A new Marist poll has been released in the race between Sen. Patty Murray (D-WA) and real estate wheeler-dealer Dino Rossi (R).

The poll of 638 likely voters (4.0% MOE) taken from 26-28 October has Murray leading Rossi 49% to 48%. Marist polls use live interviews and include a sub-sample of cell phone interviews. As we saw with the Marist poll from two weeks ago that had Murray up by +1%, the current poll is evidence against the otherwise strong trend of live-interview polls showing a modest (or better) lead for Murray.

I ran a Monte Carlo analysis that simulated a million elections using the observed preferences and sample size percentages observed in the new Marist poll. Murray won 567,112 elections and Rossi won 422,059 times. The poll (by itself) offers evidence that Murray would win with an election held over the past two weeks with a probability of 57.3%. Rossi would have a 42.7% probability of winning. Clearly, this result is a statistical tie, even if Murray’s odds are a bit better than Rossi’s.

With today’s poll, we have now had seven polls released that cover the past two weeks. Murray has led in five of the polls, Rossi has led in one, and one was a tie. A combined analysis of all seven polls provides a way to use all the recent evidence to assess this race. The resulting meta-poll had a total of 5,778 “votes” of which 2,797 go to Murray (48.4%) and 2,712 go to Rossi (46.9%). An additional 269 (4.7%) “votes” went to neither candidate.

The Monte Carlo analysis gives Murray 789,523 wins to Rossi’s 207,597 times. Thus these seven polls provide evidence that Murray would win an election held over the past two weeks with a probability of 79.2%. Rossi would win with a 20.8% probability.

With only three days to go until the Big Poll is tallied, it is worth a few minutes examining the recent trend in this race. Here is the collection of polls in September and October (I’ve excluded polls released by a candidate or party):

The very recent polling suggests a tightening race. Even so, Rossi has only led in three of 19 polls over the past two months, and all three were conducted by Rasmussen (and Murray has led in two other Rasmussen polls). Beyond that, Murray has led in all other polls conducted by seven other pollsters except for one of three SurveyUSA polls giving a tie.

The bottom line is that Murray looks like she will win this one, but there are some uncertainties that may be distorting the map between polling and actual voting. My hunch is that the uncertainties don’t help Rossi’s chances much and some favor Murray:

• Voter “turnout.” There really isn’t “turnout” in Washington, which has a very high proportion votes cast by mail-in ballots. Voter motivation should be less of an issue in this race than it will be in other states. You may have heard media reports about how a rainy day across the country on Tuesday will negatively affect Democratic candidates. While that may be true in states with poll voting, rain in Washington state can potentially give Democrats a boost by simply keeping inattentive voters in closer proximity to their unspent ballots.
• The “cell phone problem.” There wasn’t much evidence that this problem biased polls in 2008, but there is some more recent evidence to suggest it can now. Some pollsters, like Marist, include a cell phone sub-sample, but it is hard to do. If the phenomenon is real, Murray will do better than many polls suggest.
• The robopoll—live interview difference. I’ve been discussing this for weeks after Stuart Elway first raised it. The phenomena is observed nationally as well as in this race. Here, I suspect it will result in a 3%-4% boost for Murray over the robopolls, but nobody really knows what is causing the phenomenon. We’ll find out soon.

(And there is more good media stuff from the past week in politics at Hominid Views.)

As I mentioned here and here, we have been treated to two new polls today in the race between Sen. Patty Murray (D) and Dino Rossi (R).

A SurveyUSA was a robopoll taken on 678 likely voters (3.8 MOE) from 24-27 October and has Rossi and Murray tied up at 47% each. The KCTS/KPLU/Washington Poll poll used live interviews of 500 registered voters (4.3 MOE) taken from 18-28 October, and has Murray leading 49% to 45%.

In some sense, both of these polls are older than yesterday’s Rasmussen robopoll of 750 likely voters (4.0 MOE) taken completely on the 26th of October that has Rossi up +1% (48% to 47%). Therefore, the only reasonable thing to do is combine all three polls into one meta-poll and do a Monte Carlo analysis.

