I have worked in politics in Washington state in one capacity or another since 1982. I have been a campaign staffer, a candidate, a party chairman, an office-holder, and a consultant, and I have never seen polls as inconsistent as the polls we are seeing in the current Murray-Rossi Senate race. Some polls show Rossi slightly ahead, others show it dead even, and the Elway poll shows Senator Murray cruising to reelection.
The Washington polling confusion is unusual enough that it is drawing national attention. In releasing their latest results SurveyUSA included the following explanation: "There is unique and in some ways unprecedented disagreement among pollsters as to where this contest stands, and anyone trying to digest these SurveyUSA results needs to know that within the past 72 hours, Elway has the Democrat up 15 points and Fox News has the Republican up 1 point. While it is possible to see 16 points of pollster disagreement on ballot measures and in low-turnout primaries, it is unusual to see so much variation in a high-profile, statewide general election…. The New York Times describes Washington state as the Bermuda Triangle of polling."
What's going on? According to conventional wisdom the difference lies in how the polls are being conducted, and that Murray has an advantage in live interviewer polls versus automated polls. This is essentially the argument Stuart Elway offered for his outlier results. Mark Blumenthal wrote about this thesis last week: "The live-interviewer pollsters, CNN/Time and Elway, show Murray doing consistently better than the automated polls from Rasmussen Reports, Fox News and Rasmussen subsidiary Pulse Opinion Research and SurveyUSA. This week, for example, two new Rasmussen and Fox/Pulse polls showed a very different result, Rossi leading Murray by 3 and 1-point margins respectively."
There are two propblems with this argument. The first is not all polls using live interviewers are finding results consistent with the Elway and CNN polls. The recent Fabrizio poll used live interviewers, and it showed Rossi ahead, 48-42. In addition, I am aware of internal campaign polling done recently with live interviewers which yielded far different results than did the Elway and CNN poll.
Also, this discrepancy in live vs. automated polls is not showing up in other parts of the country. In the Wisconsin Senate race, for instance, the latest CNN poll is virtually identical to the latest Rasmussen poll.
A better explanation for all the variation, I believe, lies in how different pollsters screen for likely voters — not the method used to poll. Specifically it lies in the assumptions pollsters are making about the partisan makeup of the 2010 electorate.
To be accurate, a poll must be done on a statistically valid sample of the population. Your sample must have the right number of men and women, young and old, and the right geographic mix. As a poll is being conducted, pollsters will work with their callers to get this right mix, or they will weight the results after the poll is done. If you didn’t get enough responses from men, for instance, you can look at the responses from the men sampled, extrapolate, and adjust (or weight) the results accordingly.
This weighting is easy to do when you have known demographics, such as age. But what about trying to weight a sample to reflect party identification? Party identification shifts over time, and no one really knows how many Republicans and Democrats are going to vote in each election (the so-called enthusiasm factor). Quite often assumptions have to be made, and theses assumptions can dramatically affect results.
The question comes down to this: How large is the gap between Republicans and Democrats in Washington state in 2010? The Elway sample included 39 percent Democrats, 26 percent Republicans — a 13 percent gap for the Democrats. At a time when the national generic ballot shows Republicans with an unprecedented advantage, is there still a double-digit party ID advantage here for the Democrats?
One pollster I talked to said no. He said pollsters working for Republican and Democratic candidates are finding a Democratic advantage of between 3 and 5 percent. This seems credible given the findings of a Gallup poll done in July which showed Republicans closing the party ID gap all across the county. That poll showed only a 7 percent Democratic advantage in Washington state among all adults, so the gap is likely smaller among likely voters.
In 2006 and 2008, SurveyUSA’s final polls, which were quite accurate in predicting the outcome of races, found Democratic identification advantages of 7 and 11 percent, respectively. So to believe The Elway Poll one must believe that the 2010 electorate will include a greater Democratic advantage than existed in the 2006 and 2008 elections, both big years for the Democrats.
Now let's look at the most recent SurveyUSA and CNN polls with this party ID issue in mind. The most recent SurveyUSA poll showed Murray leading Rossi 50-47, but shows Rossi 14 percent ahead among Independents. Murray’s small lead is the result of a sample that included 36 percent Democrats, 27 percent Republican, plus 9 for the Democrats. How does SurveyUSA determine likely voters? Assumptions were made in weighting the sample: "Filtering: 900 adults with home telephones in the state of Washington were interviewed by SurveyUSA 10/11/10 through 10/14/10, using Random Digit Dial (RDD) sample from Survey Sampling Inc. Of the adults, 767 were registered to vote. Of the registered voters, 606 were determined by SurveyUSA to be likely to vote on or before election day, 11/02/10, 18 days from now."
If you make a different determination and assume only a 5 percent Democratic advantage you end up with both candidates deadlocked at 48 percent in the SurveyUSA poll.
CNN readily admits that they weighted their sample and made assumptions based on “past voting behavior”: "Washington is one of the few states CNN has polled in this fall in which Democrats and Republicans are equally enthusiastic about voting," says CNN Polling Director Keating Holland. "Combine that with a voting population that has trended Democratic in its past voting behavior and we see a state in which the likely voter model adds a couple of points to the Democratic candidate."
The CNN poll found Rossi ahead by 10 percent among Independents, but trailing overall by 8 percent. Assume only a 5 percent Democratic ID advantage and a comfortable Murray lead becomes a 48-45 dead heat.
Rasmussen also uses past voting history to weight its surveys: "Rasmussen Reports determines its partisan weighting targets through a dynamic weighting system that takes into account the state’s voting history, national trends, and recent polling in a particular state or geographic area."
Not all pollsters, however, use this methodology, and with different assumptions about likely voters, the race looks very different. According to the Washington Post’s political blog, “The Fix,” a recent Moore Information poll shows Rossi ahead: "The Moore Information poll, which was conducted for the National Republican Senatorial Committee and obtained by The Fix, shows Rossi taking 47 percent to Murray's 46 percent among likely voters. When only the most likely voters are sampled, Rossi leads Murray 50 percent to 45 percent."
In all of this, I am not accusing pollsters of being biased in favor of the Democrats. But polling is part art and part science, and different pollsters are clearly making different assumptions about Washington state. My guess is no better than anyone else’s, but I find it hard to believe that Democrats still enjoy a party ID advantage as large as we saw in 2006 and 2008.
This phenomenon is playing out in the Senate race, but it affects all partisan races on the Washington state ballot. We are either headed towards a Republican landslide that will oust a three-term U.S. Senator, flip two or three seats in Congress from Democrat to Republican, and give the GOP control of the legislature — or towards a fairly typical mid-term election in which the GOP make minor gains.
All year long we have wondered how big the Republican wave would actually be. It seems that in Washington state we won’t know until election night. If then.