New polling data from Virginia shows a decline in Gov. McDonnell's approval rating and not much change in the governor's race. We ask two true survey sampling gurus to assess PPP's unorthodox “RRR." And we could all use a vacation, but that goes double for the Boston Globe's vacation tracker infographic team. This is the HuffPost Pollster update for Wednesday, July 17, 2013.
VIRGINIA GOVERNOR'S APPROVAL DECLINES - Quinnipiac: “Virginia women have turned on Gov. Bob McDonnell, pushing his job approval rating to a new low of 46 - 37 percent, according to a Quinnipiac University poll released today. Voters still say 44 - 36 percent that the governor is honest and trustworthy and only 16 percent of voters, most of whom are Democrats, think he should resign. 'The lofty levels of 2--1 job approval that Virginia Gov. Bob McDonnell once enjoyed have slipped away with six months left in his term. He's under 50 percent for the second poll in a row, with just a 9-point net approval after substantial media coverage of his relationship with a campaign donor and associated problems,' said Peter Brown, assistant director of the Quinnipiac University Polling Institute. 'Almost 80 percent of voters are aware of the controversy and seem to be somewhat concerned. The bottom line seems to be that they view him as just another politician, but at this point they are not clamoring for his scalp.'" [Quinnipiac]
Another poll this week put his numbers even lower - PPP, on Monday: “PPP's newest Virginia poll finds that Bob McDonnell's net approval rating has dropped 12 points in the last month, and that for the first time since taking office he's under water. Only 36% of voters approve of the job he's doing to 41% who disapprove." [PPP]
Clinton leads 2016 match-ups - The Quinnipiac poll also included questions on potential 2016 match-ups: "[F]ormer Secretary of State Hillary Clinton leads New Jersey Gov. Christopher Christie 45 - 40 percent and tops Kentucky Sen. Rand Paul 51 - 37 percent. Christie would beat Vice President Joseph Biden 46 - 38 percent and Biden would beat Paul 47 - 40 percent." [Quinnipiac]
-The National Journal's Ron Brownstein: "Does GOP want to bet 2016 on improving w/whites? In @QuinnipiacPoll in VA HRC fav w/whites is 15 points higher than Obama approval w/them." [@RonBrownstein]
-The Guardian's Harry Enten: "Gonna be real tuff in a VA dem primary for Clinton with a favorable rating of 97% among Dems. [@ForecasterEnten]
ROANOKE COLLEGE POLL GIVES EDGE TO CUCCINELLI - Two new results in two days paint somewhat different pictures of Virginia's gubernatorial contest. A new live-interviewer Roanoke College poll of the 525 of the Commonwealth's registered voters gives Republican nominee Ken Cuccinelli a six-point advantage over Democrat Terry McAuliffe (37 to 31 percent), but also finds more than a quarter (27 percent undecided). An automated survey by the Democratic firm Public Policy Polling, released on Tuesday, also showed Cuccinelli with 37 percent of the vote, but showed McAuliffe ahead with 41 percent of the vote (and just 15 percent undecided). The new poll helps give McAuliffe a roughly 4 point lead on the estimate produced by the Pollster chart (41.3 to 37.8 percent). [Roanoke University, Pollster chart)
Little evidence of change - Whether the Roanoke poll or others in Virginia, none have yielded much in the way of trend. The survey has been consistent about producing far more undecided respondents than other polls, but Roanoke's three polls have all given Cuccinelli at 5 or 6 percentage point advantage. Via email, Dr. Harry Wilson, who directs Roanoke's Institute for Policy and Opinion Research, explains that they "typically have more undecided than other polls. I've never known precisely why. I try to get my interviewers to push, but perhaps others push harder. Still, our undecideds are dropping, as we would expect. People are really not tuned into this race."
Quinnipiac will make three - In an email, Quinnipiac says it will release results on the governor's race on Thursday morning.
AMERICANS DIVIDED ON ZIMMERMAN VERDICT - HuffPost's Emily Swanson: "Americans are divided over whether they'd have found George Zimmerman guilty of a crime in the shooting of Trayvon Martin, according to a new HuffPost/YouGov poll. But the survey reveals a stark divide between black and white Americans over the outcome of the case.