The combined 1,928 “votes” are split 916 for Murray (47.5%), 904 for Rossi (46.9%) and 108 (5.6%) for neither candidate. From a million simulated elections at these proportions and sizes, we learn that Murray wins 579,294 times and Rossi wins 414,495 times. In sum, these three most recent polls support a Murray victory with a 58.3% probability and a Rossi win with a 41.7% probability. This is very close to a tie (statistically, it is a tie), but Murray has a slight edge.

My usual near-election practice is to analyze all polls taken in the past two weeks. There are six such polls:

	Start	End			%	%	%
Poll	date	date	Size	MOE	Murray	Rossi	Diff
Rasmussen	26-Oct	26-Oct	750	4.0	47	48	R+1
SurveyUSA	24-Oct	27-Oct	678	3.8	47	47	0
WA Poll	18-Oct	28-Oct	500	4.3	49	45	D+4
Rasmussen	17-Oct	17-Oct	750	4.0	49	46	D+3
Marist	14-Oct	17-Oct	589	4.0	48	47	D+1
PPP	14-Oct	16-Oct	1873	2.3	49	47	D+2

The resulting meta-poll has a total of 5,140 “votes,” of which 2,484 go to Murray (48.3%), 2,406 go to Rossi (46.8%), and 250 go away. From a million simulated elections, Murray wins 785,190 times and Rossi wins 211,969 times. Thus, from the evidence found in polls taken over the past two weeks, we find that Murray has a 78.7% probability of winning and Rossi has a 21.3% probability of winning.

Statistically, the race is a tie because the probability of winning is under 95%, but the odds do favor Murray a little bit.

King 5 is reporting on a new SurveyUSA poll today in the race between Sen. Patty Murray (D) and perennial candidate Dino Rossi (R).

The poll has the race tied up at 47%. This is a robopoll of 678 likely voters with a margin of error of 3.8%. More analysis to come….

This morning KPLU reported on a new poll from KCTS/KPLU/Washington Poll in the race between Sen. Patty Murray (D) and real estate opportunist Dino Rossi (R). The poll has results presented in a couple of ways.

Among registered voters Murray leads Rossi 49% to 45%. However when a likely voter screen is used, Murray leads Rossi 51% to 45%. The survey was done using live interviews and has a margin of error of 4.0%. Full details on the poll will be available around noon. I’ll do a more in-depth analysis of the poll (and possibly another poll) later today.

We have fallen into a relatively dry spell for polling in the race between Sen. Patty Murray (D-WA) and real estate salesman Dino Rossi (R). Rasmussen has now released their latest Washington state poll showing Rossi leading Murray 48% to 47%. The poll of 750 likely voters was taken on Tuesday and has a margin of error of 4%.

This poll breaks Murray’s streak of eight consecutive polls with the lead. Before today’s release, the most recently taken poll was also by Rasmussen (17th Oct), and showed Murray with a +3% lead over Rossi (49% to 46%).

Using my usual Monte Carlo analysis employing 1,000,000 simulated elections, we find from the newest Rasmussen poll that Murray wins 415,950 elections to Rossi’s 573,763 wins. Thus, evidence from this poll alone suggests that, if the election had been held last Tuesday, Rossi would have a 58% probability of beating Murray to Murray’s 42% probability of beating Rossi. Given that the winning probability is under 95%, a statistician would tell us that the results suggest a statistical tie. Here is the distribution of simulated election outcomes:

A more comprehensive picture of the election emerges by combining all polls from the past two weeks. Besides the two Rasmussen polls already mentioned, we include the Marist poll taken from 14-17 October on 589 likely voters, giving Murray a +1% lead and a Public Policy Polling poll taken from 14-16 October on 1,873 likely voters and gave Murray a +2% lead. (Other polls in this race started the survey prior to the 14th of October so aren’t included.)

The combined meta-poll has 3,962 “votes” of which 1,920 go to Murray (48.5%), 1,862 go to Rossi (47.0%), and 180 (4.5%) just go away. In the Monte Carlo analysis, Murray wins 746,418 times and Rossi wins 249,788 times. In other words, the four polls suggest that Murray would win an election held over the past two weeks with a probability of 74.9%, and Rossi would win with a probability of 25.1%.

