The uber-analytic consultant Bruce Gyory has penned another memo to mayoral candidate Bill Thompson's campaign manager Jonathan Prince. Obtained by Newsday, the latest piece dissects and contrasts recent public polls -- and the conflicts in the snapshot stories that they tell. For an interesting chunk of insider-ish discussion, here's the text:
On August 16th , I sent you a memo analyzing the Siena/NYT poll released on August 9, the Quinnipiac poll out on August 13th and the Marist/WSJ poll released on August 15th.
Each of these polls presented a completely different portrait of the race. I wrote, “the stark disparity, both in the cross tabs and the percentage of undecided, render these three polls a kaleidoscope of unanswered questions, rather than a clear prism through which to project the outcome of this primary. To pretend otherwise is folly.”
That statement is even truer this week after the release of 3 new polls:
1. On August 28th Quinnipiac released a poll with DeBlasio at 36%, Quinn at 21% and Thompson at 20% with 8% undecided (margin of error 4.1%).
2. On August 29th Newsday released a poll by Penn Schoen and Berland showing the race at DeBlasio 29%, Thompson at 24% and Quinn at 17% with 13% undecided (margin of error 4%).
3. On August 30th the Siena/Times poll has the race at DeBlasio 32%, Thompson 18% and Quinn 17% with 16% undecided (margin of error 4.4%).
These polling results remain irreconcilable, especially given the Q poll being an outlier on the percentage of undecided. Is DeBlasio on the cusp of hitting 40% avoiding a run-off?; or are DeBlasio and Thompson in a real contest for first and second with Quinn fading?; or is DeBlasio enjoying a large lead with Thompson and Quinn in a dog fight for second?
At least one of these polls is wrong. It would be a distinct mistake to just average these to the middle and presume that the Siena/Times poll was correct. For their samples and timing of taking those samples are very different.
There is no certainty in which of these portraits of the primary electorate is a Rembrandt, which a Dali, or if there is a forgery in our midst. Nor should we assume the Newsday poll is accurate, just because we like those numbers better. Instead, I suggest we look at the make up of the samples and their timing to develop a feeling for which is less and perhaps which is more, accurate.
Before doing that assessment there is a critical point to be made. While the margins for error in all three of these public polls are between 4.0 and 4.4%, the margin for error in their cross tabs is far higher, at least 15%, due to the relatively small overall sizes of their likely voter samples (Q poll’s was 579, Newsday’s Penn Schoen was 600 and the Siena/Times was 505).
In this primary, as I pointed out in my August 13th memo, where the impact on outcome of how the Black vote divides between Thompson and DeBlasio, the Hispanic between all 3 top tier candidates and the Jewish, as well as white Catholic voters, means that if these cross tabs are off by 10-15% in their reporting of where these voters are heading, it will have a dramatic impact on the accuracy of their projection (i.e., given that each of the candidates are drawing strength from distinct pools or blocs of voters).
The sample underlying this Q poll is particularly troubling. The Black share of the Q poll sample drops from 35% in their August 13th poll (which was high) to 27% which may be low. That is wild for such a huge drop for the same pollster. Women drop from 59% to 55% in its share of the sample from August 13th to August 28th. Women are usually 57 or 58% of a Dem primary, and the Asian drops to 2% share.
So the Q poll’s August 28th sample is less Black, less female and way low on Asians (who will be 5-7% share when the votes are counted) than their last poll and no doubt than the September 10th electorate.
Wonder if the Q poll folks wanted to strain to create a headline that DeBlasio had a chance to reach 40%? If their sample was 58% Female, 30% Black and 6% Asian, this poll would probably have shown DeBlasio 32%, Thompson 23% and Quinn 21% with Liu at 9%.
To repeat, while the Q poll’s margin of error is 4% on top, it is much higher in the subsets. And surprise, surprise that from the August 13th sample to the August 28th, Thompson drops from 39 to 25% and DeBlasio goes from 22% to 34% amongst Blacks. A dramatic swing. Yet this poll also shows Thompson going from 12 to 18% amongst whites. Pretty wide and contradictory fluctuations in all these cross tabs. Did Thompson really decline dramatically amongst Blacks, while he grew sharply amongst whites?
The Q poll also shows Quinn losing white males to DeBlasio, especially moderate and conservative white males to DeBlasio, while he rises from 40 to 50% amongst liberal primary voter. In fact, this poll shows DeBlasio snaring 29% of moderate and cons Dems with Quinn at 22% down from 25%. So with tabloids endorsing her and bashing him on their front pages, he rises and she falls amongst outer borough male white conservative ethnics? A curious finding.
Disappointingly, the Q poll once again had no breakdown of Hispanics although this newer poll has a more realistic sample of Hispanics at 18% vs. 15% share from their August 13th sample. I project the Hispanic share at a full fifth of the electorate: 20%.
