Judging from both the volume and the tenor of the comments in the Tuesday edition of the Wrap, the polls that got the most attention were the pair of polls by Michigan-based McCollum Foster White, in conjunction with Baydoun Consulting. With new polls giving Mitt Romney a marginal lead in Michigan, and a gigantic one in Florida, the old maxim seemed most apt: if it looks too bad (or too good) to be true, it usually is.
Our community immediately leapt on some demographic numbers within those polls that absolutely deserve scrutiny, and can fairly call into question the veracity of the poll.
Now, that is often the last province of the losers in the poll analysis game. When a poll is released, those on the short end of the numerical stick often dive under the hood, looking for any seemingly aberrant stat in the sample, in hopes of hanging an "outlier" tag on the entire poll.
Well, there are times where that is semi-foolish wishful thinking, and there are times where it might be more legit. That's what we'll explore after we check out the (rather varied) data buffet for this Wednesday.
And, with that, on with the numbers...
PRESIDENTIAL GENERAL ELECTION TRIAL HEATS:
NATIONAL (AP/GfK): Obama d. Romney (47-46)
NATIONAL (Gallup Tracking): Romney d. Obama (47-45)
NATIONAL (Rasmussen Tracking): Romney d. Obama (46-44)
NATIONAL (YouGov): Obama tied with Romney (45-45)
FLORIDA (Gravis Marketing): Romney d. Obama (48-45)
GEORGIA (20/20 Insight for A Better Georgia—D): Romney d. Obama (49-46)
MASSACHUSETTS (PPP): Obama d. Romney (55-39)
MONTANA (Rasmussen): Obama d. Romney Romney d. Obama (55-38)
NEVADA (SurveyUSA for Las Vegas Review-Journal): Obama d. Romney (47-45)
NEW MEXICO (Rasmussen): Obama d. Romney (52-38)
VERMONT (Castleton State College): Obama d. Romney (62-25)
WISCONSIN (Marquette Law School): Obama d. Romney (49-46)
DOWNBALLOT POLLING:
AZ-02 (OnMessage for the McSally campaign): Rep. Ron Barber (D) 50, Martha McSally (R) 45
CT-SEN (Rasmussen): Linda McMahon (R) 49, Chris Murphy (D) 46
FL-SEN (Gravis Marketing): Sen. Bill Nelson (D) 46, Connie Mack IV (R) 39
VT-GOV (Castleton State College): Gov. Peter Shumlin (D) 60, Randy Brock (R) 26
WI-SEN (Marquette Law School): Tommy Thompson (R) 50, Tammy Baldwin (D) 41
WI-SEN (PPP): Tommy Thompson (R) 49, Tammy Baldwin (D) 44
A few thoughts, as always, await you just past the jump...
One of the most tried, true, and more than occasionally errant critiques of a poll is the decision to dismiss a poll based solely on the demographics of a sample. Old dog, still hunts. Everyone does it, including me (check the diary history if you doubt).
For example, when the Pew poll was released a couple of weeks ago showing President Obama had a wide lead, Fox Nation spewed forth the following banner headline: "Tainted Polls? Pew Oversamples Democrats To Boost Obama."
Many of those critiques add up to little more than infantile whining. My favorite was a recent right-wing critique that was blast emailed to me which claimed that a sample (the pollster is lost in the recesses of my memory) assumed a six-point Democratic advantage. The audacity!
Of course, that is actually less than the Democratic edge in the 2008 exit polls, and is only two points higher than the average of the last three presidential exit polls. The idea that a two-point shift would yield dramatically different results defies basic math, let alone political realities.
Also, for what it is worth, critiquing a poll on party ID is shaky ground in the best of circumstances. Party ID, more than any other demographic variable, is fluid. Friday's pro-choice suburban Republican could easily be an Independent voter today, given that Todd Akin opened his mouth and seems incapable of closing it.
Yesterday's FMW/B polls in Michigan and Florida were immediately tagged by the readers in our community, and they were tagged on what I would suspect are far more valid grounds.
Were the samples too Republican? Sure, but not dramatically so. The Florida sample was 41-38 GOP, where in previous years, the inverse (a narrow Democratic edge) would probably better reflect reality. Even in the Democratic apocalypse that was 2010, the split was dead even between the two parties.