This collection of four recent polls gives the appearance that the race has tightened up a bit (compare this to the 98% win probability for Murray from pooled polls early last week). Keep in mind, however, that three of the four most recent polls are robopolls. As I discussed previously, there is a strong trend of robopolls showing Murray under performing relative to live-interview polls. Perhaps we will get a live-interview poll tomorrow….

It’s the Tuesday before the election. And that make for an excellent excuse to enjoy an evening of politics under the influence. Please join us at the Seattle chapter of Drinking Liberally at the Montlake Ale House, 2307 24th Avenue E. beginning at about 8:00 pm. Some of us will be there early for dinner.

Not in Seattle? There is a good chance you live near one of the 250 other chapters of Drinking Liberally.

On my way home this evening, I came across a pick-up truck parked on the street in Redmond, towing this trailer:

Without even thinking about, I started glancing around for evidence of a Dino Rossi campaign event….

(And there are almost 50 more links to media clips from the past week in politics at Hominid Views.)

This poll from Marist makes eight consecutive polls in which Sen. Patty Murray (D-WA) has the lead over Dino Rossi (R), even if her 48% to 47% lead is smallish.

I do a more detailed analysis of the poll, and an analysis using all nine October polls, at Hominid Views.

Man, after all this number crunching and writing…I needs me a drink!

It’s Tuesday following the ballot drop…making it a great opportunity to enjoy and evening of politics under the influence. Please join us at the Seattle chapter of Drinking Liberally at the Montlake Ale House, 2307 24th Avenue E. beginning at about 8:00 pm. Stop by earlier for dinner.

Not in Seattle? There is a good chance you live near one of the 249 other chapters of Drinking Liberally.

Washington state voters are, seemingly, under a polling blitzkrieg. Today another poll is released in the race between Sen. Patty Murray (D) and real estate salesman turned foreclosure opportunity seminar motivational speaker Dino Rossi (R).

This poll comes from Public Policy Polling (PPP) and shows Murray leading Rossi 49% to 47%. The poll was taken from 14th to the 16th of October on a sample of 1,873 likely Washington voters. The margin of error is 2.3%.

PPP makes an interesting observation about the race (emphasis added):

There’s a pretty strong argument that the Washington Senate race is the most stable in the country: PPP finds Patty Murray leading Dino Rossi 49-47, basically identical to our July poll of the race that found her ahead 49-46.

The reason for the stability is that voters know these candidates, they know what they think of them, and nothing they’ve heard during this campaign has changed those opinions in one direction or the other. In July Murray’s approval rating was 46/45. Now it’s 47/48. In July Rossi’s favorability was 43/48. Now it’s 44/49.

Compare this with yesterday’s Rasmussen lede (emphasis added):

Democratic incumbent Patty Murray is barely ahead of Republican challenger Dino Rossi now as the lead seesaws again in Washington’s neck-and-neck U.S. Senate race.

The fact is, Murray has now led in the last seven of the eight polls taken in October with margins from +1 to +13 and with an (unweighted) average of just over +5%. Take a look at the entire polling history:

Note to pollsters and reporters…that meme about how close and back-and-forth this race is? Yeah…that last’s month meme based on a tiny blip in the data.

Note: The graph in the first version of this post included a poll by DSCC. I’ve reposted the graph without that poll.

Update: At Hominid Views I do some statistical analysis of the poll and further examine the differences between robopolls and live-interview polls in this race.

(And there are links to some 50 more clips from the past week in politics at Hominid Views.)

I posted the results earlier today of the Washington Senate race based on KPLU’s news reports. At noon today, the KCTS/KPLU/Washington Poll top lines were released for public consumption.

In that all-important race between the single most powerful mom-in-tennis-shoe in the Senate & nation, Sen. Patty Murray (D-WA), and the single most sought-after motivational speaker on the foreclosure profiteering seminar circuit, Dino Rossi (R), we find Sen. Murray leading 50% to 42%. This makes five polls in a row that has Murray in the lead.

