No independent poll in 45 of 65 close House races

Of the 65 House seats our forecast rates as competitive, 45 have never had a poll from anyone outside the two parties' orbit. Twelve have had no poll at all. Another 33 have been polled only by campaigns, party committees or allied groups. Just 20 have had at least one survey from a pollster working for itself, a university or a news outlet.

The imbalance runs one way. Since mid-February, Democratic-side groups have released 118 House polls that reached our file and Republican-side groups 31. Three weeks before an election that will decide control of Congress, much of what the public knows about individual districts comes from the people trying to win them.

The same thinness reached our own Senate forecast. In Idaho, one unsponsored poll carried three-quarters of the weight in our average and moved a race we had at effectively zero to a coin flip, until we changed the method this week.

How we counted

We took every House district where our forecast gives either party between a 15 and 85 percent chance: 65 seats. For each we listed every general-election poll our model uses, back to mid-February. A poll counts as partisan when a campaign, a party committee or a group aligned with one side paid for it, or when the pollster works only for one party. Polls for newspapers, television stations and universities count as nonpartisan.

On that definition, 132 polls cover the 65 seats, and 22 of them are nonpartisan. A stricter test that also sets aside media-sponsored polls cuts the number of seats with an independent poll from 20 to 18. A looser band, 10 to 90 percent, gives 78 seats, of which 19 have no poll. Every cut tells the same story.

The polls are old as well as few. Only 26 of the 65 seats have had any poll since 1 September, and 13 have had a nonpartisan one.

The gap is sharpest in the closest races. The Cook Political Report listed 21 House toss-ups on 9 October. Thirteen of them have only partisan polls in our file, and California's 22nd has none. Wisconsin's 3rd, where Rebecca Cooke faces Derrick Van Orden and our forecast sits at 50 percent, has three polls, all from the Democratic side, the latest finished on 25 June. In Pennsylvania's 1st, four polls of Brian Fitzpatrick's race are all partisan.

Who pays for the polls

Of the 33 seats with only partisan polls, 23 have been polled only by Democratic-side groups, three only by Republican-side groups and seven by both. Across all House races, Democratic-side releases outnumber Republican-side ones 118 to 31.

That pattern flips above the House. In Senate races Republican-side releases lead 101 to 68, and in governor races 62 to 38. Which side publishes depends on the race, not only the party.

Released partisan polls lean to their sponsor, by about 2 to 3 points this year. Where a partisan poll and a nonpartisan poll of the same race were taken within 30 days of each other, Democratic-side polls ran 3.4 points more Democratic and Republican-side polls 2.1 points more Republican. That is 216 comparisons, almost all from Senate and governor races, since only 18 involve a House district.

The House data point the same way. Measured against our fundamentals for each district, Democratic-side House polls came in 3.4 points more Democratic, while nonpartisan House polls came in 0.2 points more Republican. Republican-side releases came in 5.7 points more Republican. The districts differ, so this is not a clean test, but it suggests the flood of Democratic releases paints a rosier picture for the party than independent pollsters do.

Our model already discounts partisan polls: it halves their weight and takes 2.5 points off their margin toward the sponsor, a figure measured from past final polls. In a seat with only partisan polls, the polls carried on average 34 percent of the weight against the fundamentals, against 40 percent where a nonpartisan poll exists. Where there is no poll, the fundamentals do all the work.

Idaho: our own single-poll problem

The sharpest case of thin polling is in our own Senate forecast. Todd Achilles, an independent and former Democratic state representative, is challenging Senator Jim Risch. Idaho voted 35 points more Republican than the nation in recent presidential elections, by our measure.

On 17 September our projection had Risch ahead by 25 points, with Achilles's chance effectively zero. The next day we added a poll by Advanced Targeting Research, taken 13 to 16 September among 700 registered voters, mostly by text message. It found Achilles on 50 percent and Risch on 39 in a two-way question. Our chance for Achilles went to 51 percent and stayed there for almost four weeks.

That one poll carried 75 percent of the weight in our Idaho average. The other five polls are campaign or partisan releases, which our method halves and adjusts, so the newest unsponsored poll dominated by design. With it, Achilles led our average by 2.7 points, and because our model let the polls count for 92 percent against the state's history, that average set the forecast.

The poll is a published, sourced survey with its full figures released, and it is not alone in showing a closer race than Idaho's history suggests. EastIdahoNews reported in August that a Bullfinch Group poll released by the Achilles campaign had him ahead 38 to 33 in a head-to-head. But polls released by the Risch campaign from Peak Insights put Risch ahead 52 to 18 in August, and Poliwave lists a further Peak Insights poll from 28 to 30 September with Risch ahead 49 to 24, which is not yet in our file. Silver Bulletin's average stood at Risch plus 18.9 points in mid-August, according to the race's Wikipedia page, and the Cook Political Report, Inside Elections and Sabato's Crystal Ball all rate the race safe or solid for Risch.

On 14 October we changed the method. Our forecast now counts the evidence in a race by the number of separate polling firms rather than the number of polls, so a race polled by one independent firm leans more on the state's voting record, and a registered-voter poll in a race with an independent challenger counts for less. In Idaho the polls now carry 65 percent of the forecast, and the Advanced Targeting poll 52 percent of the average, which moves from Achilles ahead by 2.7 to Risch ahead by 6.3. Achilles's chance falls from 51 percent to 14 percent, and we now rate the race Likely Republican. Idaho decides Senate control in 2 percent of our simulations, down from 8 percent.

Other races resting on one voice

Idaho is not the only statewide race where one firm or one side carries the average. In Louisiana's Senate race, one poll for the Democratic nominee Jamie Davis's campaign holds 89 percent of the weight. In South Dakota, a Zenith Research poll carries 65 percent. In South Carolina, three firms classed as working mainly for Republicans, Rasmussen Reports, InsiderAdvantage and Trafalgar Group, hold 86 percent of the weight across ten polls. Arkansas's governor's race has a single poll, from Hendrix College for Talk Business and Politics. Our chances for the challengers in those races range from 6 to 33 percent, and none hinges on one survey the way Idaho did.

What it means

The thin polling cuts both ways for our forecast of a Democratic House, which we put at 81%. Where polls are missing, our numbers rest on district voting history and the national environment. Where only partisan polls exist, we discount them. The risk is in the races where a single survey, unchallenged, does most of the work. For readers, the useful habit is to check how many polls stand behind any race figure and who paid for them. Our race pages list every poll we use and the weight each one gets.