Where these polls look way off the fairway is in their assessment of the ages of the electorate. Even with a re-weighting, the pollsters are making some extraordinary assumptions about the electorate in Florida:
Voters ages 18 to 30 years old:
1.33% of respondent universe versus 1.8% of (FMW)B PVBA model projections for 2012 November general election and 16.3% of Florida’s overall registered voter base.
Voters ages 31 to 50 years old:
7.65% of respondent universe versus 10.1% of (FMW) BPVBA model projections for 2012 November general election and 21.6% of Florida’s overall registered voter base.
In other words, voters that are 50 years of age or less make up 38 percent of the registered voters in Florida (which is, admittedly, one of the oldest states of the Union). The pollster is assuming they'll make up 12 percent of the electorate.
What did they do in 2010, a crappy Democratic year? 34 percent of the electorate.
In 2008? 51 percent of the electorate.
That's nothing short of absurd.
Michigan's poll, which you might recall is the only poll in 2012 to give Peter Hoesktra a lead in that state's Senate contest, is tethered to the planet at a slightly better clip. It assumes that 18-50 year olds will make up about two-fifths of the electorate. But in 2008? It was a full 60 percent of the electorate.
We see a slightly less ridiculous example of this in today's SurveyUSA poll of Nevada. If you buy stock in SUSA's numbers, this swing state is living up to its billing, with Barack Obama up by just a couple of points when paired with Mitt Romney. However, this poll also assumes that the two are essentially tied with Latino voters, something even SUSA is acknowledging is strange:
The new SurveyUSA poll showed Obama and Romney dividing the Hispanic vote 48 percent for the president and 47 percent for the former Massachusetts governor - a result even the pollster questioned. In June, a Latino Decisions poll showed Obama leading Romney among Hispanics, 69 percent to 20 percent.
"We were surprised by the Hispanic result, too," Leve said, adding it's difficult to accurately survey Latinos. "It raised some eyebrows, but that is a snapshot and that is what we see."
With a separate poll of only Latino voters
released today showing Obama up by a mere 63-28 margin over Mitt Romney, that is a bit fanciful.
However, that does not mean Barack Obama is really up huge, as some folks often conclude when a sample fails to meet their preconceived assumptions. For one thing, you are talking about a fairly small sample size (by my best guess, about 100-120 respondents). Even if you played with the numbers to give Obama a "more realistic" slice of the Latino vote, it would move the needle (if the math on the back of my envelope is right) to about a 6-7 point Obama lead.
That said, those kind of considerations are worth considering when you see a sample of African-Americans with Mitt Romney garnering 25 percent of the black vote, or the like. Bear in mind: (1) the margin of error with those groups is huge and (2) altering the numbers, given the tiny slice of the electorate represented, doesn't account for a whole hell of a lot.
Of course, in the final analysis, the best solution is to look at a large sample of polls in the same state or place, and figure out where the critical mass of them tend to congregate. Of all the ways to read polls, this is the one that tends to give you the straightest shot, and saves you the irrational exuberance (or instant grief) of a single, errant poll.
In other polling news...
- Because they don't reveal their demographics, and because most of their stuff is behind a paywall these days, I can't tell you what is up with that Connecticut poll. I will only say that if any of my conservative friends or readers wants to take Linda McMahon, and lay me the three points represented in this latest offering by the House of Ras, I'll lay cold cash on the table. Right. Freaking. Now.
- On the other end of the spectrum, and with apologies to my most optimistic and devout Obama friends and fans, I absolutely do not believe that the president is close to a coin flip in Georgia. Now, having said that, I did look at that poll, and I don't see any stats that immediately fall into "gotta be shitting me" territory. The poll is a little too Democratic (D+6 with the likely electorate, compared to D+3 in the 2008 exit polls), and 2008 was a year that Georgia was closer than it has been in recent years, so that could easily represent a high water mark. And the poll is not quite "white enough" when compared to the 2008 exits. Depending on how they define it, though, the issue here could be geography. The poll says that 63 percent of their sample came from Atlanta. Locals can chime in, but that seems a bit high, unless they are including suburbs, exurbs, and anyone who is a Braves fan, presumably.
- A race to watch in the next few weeks is in the Badger State. Today's pair of polls look pretty damned good for Republican Tommy Thompson. The question now is whether this is a bump that came from the party unity that came out of his primary victory, or a cementing of the state of this race. Worth noting: these polls were in the field Thursday through Sunday, which was largely before Thompson's mini-meltdown on the issue of the release (or lack thereof) in his tax returns.