This poll surveyed 500 people from October 5th to the 14th. Hence, it overlaps every one of the other five October polls. Here is summary of all the polls this month:

	Start	End	Samp.		%	%	%
Poll	date	date	size	MOE	Murray	Rossi	Diff
WA Poll	05-Oct	14-Oct	500	4.3	50	42	D+8
SurveyUSA	11-Oct	14-Oct	606	4.1	50	47	D+3
CNN/Time/OR	08-Oct	12-Oct	850	3.5	51	43	D+8
Elway	07-Oct	11-Oct	450	4.6	51	38	D+13
Fox	09-Oct	09-Oct	1000	3.0	47	46	D+1
Rasmussen	06-Oct	06-Oct	750	4.0	46	49	R+3

What does this poll tell us by itself? A Monte Carlo analysis using the observed sample size and percentages gives Murray 902,830 wins to Rossi’s 92,001 out of a million simulated elections. That is, the poll provides evidence that Murray would have a 90.8% probability of beating Rossi in an election held now.

But we can pool the results from all six of the polls shown in the table. This gives us a total of 4,156 “votes”, of which 3,890 of them go to either Rossi or Murray. Murray gets 48.9% to Rossi’s 44.7% (or, if we look at just the votes for Murray and Rossi, Murray is up 52.2% to 47.8%). A simulation analysis finds Murray beating Rossi with 975,049 wins to his 24,332 wins.

The new KCTS/KPLU/Washington Poll has added a bit more certainty to Murray’s lead. Yesterday, this same analysis using all but today’s poll gave Murray a 94.5% chance of winning. Today, with six polls, we find that Murray would have a 97.6% chance of beating Rossi in an election right now.

Finally, let me address a meme about this race that has been quite prevalent in the recent media. There is an idea the the polls are “all over the place.” Maybe. But not really. Let’s look at results of polls taken in September and October:

Since these polls are taken on a sample of the underlying voting population, there is some uncertainty about the results. The vertical bars on the poll results show the plausible range that the voting population could have, given the margin of error inherent in the poll. (Other problems can lead to biased estimates; the margin of error only incorporates uncertainty reflecting the size of the sample.) And even then, by chance, we expect the true value to lie outside of the plausible range in about one in twenty polls.

Looking at the recent polls, it seems the last five polls mostly overlap, suggesting that the truth lies somewhere near a 54% Murray, 46% Rossi result. We also see this in early September. From mid-September to early October, Murray appears to do significantly worse.

Remember the discussion with the Elway poll last Tuesday? Elway was compelled to justify his noticeably higher numbers for Murray, and suggested that there was a difference between robopolls and live-interview polls. (Of course, if we consider the plausible range, the Elway poll fits right in.) The polls released since Tuesday support Elway’s suggestion.

Elway did not offer a hypothesis why robopolls would turn in different results than the live-interview polls, but I did. Via Goldy:

As Darryl suggested the other night at Drinking Liberally, what if the enthusiasm gap we’ve all been hearing about is largely manifested in who is or is not willing to give up ten minutes to interact with an automated poll? I know I’ve hung up on a couple robo-polls this year, but find it harder to do so with a real live person. Interesting hypothesis.

This looks right for all the recent polling. We see that the three live-interview polls (Elway, WA poll, and CNN) all post numbers on the high side and the robopolls (SUSA, Rasmussen and FOX News) all come in on the low side. That big dip in the middle is formed from six robopolls taken in a row. In fact, Rasmussen and FAUX News both use the same company—Pulse Opinion Research–to do their polling. It may not be chance that FOX New and Rasmussen both give Murray her worse performance in this series. And this would explain the question I posed in August, “What the hell happened to Survey USA?”

The robopoll/live interview/enthusiasm gap hypothesis was offered Tuesday before the three most recent polls were released, and I am now pretty confident that this is what is happening in the Murray—Rossi race here in Washington.

I am less confident about other House and Senate races nationally, but if it can happen here, it might well be happening elsewhere. I think there is a fair possibility that robopolls will systematically underestimate the performance of Democrats nationwide, which will make for a pretty darned interesting election night!

(Cross posted at Hominid Views.)