By and large, this first week of the "real campaign season" was a decent one for Democrats, on the polling front. Even nominally Republican pollsters (We Ask America, Harper, and Gravis) had decent polling numbers for Democratic challengers in Wisconsin, Pennsylvania, and Florida (though that race in Pennsylvania has been strongly tilting to the Democrats from the outset).
One poll, however, did merit some attention, and not necessarily for the right reasons. One of the weaker results for the Democrats was a new poll, taken by the Bob Graham Center from the University of Florida, which showed incumbent GOP Gov. Rick Scott with a five-point edge over Democratic challenger Charlie Crist.
The concern on this poll was the use of a fairly unusual component to their weighting, and a possible misuse of said weighting, which may well explain why the UF poll led to a more pessimistic outcome for Crist than other polls released in the week.
More on that after the jump, however. For now, feel free to peruse the poll-a-topia for the past week (Aug 29-Sep 4—we'll resume our regular schedule next week):
GA-Sen (GAPundit.com—R): Michelle Nunn (D) 45, David Perdue (R) 43
IL-Sen (Harper Polling—R): Sen. Dick Durbin (D) 44, Jim Oberweis (R) 38
IL-Sen (We Ask America—R): Sen. Dick Durbin (D) 48, Jim Oberweis (R) 41
KY-Sen (CNN/ORC): Sen. Mitch McConnell (R) 50, Alison Lundergan Grimes (D) 46
KY-Sen (Rasmussen): Sen. Mitch McConnell (R) 46, Alison Lundergan Grimes (D) 41
KY-Sen (SurveyUSA): Sen. Mitch McConnell (R) 46, Alison Lundergan Grimes (D) 42
LA-Sen (Rasmussen): Bill Cassidy (R) 44, Sen. Mary Landrieu (D) 41
NH-Sen (PPP—D): Sen. Jeanne Shaheen (D) 50, Scott Brown (R) 44
NH-Sen (Public Opinion Strategies—R): Sen. Jeanne Shaheen (D) 44, Scott Brown (R) 41
OK-Sen-B (Rasmussen): Jim Lankford (R) 58, Connie Johnson (D) 29
SD-Sen (PPP—D): Mike Rounds (R) 39, Rick Weiland (D) 33, Larry Pressler (I) 17
WV-Sen (RL Repass and Partners): Shelley Moore Capito (R) 54, Natalie Tennant (D) 37
AZ-Gov (Rasmussen): Doug Ducey (R) 40, Fred DuVal (D) 40
CA-Gov (Field Poll): Gov. Jerry Brown (D) 50, Neel Kashkari (R) 34
FL-Gov (Gravis Marketing—R): Gov. Rick Scott (R) 37, Charlie Crist (D) 37
FL-Gov (SurveyUSA): Charlie Crist (D) 45, Gov. Rick Scott (R) 43
FL-Gov (Univ. of Florida): Gov. Rick Scott (R) 41, Charlie Crist (D) 36, Adrian Wyllie (Lib) 6
GA-Gov (GAPundit.com—R): Gov. Nathan Deal (R) 44, Jason Carter (D) 42
IL-Gov (We Ask America—R): Bruce Rauner (R) 46, Gov. Pat Quinn (D) 37, Chad Grimm (Lib) 7
MA-Gov (MassINC): Martha Coakley (D) 40, Charlie Baker (R) 31
MA-Gov (MassINC): Charlie Baker (R) 34, Steve Grossman (D) 29
MA-Gov (MassINC): Charlie Baker (R) 37, Don Berwick (D) 19
MA-Gov (Univ. of Mass-Lowell): Martha Coakley (D) 41, Charlie Baker (R) 32
MA-Gov (Univ. of Mass-Lowell): Charlie Baker (R) 33, Steve Grossman (D) 31
PA-Gov (Harper Polling—R): Tom Wolf (D) 52, Gov. Tom Corbett (R) 41
PA-Gov (Robert Morris Univ.): Tom Wolf (D) 56, Gov. Tom Corbett (R) 25
VT-Gov (Rasmussen): Gov. Peter Shumlin (D) 48, Scott Milne (R) 36
WI-Gov (We Ask America—R): Mary Burke (D) 48, Gov. Scott Walker (R) 44
FL-18 (Frederick Polls—D): Rep. Patrick Murphy (D) 54, Carl Domino (R) 33
ME-02 (Public Opinion Strategies—R): Emily Cain (D) 37, Bruce Poliquin (R) 33, Blaine Richardson (I) 6
A few thoughts, as always, await you just past the jump ...