On the Siena/NYT poll I did notice one stray factor: the length of time the sample was taken. The calling period for this poll stretched from August 19-28, a full 10 days (whereas Quinnipiac was from August 7-12th and the Penn Schoen for Newsday from August 22-27th: each 6 days). Usually a single poll is taken over 3-6 days, not a full 10 days, especially in a contested and heated primary.
For example, two nights of polling in this Siena/Times poll came before the TV debate on August 21, and 3 nights before the Daily News endorsement and 7 days before the Times and 8 days before the Post.
Not to mention that much of the polling by Siena came before we countered with a TV ad pointing out that DeBlasio had lied in his ad on Thompson’s Stop and Frisk position (citing that the Times ad watch found DeBlasio’s ad misleading) and when Bill Thompson looked straight into the camera explaining his position on Stop and Frisk with clarity and emotion.
One of the reasons pollsters usually take polls within a 4-6 day period is to get a single accurate snapshot of the electorate in a time frame where voters are focused on the same events. Now let me be clear, the Times picked Siena for a reason. Of all public pollsters, they have been the first with the most on transparency and disclosure of their samples. They have always been quite consistent on their samples and fair in drafting their questions.
Did Siena make a scientific mistake – no, but did the 10 day polling period leave their poll vulnerable to missing the impact of the editorial endorsements for Quinn and our response ad to DeBlasio on Stop and Frisk, especially amongst Black voters – yes.
Which leads to the Penn Schoen poll for Newsday. Its calls came the most recently from August 22-27th. So its results came when voters had already read and heard about the editorial endorsements for Quinn and after Black voters had seen our response ad to DeBlasio’s “misleading” ad.
If the Penn Schoen poll is accurate, we have a lot to take comfort in and the Quinn campaign should be depressed. In this, the last poll taken chronologically, unlike the Q poll, we lead DeBlasio amongst Blacks 35-29%. This poll also shows us breaking 20% amongst whites (tracking the progress we made from the early August to the late August Q poll amongst white voters from 12-18%). This poll suggests, especially given our dead heat with DeBlasio amongst Jewish voters, that we may grow toward 25% amongst whites by primary day. This poll also shows when compared with earlier polls that we are starting to grow amongst Hispanics.
The bad news for Quinn is that Penn Schoen calls were made during her endorsement run (the Daily News, Times and Post endorsed between August 22-August 26). If the Penn Schoen data is accurate, the endorsements carried large sails but Quinn had no wind to skipper her boat.
The Penn Schoen data shows the Hispanic vote is heading towards becoming a three way dead heat, with us second to Quinn. With high functional undecided amongst Hispanics, and our holding the strongest validators going into that race: the endorsements of El Diario, as well as Ferrer, Serrano, Diaz, Jr., Espaillat, Dilan and Peralta, if we get this late break amongst Hispanics, given the trendline amongst Black and white voters, it will be a good omen. In short, if the Penn Schoen data is correct, our contest with DeBlasio could become a dead heat, with Quinn trailing in an intermediate tier, if Hispanics join Blacks in coming home to Bill Thompson.
As an aside, I think all 3 polls, with Q poll sample being the most egregious, is underestimating the Asian vote and doing a disservice by not sampling enough Asians to get a read on that growing factor in NYC politics. I would be shocked if the Asian vote were not at least 5% of the total vote (it could be as high as 7 or 8%) and that will leave John Liu close to or in double digits when the votes are counted. In fact, I suspect Liu will nip Anthony Weiner for fourth place given the polling data’s undercounting and undermeasuring the Asian voter in this primary.
In the final analysis, what this polling data really tells me is that the pollsters have no clue what is going to happen on September 10th. At least one of these pollsters is going to need to buy a spatula to scrape the egg off their face.
Here is what this race will boil down to – turnout. I suspect DeBlasio’s camp is projecting a turnout in the 500,000 range, tracking Professor Mollenkopf’s public projection. If Mollenkopf is correct, that will mean Manhattan will cast the largest share of turnout (just shy of a third of the total vote).
Our model projects a 700,000 or higher turnout. We have listened to Jerry Skurnick who in the Times reminded everyone that in an open mayoral election where the winner of the Democratic primary is perceived to have a strong shot at actually becoming Mayor, turnout has never been less than 700,000 in a Democratic primary.
I still believe Skurnick has the better argument over Mollenkopf. It is not just an empirical argument. High turnout is driven by racial and ethnic groups in play, the outcome in doubt, heavy horse race coverage from the media and a consequential office at stake. All exist in abundance in this primary.
If TO hits 700,000 that means the outer borough vote will cast over 70% of the total primary vote. I think our turnout model is sound and that the Penn Schoen poll is probably more accurate than the others because of its later time frame and unlike the Q poll, Penn Schoen did not juggle its sample.
Let me update the conclusion from the August 16th memo. A fair analysis, distilling all of this public polling data, would conclude this is on the verge of becoming a two-way race where DeBlasio has the clear lead, but we are in striking distance of contesting for first place.