Final Pre-election View, Pt. 2: The House

Forget about your House of Cards/
And I'll Do Mine.

Radiohead, In/ Rainbows, Track #8

WARNING: WE WILL BE DRAWING INFORMATION FROM SEVERAL WONKY ELECTION ANALYSIS WEBSITES. I'll try to link to them at, or soon after, first mention.

For all of its gerrymandered faults, the House is the most democratic of Federal institutions--there is little brake on the whims or delusions of the majority. Then, too, despite the fact that a majority of House districts are set up for a guaranteed party victory (about the same number, some 160-170 for each party), this decade has shown that about 100 are legitimately contested ones, in purplish shades varying from scarlet to indigo. These districts need to be won, or defended, every two years.

Nancy Pelosi did much that many Americans did not want, but I can not fault for a moment her proven ability to run legislation through the House on a timely basis (props also to ex-Congressman Rahm Emanuel, who is leaving his job as President Obama's chief of staff to go back home to try to run Chicago). A Republican majority, no matter how small, should work just as well, so it is much to be feared and opposed. I use the word "work" very loosely--they will be able to pass the legislation they want through the House, though Senate dithering and Presidential vetoes will both await if they get too aggressive--and by "well", I mean "badly", in terms of the short- and long-term interests of Americans, the economy, and the rest of the world. But what I mean is that 218 seats gets them control of all the committees, the bills put on the floor, and the rules which will govern their passage.

So, it's pretty much all or nothing: the Republicans need to gain 39 seats (plus the two vacant ones which are pretty much conceded to them). Go through the following and see if you think they'll get there.

Meet The Wonks

Nate Silver of Five-thirty-eight.com, now housed on the New York Times' blogs, is tops among those who study politics with a serious quantitative discipline. Part of his appeal is his willingness to write honestly and openly, and his willingness to engage in give-and-take with the blogosphere. He is also alone among those I'll mention here who makes his forecasts based on Election Day and not If-Today-Were-Election-Day.

With regard to forecasting individual House races, Silver has taken all the data available and come up with projected Election Day vote percentages, which he discloses in his district-level map. He also has for each a measure of standard error in those estimates (which he doesn't disclose); that error will reduce (and the projected percentages will change, albeit gradually) as Election Day nears and more polling data comes in--he commented recently that only 116 of the 435 races had local polls. The combination of the projected vote and the standard error has allowed him to quote a "percentage chance of takeover" for each individual race.

Silver is thought to be a liberal Democrat, but he has been very clear in establishing a distance from Democratic partisan viewpoints, respecting the work of conservative pollster Rasmussen and others of his ilk, and taking a firm, early position that the Democrats are in serious trouble, particularly in their bid to hold onto the House. His current assessment is that the Republicans have a 67% chance of winning control of the House, his mean simulation result is Republicans 224.3 seats, Democrats 210.7, and he shows a nice, real-life normal distribution of probable Democratic seats (done from "10,000 simulations") that looks like it peaks around 210-212, but with a neat little secondary spike right at 217-218.

I have two beefs with Silver: one is his habit of referring to his model anthropomorphically, as though it were a person with opinions ("my model is skeptical about...") instead of a mathematical construct he made. The other is that the notion of speaking of elections probabilistically is problematic: it has sense in the idea of putting stakes in the ground for bettors to take one side or the other, but not in the sense that political outcomes are random events. They are deterministic in their nature; the uncertainty comes from the difficulty in knowing who will come out and vote, whether people tell the truth to pollsters, how strategic decisions in the last days of the campaign will fall out, and whether past patterns of behavior will accurately predict future behavior, none of which really are random events like cutting cards or rolling dice.

Next up: Congressional Quarterly is a journal which studies developments in Congress and reports on them to the public. A few years ago, they got more seriously involved with detailed study of political races, and their site for this is called CQ Politics. They have done their own district-by-district ratings, which are done in a more traditional, less numbers-driven style. Essentially, they start with a detailed knowledge of individual Congresspeople, their districts, and how politics work, then add data as the race goes along.

The ratings start out with a pretty strong incumbent bias, but gradually filter in threats to incumbents. The only critical data point in the analysis is the last one before Election Day. In this study, I'll be using the ratings from September 17, though they came out with some shifts--entirely predictable ones--on September 30.

Cook Political Report appears similar to me in its base and approach. It makes a lot of good background information available to its paid subscribers--not so much to us Internet interlopers, though I did get a good description of the Congressional district "PVI"--the Partisan Vote Index. A description of a district as "R + 6" means that in the last two Presidential elections (i.e., 2004 and 2008), the district averaged 6% more Republican than the national average (which they quote as 48.7%, ignoring 3rd-party votes).

Larry Sabato is head of a non-partisan political institute at the University of Virginia and a long-time political observer and commentator. His "Crystal Ball" seems to utilize the national generic poll as the primary input to project national House campaign results.* I faulted him for not lining up his race-by-race assessments with his overall prediction ("if the election were held today"--or anytime in the last month--he's predicting a gain of 47 seats for the Republicans), but that overall prediction lines up very well with Silver's moving-target predictions. Sabato's organization just changed ratings of about 20 races on September 30 (unsurprisingly, all in the direction of improved Republican chances), but he has yet to update that attractive, but defective, chart published a few weeks ago.