Ever since the 2012 election, and the Dean Chambers saga, polling analysts have become loath to discuss the partisan composition of polls. Any talk that a poll is too rife with Republicans (or Democrats, depending on the partisan tilt of the analyst in question) leads to mocking calls of someone trying to "unskew" the poll.
This Univ. of Florida is begging for the treatment, however, and it has squat to do with "unskewing." Allow me to explain.
Back in the Fall of 2012, I took the whole unskewing gimmick to task. At the time, I offered this critique of the architect of that Republican movement (blogger Dean Chambers):
But the most basic problem here, as several responsible journalists have managed to point out during Chambers' 15 minutes of fame, is that his method of "unskewing polls" is to simply take a party weighting that favors his political ideology, and apply it to every poll. Voila!
In Chambers' case, he uses a recent survey conducted by the Republican polling firm Rasmussen Reports which claimed that Republicans make up 38 percent of the current electorate, while Democrats make up 33 percent and independents 29 percent.
There are a number of inherent problems in this kind of "analysis."
For one thing, as has often been noted in the past two weeks, party ID is easily the most fluid of the demographic subcategories often associated with polling. People change their self-identified preferred political party with some frequency. And it goes without saying that they will change that more often than they will change ... say ... their gender or ethnicity. Even ideology, which is a state of mind as much as anything, is more permanent than one's party ID, which can flit from Democrat-to-independent-and-back-to-Democrat based on the news of the day with some voters.
This is why responsible pollsters tend not to weight for party ID (Rasmussen and Voter/Consumer Research are two noteworthy exceptions). Simply put, to weight for party identification is to substitute where the electorate actually stands with where you think (or, more appropriately in this case, hope) the electorate stands.
In the end, that is why the "unskewing" movement was bunk, and why it was ultimately and forcefully discredited by the outcome of the 2012 elections. You can't just assume the rosiest party composition for "your side", and then recalculate a poll filling in those partisan parameters.
However, in a way, that's exactly what the UF poll did, though I am more than willing to acquit them of any naked partisan preferences.
The UF poll's methodological statement reads as follows:
Results were weighted by age, party identification, and media market, thus allowing the results to mirror the distribution of age groups, party identification, and media market in the Florida Voter File.
These were, to say the least, unusual choices for
which demographic variables to wait. Race is typically a variable that gets weighted, as does gender. Neither of them were utilized, but partisan ID was?
What's more: look at how the partisan ID was weighted. According to a summary of the poll by one of the poll clients (the Tampa Bay Times), the chosen weight by party was 35 percent Republican, 30 percent Democrat, and 26 percent Independent. Aside from one glaring question (what the hell were the other 9 percent?!), there is another problem.
In 2012, the ratios were totally different. It was 35 percent Democratic, 33 percent Republican.
So, you might say, isn't a presidential election year awfully optimistic, Mr. Unskewed?
Yup. So let's look at 2010. Then, it was dead even: 36 percent for each.
Indeed, going through exit polls dating back for the last decade, there has not been a single instance where Republicans have enjoyed a five-point edge in turnout in Florida.
There is a reason why weighting by party ID is a bit dangerous. There is every chance that Florida could tilt as Republican as the UF poll implies. But it is just a chance. And, if the electorate doesn't fit that preconceived notion, the poll is permanently going to be wanting. At the end of the day, party identification is simply too fluid a variable to be a safe parameter upon which to peg a poll. Which, ultimately, is why so few do it.