL, CA-04, IL-10, LA-04, MN-03, MN-06, NJ-07, WA-08) and wrongly predicted some seats would hold (KS-02, LA-06, MD-01, PA-11, VA-02, VA-05).

  33. But even with the Tea Party throwing money into Delaware, the chances are that Castle loses are pretty slim (New Hampshire is a different story and I’m interested to see how that one goes).

  34. The trouble for this analisys is what is affected by the Rasmussen numbers. I would like to see the results excluding Rasmussen polls. Rasmussen numbers are enough for distort the analisis for all the groups. Include the effect of the generic ballots for the House is making the Rasmussen numbers affect too to the US House rank.

    Despite that, I agree with the end of the diary. They are some house races for fight especially hard, but I think they are more than five.

  35. but well, I’m in Germany, so it was 4 AM here or so when your comment posted. First, we merge the predictions of the various primary scenarios using Intrade numbers/gut feelings, so we do account for the possibility that the candidate in NH could be Lamontagne or Binnie. Still, it doesn’t look good for Hodes even if they are the candidates.

    We do include released internals in our projections, and account for House effects, so that, basically, if the internals were much more positive for Democrats, they’d drag the projections behind.

    A second point that’s not related to modeling is that I don’t know why I should trust Democratic internal polls over public polls (from neutral companies like PPP, who IS in fact also working as an internal pollster and therefore has the same quality standards) or, in fact, GOP internal polls.  

  36. That mean King vs (Boswell or Zaun). (I think Boswell will win).

    I remember look to Iowa some months ago and I get with the idea what was very easy send King out of his district and mix him with IA-03 or IA-04.

  37. I think GOP primary electorates are extraordinarily unpredictable right now, and certainly there’s a track record of the far more conservative candidate massively outperforming expectations, and in several races (KYSEN, AKSEN,etc) shocking the establishment. With an electorate as volatile as GOP primaries are right now, and with such a tiny group of people who are so expensive to advertise to as the DE GOP primary electorate, I’m wary of discounting almost any possibility.

    I wouldn’t have thought it was likely a month ago, and I’m not saying it will happen. But on almost every primary, I’ve expected the more conservative or wingnutty candidate to do better than was expected a few weeks out from the primary, and in most cases I’ve come closer to the result than the CW suggested.  

    I’m almost disappointed there won’t be any more GOP primaries after a week from tomorrow; they’ve been entertaining as hell.  

  38. should have been more exact, the 1993 special result was Barca 49.9 to Neumann’s 49.3 …

    also, re. the special elections, meant to say, didn’t see any circumstances over last 20 years, but would guess there were some over the previous 200 years, though rare ??

    (it’s past my bedtime) …

  39. I do not buy their PA-10 number.  If you favor they did not take fundraising into account, their numbers there make sense, but I am highly skeptical there is a 90 percent chance PA-10 flips.  I would rank it 5th on my list of Pennsylvania seats to flip and only flipping if the Democrats lose 50+ seats.

  40. Sestak/Specter were running then and no noteworthy Republican primary was on that day. Dems came out in droves for their respective candidates (both the Conservative Dems of which there are a lot, and the Progressives) and carried Critz along. In the General election it will be almost the other way ’round, with Republicans turning out and Democrats improving not much over their primary turnout.  

  41. there’s LA-06, where Don Cazayoux (D) beat Woody Jenkins (R) by about 3% in the special election in May 2008, only to lose to Bill Cassidy (R) in November by 8%, despite it being a good Democratic year.  And there was nothing scandalous about Cazayoux that would’ve accounted for that loss.  It was simply the makeup of the district.

    Though had Jenkins not been an extremely flawed candidate to begin with, Cazayoux probably never wins that special election in the first place.

    BTW, I had no idea Cazayoux is now one of the U.S. Attorneys for Louisiana.

  42. The GOP primaries have been pretty damn amusing, and I suspect that O’Donnell will probably do better than some are expecting (in that I think she gets into the 40s, maybe even getting within single digits of Castle).

    Having said that though, Delaware Republicans aren’t as conservative as Kentucky Republicans or Alaska Republicans, it’s sort of like the difference between Arkansas Democrats and Vermont Democrats, there should be enough moderates/pragmatic conservatives in Delaware that we should operate under the presumption that it will be Castle who wins the primary (though I am, of course, rooting for O’Donnell because that would take a nearly guaranteed senate seat for the Republicans and give it right back to us, which would be most helpful this cycle!)

  43. …when the data is quite possibly more compromised than in the past.

    Remember, there was that (in)famous poll that showed FDR getting destroyed in 1936. Considering that a hugely disproportionate number of polls that are contributing to their data are from only a few pollsters, and that most of the polls are robopolls, I think these models using relatively new ways of collecting poll data and comparing the new stuff from one method to the older data collected in a different manner is a recipe for some serious problems.  