Cook and CQPolitics have current analyses which are deceptive: they are only calling for net pickups of 10-20 seats for the Republicans, similar to Sabato's seat-by-seat analysis. I don't believe those are anything like what their final forecasts will be, and it would be dangerous for anyone to take them seriously. You have to assign all the many races called "toss-up" to the Republicans to get to their realistic forecast, where I expect them to end up. As I said, shifting each column over (or just relabeling it--Sabato's chart had zero Democratic-held seats as "Likely Republican") makes their analyses comparable to Silver's.

All three of those--Cook, Sabato, and CQ--will go through the same tiresome process of gradually, and selectively, moving races from Leans Democratic to Toss-up, Toss-up to Lean Republican, Leans Republican to Likely Republican, etc. as the days to Election Day diminish. They expect that memories will be short and they will not be faulted for failing to identify many of the seats that will turn over until the last weeks--something that Silver has been willing to do quite boldly. They should converge on similar final charts of race predictions, probably also similar to Silver's as he shifts his, too.

The final site I would recommend to those who want to do some research on their own is electoral-vote.com, which puts faces, PVI's, and narrative together about most, but not all, of the 100 contested races.

I did a study using Silver's percentages, categorized in the style of the others, and testing the hypothesis that the others identify the same races with the same results, only shifted one category toward the status quo. It worked pretty well, but not perfectly--on average, it was the right adjustment, but some races were two columns off or even more, while some were in the same column.

My Turn

This leads me (finally) to my own analysis. Basically, I looked at five pieces of data for 100 races. The selection of the races came easily: there is some variation around the margins (the "Likely Dem" and "Likely Rep" columns of one analyst might have races considered "safe" by others), but in the key areas of likely takeovers and tossup or near-tossup races, with a few exceptions, they're talking about the same ones.

The five pieces of data were the Cook, CQ, and Sabato forecasts, shifted a column away from the status quo as I've tried to describe, Silver's turnover percentages, and the PVI, which I used simplistically: a seat with a Republican-leaning PVI of "R + 6" or higher, in other words with a 55% or higher average Republican vote in the last two elections, was considered for this purpose a seat mildly trending Republican. I sought the center of those five indicators. A consensus is indicated if three or more of them (as adjusted) agree; those where there was too much difference of opinion are highlighted separately.

Democratic-held seats:
Gone (predict 100% to lose):
IN-3 (vacant); TN-6 (Open-Gordon); AR-2 (vacant- Snyder); NY-29 (Vacant); KS-3 (Open-Moore) (5 total; the two vacant were mentioned above, so Republicans need net +41 with them included).

High likelihood of loss (80% to lose):
LA-3 (Open-Melancon); ND-AtLg. (Pomeroy); MD-1 (Kratovil); AR-1 (Open-Barry); TN-8 (Open-Tanner); OH-1 (Driehaus); IN-8 (Open-Ellsworth); CO-4 (Markey); NM-2 (Teague); OH-15 (Kilroy); WA-3 (Open-Baird). (11 seats--note how many were seats vacated, either because Democrats sensed loss and abandoned them, or, in several cases, because they chose to run for the Senate, generally unsuccessfully).

High vulnerability, but forecasts varying widely (lose 2 of 3):
MI-1 (Open-Stupak)--91% turnover chance from Silver; PVI R+3; Cook Leans Republican (LR); Sabato Tossup (TO) (but moving to LR now); CQ TO.
MS-1 (Childers)--84%; R+14; Cook, CQ TO; Sabato LR.
PA-7 (Open--Sestak)--76%; D+3; Cook LR; CQ, Sabato TO.

Medium vulnerability (24 seats--lose 60%):
MI-7 (Schaer); FL-24 (Kosmas); WV-1 (Mollohan); NH-2 (Open-Hodes); PA-11 (Kanjorski); NH-1 (Shea-Porter); SC-5 (Spratt); PA-3 (Dahlkemper); PA-10 (Carney); AZ-1 (Kirkpatrick); IL-14 (Foster); GA-8 (Marshall); PA-12 (Critz); FL-8 (Grayson); VA-2 (Nye); IN-9 (Hill); NV-3 (Titus); MO-4 (Skelton); NY-24 (Arcuri); PA-8 (Murphy); KY-6 (Chandler); AZ-8 (Giffords); TN-4 (Davis); FL-2 (Boyd).

Medium, but varying forecasts:
On the high side (lose 3 of 4)--
IL-11 (Halverson): 88%; R+1; CQ LR; Sabato, Cook TO.
VA-5 (Perriello): 88%; R+5; CQ, Sabato, Cook all TO (Sabato moving to LR).
OH-16 (Boccieri): 84%; R+4; CQ, Sabato, Cook all TO.
WI-7 (Open-Obey)): 83%; D+3; CQ, Sabato, Cook all TO.
SD-AtLg. (Herseth-Sadler): 82%; R+9; CQ, Sabato, Cook all TO.
TX-17 (Edwards): 68%; R+20 (!); CQ LR; Cook TO; Sabato TO (moving to LR).

