Washington State plays a pivotal role in the battle for the Senate this year, since the Republicans’ path of least resistance runs right through it. Were Republicans to win the three contests that can be described as tossups — in Illinois, Nevada and West Virginia — and avoid surprises in other states like Colorado and Kentucky in which the polling is still fairly close, then a victory by Dino Rossi in Washington would give them their 51st senator.
But if Patty Murray, the Democratic incumbent wins, the Republicans’ task would be much tougher: they’d have to win a state like California, where the polling has run against them of late, or one like Connecticut, where there are probably too few undecided voters to give their nominee, Linda McMahon, the victory even under best-case assumptions. We’ve sometimes used the phrase tipping point state to describe a state that will make or break a party’s electoral fortunes, and it may be Washington in this case.
The inconvenient thing about Washington State, however, is that the polling there has been all over the map.
Indeed, this has been a problem for some time. In April, one survey showed Mr. Rossi ahead by 10 points, while another had Ms. Murray ahead by 17.
Things seemed to have improved last month, when several polls were in the field at the same time, and essentially all showed Ms. Murray with a small but meaningful lead.
Since then, however, the problem has gotten worse again. Rasmussen Reports has deployed four of its ubiquitous surveys in the state in the past three weeks, each showing the race essentially tied. And a Republican pollster, working for the conservative-leaning American Action Forum, showed Mr. Rossi ahead by 6 points. But yesterday, a poll by a local nonpartisan polling firm, Elway Research, gave Ms. Murray a 15-point advantage.
Ignoring the American Action Forum poll for the time being — we don’t use explicitly partisan polls in our Senate or governor models and they often diverge significantly from nonpartisan ones — we still have a pretty big difference to split between the Rasmussen and Elway numbers.
The usual thing that people do in these situations is to engage in a “deep dive” on the demographics — “Rasmussen has only 36 percent Democrats in its sample when, like, everyone knows it should be 38 percent, dude!” Although I’ve engaged in such expeditions on occasion, most of these differences are caused by random statistical variations and are not meaningful. And the others usually result from legitimate differences of opinion about how to model the electorate.
So, let’s go “macro” here instead: why might Washington be a difficult state to poll?
One reason could be that it is one of two states, along with Oregon, where voting takes place almost entirely by mail. This can wreak havoc with traditional likely voter models, which often ask questions like, “Have you voted in the election precinct before?” and “Do you know where people in neighborhood go to vote?” — questions that are nonsensical in the context of an election that takes place by post. Also — probably because of mail balloting — turnout in Washington and Oregon has generally been very high, so targets that might work well in other states could fail there. Finally, since many voters in Washington return their ballots well in advance of Election Day, a pollster surveying the race close to Election Day will encounter another type of voter — those who claim to have voted already — which traditional likely voter models are not well designed to handle.
If this were to cause problems for pollsters, you might think the effects were felt predominantly among national polling firms, which might not adapt their assumptions to cater to the peculiarities of the state. And indeed, Rasmussen Reports has had some problems in Washington in the past, overestimating the performance of the Republican candidate by an average of 4 or 5 points since 2000:
SurveyUSA, another automated polling firm which has shown better results for Mr. Rossi this cycle than most other pollsters, also demonstrates such a pattern. Their final poll of the general election has overestimated the performance of the Republican candidate in each of the nine surveys in our database, and by 4 to 5 points on average:
So the two polling firms that have shown the best results for Mr. Rossi — Rasmussen and SurveyUSA — have a longstanding history of overstating Republicans’ numbers in Washington. Can Democrats rejoice?
Not entirely, because the local polling firm we mentioned, Elway Research, has tended to show a bias running the other way, with their surveys overestimating the performance of Democratic candidates by an average of 3 points:
In the absence of any other information, the thing you usually do when polls diverge is simply to average them and hope for the best. There is a special reason to do so in this case, since the polling firms that show the outlying results have some history of being biased in the same directions that they are now: Rasmussen and SurveyUSA, too favorable to Republicans; Elway, a bit optimistic for Democrats.
That would point toward Ms. Murray indeed holding a small lead, one that could be somewhat more meaningful than usual given that many ballots in Washington are sent in before Election Day and it is hard to make up ground late — although Mr. Rossi obviously retains decent chances.
Still, we haven’t quite addressed the question of why these biases might exist. If postal voting is the culprit, we might expect to see the same problems in Oregon, which conducts its voting in the same way.
The evidence here is inconclusive, however. SurveyUSA and Rasmussen have both shown a slight bias toward Republican candidates in Oregon, but there is not much data to look at, and the differences are not statistically significant.
A few other things about Washington State are worth considering. First, Washington, a technologically savvy state, has a somewhat higher than usual number of cellphone-only adults, which pollsters may not be capturing. It also has a fair number of Asian-American voters, who can sometimes be hard to reach because of language or cultural barriers. These things could make a difference at the margins.