  44. They’ve been at it the longest, which is why academic papers tend the use them more. That, and we have access to their proprietary database, so it’s a lot easier for us to do validation analysis.  

  45. not many Democrats will win open-seat elections in R+3 seats. Obama would certainly lose KS-03 if the election was redone today, and while Moore would have a higher chance of holding the seat if he was running for reelection, a lot of popular incumbents in reddish districts will this time be swept away by the tide- like Chet Edwards, for example. Now Moore isn’t even running, which makes the hold even harder for Democrats.  

  46. Our model is fine, we just had a couple of coding errors in the tables, where we listed WY and OK as Republican-held (which effects only the graphics and not the model though, they’re correct in the calculations).

    We have a 100% chance of losing Wyoming and 98% of losing Oklahoma.  

  47. Snyder leads Bernero 51-29. That’s a prohibitive lead, larger than those of the incumbent GOP governors running in Alaska or Idaho. And that’s neither an outlier nor coming from a very Republican-leaning pollster. What’s more, Bernero has never been within single-digits in ANY poll.  

  48. Can’t find the link now but it was persuasive. Basically the Special actually drove turnout for the primary since it was larger in the district than anywhere else in the state.

  49. I haven’t been following Hare closely but didn’t realize his seat was considered highly endangered.

  50. 50:1 would be too strong even if it were a rematch between well-known politicians; with an open seat and two relative newcomers, the chance of a candidate-generated surprise should probably exceed 2%, even without the possibility of people changing their minds once they learn about the candidates.

  51. but the general election hasn’t really started yet.  It usually picks up after labor day.  As that poll indicates, a large percentage of voters are still undecided (20%).  If you look at polls from this time in 2006, Granholm was able to go from being tied (or is some cases, small leads), to winning by 14%.  Bernero was also down by 20% in primary just 2-3 weeks before the vote.  With 20% undecided and 2 months until the election, a lot can happened.  Once Bernero starts criticizing Snyder for outsourcing, the poll numbers will get much closer.  

    The race is definitely lean Republican but it’s nowhere near safe Republican.    

  52. “60-40 and we’ve decided” would be prohibitive.

    The current results are more like “50-30 sort of leaning, but haven’t thought about it much.”

    Bernero is doing far better than he was at an equivalent time in the primary; the difference is that now he has an opponent that people have heard of and don’t already dislike.  On the other hand, they might well come to dislike Snyder once they learn about him; his general-election negatives really weren’t tested by the primary.  

    Anecdotally, I know of several Snyder primary voters (of widely varying political views) who were unhappy about how little they knew, and were at least considering voting for the Democrat in the general election.  The other Republican candidates were better known, and worth voting against.  Snyder still might be pretty good — but that is a “might be” rather than a confident assessment.

  53. quality more strongly after an update to the model that most likely comes in several days. So we might see some improvement on that front soon.  

  54. I agree with your two flaws about the analysis.

    However, when it comes to the other part, it seems that you commit the same two flaws, right? It seems most of, it not all, of your basis relies on polling.

    As far as the loss of Southern seats, it’s a mixed bag. You’ve got some mentioned that tend to be less favorable than these.

  55. But I’m not sure about Herron. Hasn’t wtndem pushed back on the idea that Herroon’s actually a strong candidate? And I believe Herron is his/her State Senator.

    Kosmas is losing. Adams is a solid candidate. There’s a lot of agita over the space program. Even Rothenberg moved her to Republican Favored (which I think is too far).

  56. I asked what polling data you used to come up with the 87% likelihood of the seat going Republican.

    It is just a guess based on the generic ballot? Is that scientific?

    I also disagree about KS-03 being a reddish district. At this point, I think it could go either way. The R+3 PVI is based on a comparison to how the district voted in the last two presidential elections. In KS-03, both Wyandotte and Douglas counties have been getting more Democratic since 2004 and Johnson County has become much more Democratic since then.

    I know it’s hard to look past the state label, see Kansas and you think conservative, but this district is very urbanized and is not like most of Kansas.

  57. …trust internal polls, but people who aren’t doing models but who are responsible for winning campaigns put a lot more trust in polls that are far more sophisticated and also cost a heck of a lot more money than a cheap-o robopoll for some local TV station.  

    So, to be more explicit: you guys are spending a whole lot of time doing your models based sketchy data. It’s like doing computer modeling of weather patterns or something based on 17th century maps of the US. The fact is, you don’t have access to good polling data, and you’re not going to have access to good polling data except some of the generic stuff on a national level, especially in a mid-term where you have far fewer polls than in a presidential year.  

  58. Use for what?

    I’m saying these models based on polling are severely compromised, because they’re not based on sound polling, and they won’t be, because they only have public polls, and most of those are crap, especially on the CD level.  