On the low side (lose half):
AL-2 (Bright): 74%; R+16; Sabato TO; Cook, CQ LD.
AZ-5 (Mitchell): 66%; R+5; Sabato LD (moving to TO); Cook, CQ TO.
IA-3 (Boswell): 63%; D+1; CQ LD; Sabato, Cook TO.
CO-3 (Salazar): 61%; R+5; Cook TO; Sabato, CQ LD.
WI-8 (Kagen): 56%; R+2; Cook, CQ TO; Sabato Likely Democratic (here abbr. as PD, Probably Dem.)
NC-8 (Kissell): 50%; R+2; CQ LD; Sabato, Cook TO.
OH-18 (Space): 49%; R+8; CQ, Sabato LD; Cook TO.
ID-1 (Minnick): 47%; R+18; Cook, CQ LD; Sabato TO.

Lower vulnerability (lose one in four):
CA-11 (McNerney); TX-23 (Rodriguez); FL-22 (Klein); NY-19 (Hall); NC-11 (Shuler); NJ-3 (Adler); OR-5 (Schrader); MI-9 (Peters); IN-2 (Donnelly); CO-7 (Perlmutter); NY-23 (Owens); PA-4 (Altmire); MA-10 (Open--Delahunt); ; NY-20 (Murphy); NM-1 (Heinrich); WA-2 (Larsen); VA-11 (Connolly); GA-2 (Bishop); OH-13 (Sutton); VA-9 (Boucher); WV-3 (Rahall); NY-1 (Bishop); CT-5 (Murphy). (23 seats)

Low, but varying forecasts(low side--lose 10%):
IL-17 (Hare): 26%; D+3; Cook, CQ LD; Sabato Safe Dem (but moving to PD).
CA-18 (Cardoza) 0.5; D+4; Sabato, Cook PD; CQ LD.

Very low vulnerability, but still considered contested (all to be held)
CT-4 (Himes); CA-47 (Sanchez); NC-4 (Price); GA-12 (Barrow); AR-4 (Ross); KY-3 (Yarmuth). (6 seats)

Democratic losses counted, by category above: 5 + 8.8 + 2 + 14.4 + 4.5 + 4 +5.75 + 0.2 + 0 = 44.65 seats (of 88 considered)

Republican-held seats:
High likelihood of loss (lose 3 of 4):
DE-AtLg. (Open-Castle); HI-1 (Diou); IL-10 (Open-Kirk); LA-2 (Cao). (4 seats)
Low likelihood (lose 1 of 4):
CA-3 (Lungren); FL-25 (Open-Diaz Balart); PA-15 (Dent); PA-6 (Gerlach); WA-8 (Reichert). (5 seats)
Very low, still considered contested (hold all):
MN-6 (Bachman--unfortunately); OH-12 (Tiberi); CA-44 (Calvert)

That's it--basically there were only 12 Republican-held seats that any of these experts gave a significant chance of flipping. Not too surprising when one considers that basically all the vulnerable House seats were going Democratic in 2006 and 2008. My expectation is that the likelihood of loss of these seats will become greater for the high-likelihood ones, but lower for the low-likelihood ones. There are a couple of real weak seats for the Republicans, but the wave will probably carry the other Republicans to safety.

Republican losses: 3 + 1.25 = 4.25

Net: Republican gain of 40.4 seats, resulting in the following count (my official prediction):
Democrats--217.6 seats, Republicans 217.4 seats.

I did NOT force these results--it's what came out from these calculations, honestly, even blindly, considered. My conclusions are the following:
1) In this election, control of the House is likely to be an absolute cliff-hanger.
2) The experts will be reluctant to show a tie or near-tied House, because it suggests an inability to make a decision.
3) Strategic use of resources is critical--nothing can be wasted. If I were in the DCCC, I would pump money into the "medium" vulnerability category above, including the ones contested to the high and low sides. They make up almost half the Democratic losses, while the high-end category seats are probably not salvageable, and the low vulnerability ones are very numerous yet represent a smaller portion of projected losses. In terms of states, I see a lot of Pennsylvania, Florida, Arizona, and Virginia in those groups, and not as much Ohio as I would've thought.

For New Mexico residents: Note that NM-1 is considered to have a low vulnerability, NM-2 a high one, while our district, NM-3, is about as safe a Democratic seat as you will find anywhere. I agree with all these assessments.


*There is some data supporting use of the generic poll, though not as an unbiased indicator; there needs to be some adjustment of the raw vote percentages to get an accurate indication of expected results, and the generic results with unnamed House candidates is different from a national poll in which respondents are told the names of the candidates. Thanks again to Silver for both studying this, and for reporting on it.