Forty-one percent of respondents to the new poll said that they would have voted to find Zimmerman not guilty if they were on the jury, while a combined 38 percent said they'd have found him guilty of one of two crimes, including 15 percent who said they'd have voted to find him guilty of murder and 23 percent guilty of manslaughter. Twenty-one percent said they weren't sure...Asked which of four words best described their response to the verdict, respondents were closely divided between positive reactions (22 percent were 'pleased' and 19 percent were 'relieved') and negative ones (27 percent were 'disappointed' and 13 percent were 'angry')." [HuffPost]
ACADEMICS WEIGH IN ON PPP SAMPLING METHOD Yesterday's update reviewed an extended critique via Twitter by Republican pollster B.J. Martino of an unusual procedure used by the Democratic firm PPP. The technique, as described by PPP, involves first, interviewing "more than enough people," then "randomly reject[ing] individual surveys from demographics that are overrepresented. For example, in our polling more women answer relative to men, and not enough African-Americans answer our surveys. Our random selection eliminates any potential bias from the rejections, plus it functions like a quota, only after the fact. PPP also employs a mathematical weighting scheme that assigns a weight based on each demographic." We asked two prominent academics with expertise in survey sampling to offer their assessment of PPP's unorthodox technique. [HuffPollster, @BJMartino, PPP]
Gelman is 'baffled' - Andrew Gelman, professor of statistics and political science and director of the Applied Statistics Center at Columbia University, blogged his reaction: "I admit to being a bit baffled by all of this. If this organization is actually going to the trouble of doing full survey interviews on these people, then they definitely shouldn't be throwing away the responses. Maybe they're doing some sort of screening, where they're only asking a few questions and using these in order to decide whether to go on? That could make sense. The whole thing seems so odd to me that I wonder if I'm missing something here." In the comments section, he added: "If you poststratify [weight the data], I don't see any need to discard data...if you discard data you throw away information. If the goal is to learn about the whole population, I see no reason to throw away respondents of type X just because there are too many of them. Better to use all the respondents and get a more accurate inference." [AndrewGelman.com]
Jackman says random error should be similar - Stanford political scientist Simon Jackman, who designs HuffPost's polling models, reacts via email: "As a strategy for reducing bias, retrospective random rejection [RRR] to get down to target quotas can't be that much different from post-stratification weighting of all the data, if it really is random accept/reject within pre-defined quotas. Both are trying to reduce bias, given a set of assumptions about the target population." Both also increase random error. Jackman continues: "There is an interesting issue as to whether I'd want a data set that has been subject to less weighting but has a smaller nominal n (what you'd get from PPP's RRR/quota method), vs a larger nominal sample size but more weighting (say, if we applied conventional post-stratification to the original data), where both data sets match a set of population marginals (by construction). As the weighting gets more aggressive, the implied design effect [random error] goes up, and the effective sample size is (much?) less than the nominal sample size. But even so, it's hard to see how you'd be worse off with more data, even if you had to weight aggressively."
Fraud? Martino's most pointed comments concern shifts in partisanship and presidential candidate support he observed in the final PPP/DailyKos/SEIU poll based on the discarded responses. His worry? "The discard process can be (can be, not saying is) manipulated to produce desired results." While that is true, the same can be said for the data weighting that virtually all pollsters do after interviews are completed. A pollster willing to manipulate the data to produce a desired result could do so easily. Ultimately, the best protection against that sort of fraud would be total transparency about data and weighting procedures.
DailyKos, to their credit, took the unusual step of posting PPP's raw data online, which enabled Martino's analysis (though has he points out, the details about their data discarding and weighting procedures can be murky). Though they do not discard interviews, Martino's company, The Tarrance Group, did not publish raw data from any of the dozen or so polls they publicly released in 2012.
WEDNESDAY'S 'OUTLIERS' - Links to more news at the intersection of polling, politics and political data:
-Approval of Congress remains dismal. [Gallup]
-Most Americans rate relations between various racial and ethnic groups as "good." [Gallup]
-David Hill is skeptical that skeptical that polls can accurately gauge complex attitudinal constructs like racism and sexism. [The Hill]
-On Twitter, reactions to the George Zimmerman acquittal that offered clear views of the result expressed anger over opposition by more than 4:1. [Pew Research]
-Americans are increasingly likely to describe themselves as “independent" and “moderate." [Pew]
-David Leonhardt sees few absolutes and little movement over time in the abortion debate. [NYT]
-Nate Cohn finds it “preposterous" that Wyoming gets Senate representation. [TNR]
-The Boston Globe shows us what happens when data journalism meets vacation time. [Globe]
POLLSTER UPDATE VIA EMAIL! - You can receive this daily update every weekday via email! Just enter your email address in the box on the upper right corner of this page, and click "sign up." That's all there is to it (and you can unsubscribe anytime).