  59. Statistical models don’t look at state levels and think ‘conservative’, they just look at the data we feed them.

    No, there haven’t been any polls for KS-03, just like for about 350 other House races. Yet, we are able to do relatively accurate predictions by looking at the races that HAVE been polled, their characteristics, and work out from that how races with other characteristics will play out, using Hierarchical Bayesian regression- doesn’t have a Wikipedia page, but if you google it you’ll get to some good scholarly explanations if you understand statistics, and frankly, if you don’t understand statistics it’s not easily to explain.  

  60. This is basically a “bafumi-type” model that takes into account some district factors in a regression analysis and applies a national swing on top of it.

    Models of this type were used by Gelman, Bafumi, and others very successfully in 2006 and 2008, and they can account for about 94% of variation in house seats.

    Our innovation is that we’ve found a rigorous way to augment these regressions with polls. This should improve performance from an already high baseline.  

  61. R+3 for KS-03 and comparing it to polls of other R+3 districts, that doesn’t tell you much.

    You have to know which way the district is trending and you have to know the local dynamics such as high-profile state races that will affect turnout, factors that might not be present in other states and districts. KS-03 is trending Democratic and Kansas Democrats and independents are going to turnout in large numbers to vote against Sam Brownback and Kris Kobach.

    The R+3 number for KS-03, which is based partially on the 2004 election, is not a good measure of how the district is today. This is a district that has changed a lot since 2004.

    The first poll that comes out for this race, I think your model is going to swing wildly in another direction.

    I think you should wait until you actually have polling data before making a predictive model for a race, but that is just me.

  62. There are no polls, the election hasn’t happened, but you know that your predictions are “relatively accurate.”

    Well, thanks for responding. It’s helped me confirm my suspicions.  

  63. Candidate generated surprises and big movement is pretty rare, and we have a lot of data to estimate their frequency.  

  64. The regression indicates that Djou would lose, but there was a fairly solid poll that had him with a 20 point lead.  

  65. And that can help for the House, because Rasmussen gives only generic ballots and polls for ND and SD (if I’m not wrong). But I think as example what the alone reason for your ME-Gov rating is the influence of Rasmussen polls.

    My work with Rasmussen numbers tell me what the Rasmussen effect is not linear, is not the same for all the races. Rasmussen use not the same criterions for all the races, and that change the ranks. They want to change the ranks. I think the alone way for quit the Rasmussen effect of the mix is quit the results of Rasmussen polls of the mix. I have more logical results without Rasmussen polls in the mix.

  66. You talk about protect both, Latham and Zaun. I was thinking the way for protect Latham only.

    King is hard right of the hard right, but the republicans may think in protect Latham and King. Zaum has enough scandals for win not enough support. I think he will lose.

  67. where Des Moines is, and two-thirds of the current residents of IA-03 are, in with King’s new district. Pretty much everyone expects King to have a massive western-Iowa IA-04, while Polk County remains the center of IA-03.

    Boswell was actually thrown into King’s district in the 2001 redistricting, but he changed his residence from his southern Iowa farm to Des Moines.

  68. saved. Let me dig into the records. The numbers on our site are our final pre-election forecasts, just as for every prediction site, including

    Also, as for the Electoral Vote/states thing, models are probability based, so for the EV count we multiply the EV count of every state with the probability we projected Obama to have to win the state.  

  69. I feel the perception of Roy Herron’s strengths and abilities in this race are highly over-rated. This from a hard-core Democrat who had him as a State Senator for years. It would be interesting to see what the DCCC polling shows in TN-8, as this is a district that could easily be written off if Herron is not polling competitively. I support Herron, but I know him, and thus try to be realistic about him and this district.

  70. Cool — what is the frequency of candidate surprises?  And can you easily limit it to candidates who haven’t held any elective office before, or even run for one seriously?  (If this were between two state legislators, I would expect at least some skeletons to have already come out in earlier cycles — but that isn’t as likely with a first-time candidate.)

  71. make the least bit of sense that Boswell, Loebsack and Braley are all ahead by the same amount. We have the Dems ahead by 17 in IA-01 and 21 by IA-02, and there SHOULD be some discrepancy- the districts just aren’t the same politically.

    I don’t really believe either that in a D+1 district the electorate will be 6 points more Democratic than Republican. If that happens, Iowa is either REALLY different from the national climate, or the Democrats would have to lead in the Generic by at least 2-3 points. Which they don’t.  

  72. because in terms of PVI IA-03 should be more competitive than IA-02 and IA-01.

    Also, Ben Lange (the GOP candidate in IA-01) has the lowest name recognition by far–25 percent, compared to 60 percent for Miller-Meeks in IA-02 and 68 percent for Zaun in IA-03.

  73. Hard to know, since we don’t have any trendlines, but the various stories about Zaun did get pretty broad local media coverage, beginning on August 19. This poll was in the field August 31-September 3.  

  74. LA-02 – Only a 51% chance of a Dem takeover?

    SD-AL – 81% chance of a Rep takeover?

    IA-03 – A 63% chance of a Rep takeover?

    I don’t believe any of them.  

Comments are closed.