Tuesday, October 12, 2010 at 5:31 pm by Darryl
A new Elway poll has just been released for Washington state. Included in the poll is a head-to-head match-up in the Washington state senatorial race between Sen. Patty Murray and real estate salesman Dino Rossi. The poll was conducted using live interviews between October 7 and October 11 on a sample of 450 likely voters. The margin of error is 4.6%.
The results have Murray leading Rossi by 51% to 38% with “leaners” excluded. Murray does even better (55% to 40%) if “leaners” are included. I’ll stick with the former numbers in what follows. Here is the polling in this race to date (excluding polls released by the campaigns):
This new poll reverses a not-so-good trend for Murray over the past month. Her lead had the appearance of slipping away (although really the recent polls have all been a statistical tie).
You may also notice that all four of the Elway polls are more favorable for Murray than nearly all other polls. As Goldy mentions, Elway goes to some effort to explain this discrepancy. Elway goes to some effort to explain this discrepancy. Elway shows that there are distinctly different trends for polls done using automated questions compared to live interviews. The robopolls showed Rossi with a slight lead early in the year with the trend lines converging to a tie right now. The live polls showed Rossi and Murray tied early on, with Rossi staying relatively flat and Murray pulling away by about 7 points. No explanation is offered, but the implication is that Elway’s live polls are more accurate.
We can, to some extent, evaluate Elway’s hypothesis. Here is a graph I published on Nov 3, 2008, when Dino Rossi was running against Gov. Christine Gregoire in the Washington state Gubernatorial race:
Here again, Elway seemed to favor Rossi’s opponent relative to other polls, suggesting Murray would win 51% to 39%. The Washington Poll, another live interview poll, was also on the high side relative to other polls earlier (51.4% to 45%) and close later (50% to 48%). In the last month most pollsters had the race within two points: Strategic Vision (50% to 48% and 49% to 47% earlier), Rasmussen (50% to 48%), and SurveyUSA (50% to 48% and 48% to 47% earlier). The day after this graph was made, SurveyUSA released their final poll (52% to 46%).
The final tally in the Gregoire–Rossi race was 53.2% to 46.8% (+6.4). SurveyUSA’s last poll nailed it (+6), and the Washington Poll got the spread right (+6) in their earlier poll. Elway underestimated both candidates’ percentages and had a high spread (+12). This suggests that most of the recent polls underestimate Murray’s performance, even if the Elway poll probably overestimates it.
Back to the Elway poll…As usual, I’ll analyze it using simulated elections of 450 likely voters voting at the observed percentages in order to make a probability statement about who would win in a hypothetical election held now. From a million such simulated elections, we find that Murray wins 979,930 times and a Senator Rossi happens 18,537 times. In other words, this poll gives Murray 98.1% chance of winning an election right now. Rossi’s probability is 1.9%. Here is the distribution of outcomes from the simulated elections:
Compare today’s picture to that from last Friday’s Rasmussen poll. In fact, since the Rasmussen poll was conducted on October 6 on a sample of 750 likely voters. It showed Rossi leading Murray 49% to 46%. Given that the Rasmussen poll was taken the day before the Elway poll started, it seems reasonable to combine ‘em.
The combined Elway-Rasmussen poll sampled 1,200 likely voters of which 1113 chose one or the other candidate. Murray got 574 votes (47.8%), and Rossi got 538 (44.8%). A Monte Carlo analysis using a million simulated elections gave Murray 772,311 wins to Rossi’s 221,338 wins. In other words, these two polls combined suggests that Murray would have a 77.7% of beating Rossi if the election was held now.
To summarize, in combining the only two October polls to date—one done with live calls one by robocalls—Murray still comes out ahead, albeit not with a statistically significant lead.
(Cross posted at Horses Ass.)