Today, without question, was a day in the political data world that was both quieter (less than half the polling load) and carried less shock value (read: no NBC/Marist polls) than what readers saw in Thursday's edition of the Wrap.
And that affords us the opportunity to take a second to look at something that yesterday's eye-popping NBC/Marist numbers ginned up. Within milliseconds of those polls hitting the streets, right-wing doubters from hither and yon rose in unison to declare that the partisan configuration of the polling samples is way the heck off, and therefore "honest" polls of the race would have things far, far different.
This is hardly a new meme; indeed, it has been the topic of a Wrap or two during this cycle already. With today being a relatively light day, and with little prospect of light days going forward, however, today may be an excellent day to revisit that topic.
First, though, onto the numbers:
PRESIDENTIAL GENERAL ELECTION TRIAL HEATS:
NATIONAL (CBS/New York Times): Obama d. Romney (49-46 LV; 51-43 RV)
NATIONAL (Gallup Tracking): Obama d. Romney (49-44)
NATIONAL (Rasmussen Tracking): Romney d. Obama (48-45)
COLORADO (SurveyUSA): Obama d. Romney (47-46)
ILLINOIS (Southern Illinois Univ.): Obama d. Romney (47-34)
MICHIGAN (Foster McCollum White/Baydoun): Obama d. Romney (46-44)
NEW JERSEY (Fairleigh Dickinson): Obama d. Romney (52-38)
NORTH CAROLINA (Rasmussen): Obama d. Romney (51-45)
VIRGINIA (Rasmussen): Obama d. Romney (49-48)
WASHINGTON (Elway Poll): Obama d. Romney (53-36)
DOWNBALLOT POLLING:
AZ-09 (GBA Strategies for House Majority PAC et al--D): Kyrsten Sinema (D) 45, Vernon Parker (R) 41
MI-SEN (Foster McCollum White/Baydoun): Sen. Debbie Stabenow (D) 47, Pete Hoekstra (R) 42
NV-03 (SurveyUSA): Rep. Joe Heck (R) 53, John Oceguera (D) 40
WA-GOV (Elway Poll): Jay Inslee (D) 44, Rob McKenna (R) 41
A few thoughts, as always, await you just past the jump ...
Someone smarter than I am once quipped on Twitter that the surest way to know if your candidate is losing is if you hear a lot of people online complaining about biased samples in polls. There is, without a shadow of a doubt, more than a kernel of truth in that.
The reason for that, though, goes beyond just partisan wishful thinking. It is also because we, as devout partisans (mostly), tend to forget that party ID is a fluid characteristic in a number of people. So, too, is enthusiasm to participate in the political process.
Therefore, we fall into the trap (and I include myself in the "we" here, because I was as guilty as anyone in 2010 of this crime) of looking at a polling sample, comparing it to the most recent exit polls, and immediately dismissing it because there is some disparity between the two. That argument falls somewhere along these lines:
"Heh. That PPP poll showing Obama up by six points is so biased. They had seven percent more Democrats than Republicans! In the 2010 exit polls, the Republicans were tied with the Democrats! Stupid liberal media..."
On two levels, that is an asinine argument. For one thing, it is almost always a joke to try to conflate 2012 turnout assumptions with the 2010 exit polls. Somewhere between 40-50 million
more people will vote in 2012 than voted two years ago. The two voting universes are nowhere near comparable, nor are they hardly ever comparable.
For another thing, self-identified party allegiances are always changing. In 2008, the Democrats indeed had a seven-point edge in the presidential exit polls nationally. In 2004, it was tied between the two parties. In 2000, it was a four-point Democratic edge. Thus, for anyone to paint a party ID composition of a poll as "biased" is potentially dangerous, because these numbers do shift with such ease and such frequency.
Look, I rag on Rasmussen all the time. And I still maintain that their assumption that the likely electorate in November will be 4-5 points more Republican than Democratic is a freaking joke. And I do have history to buttress that opinion. In the last 36 years of presidential elections, Republicans have never turned out in greater numbers than have Democrats in a presidential election. That 2004 tie between the two parties remains the high-water mark for the GOP.
But that does not mean that it cannot happen. It just means that it has not happened yet. If Rasmussen is right, then (a) they truly are ahead of the curve, and (b) we can all prepare for President Romney. But if they are wrong, it means that their polls are, indeed, a bit biased. Especially since Rasmussen, unlike virtually every other pollster that is respected in the community, actually weights their national polling numbers by partisan identification. Which means that R+4 or R+5 electorate is actually baked into their final results. Which goes quite a long way to explaining how every other pollster looking at the race nationally gives the president a modest-to-healthy lead, while the House of Ras is on an island claiming a slight advantage for Mitt Romney.
So, when can we call out bias in a poll?
I'd say the charge becomes more fair in two circumstances.
1. When the party ID is drastically different from recent past indicators
One of the most idiotic charges made by casual poll watchers is something along these lines: "In 2008, Ohio was Obama +8. This NBC poll was Obama +10. This poll is biased to the Democrats!"
That's just flat-ass dumb, and leaving aside that party ID does shift, simple arithmetic can explain why. Even if Romney and Obama were getting 100 percent of their party's votes (which they never will), a two-point spread in party ID would yield a difference in the final topline numbers of ... wait for it ... two points. And that is assuming that the real state of play is a Democratic edge of 8 points.
Contrast that, for example, with today's House polling in NV-03. That poll assumed (when leaners were pushed) that the Republicans had an edge in party ID in the 3rd district of 11 percentage points. Democrats, actually, have a very slight edge in registration there. A shift of two points is meaningless. Being eleven points off? That might be more worthy of scrutiny.
2. When other demographic indicators are way off.
I don't want to pick on SurveyUSA, who produced that Nevada poll, but their numbers in the 3rd district also are instructive here. Another reasonable criticism of errant polling is when they have demographic numbers that are just whacked. But the issue here is to make sure that they are way off the mark.
The reason why is the simple fact of sample size. If only 3 percent of the electorate is black, saying that a poll showing Romney getting 20 percent of the black vote instead of 5 percent really doesn't mean squat to the final outcome.
But, let's look at that SUSA poll. They have Democrat John Oceguera up over incumbent Republican Joe Heck by 11 points among Hispanic voters. When the Hispanic vote is roughly 15 percent of the district's electorate, and the 2008 exit polls (and more recent polls) have shown Barack Obama winning the state's Latino vote by margins that were as high as over 50 percent, an 11-point margin may be underselling it in a way that could matter to the outcome.
Again, the issue here is impact. Being "off" by a few points on any particular sub-category or demographic isn't all that important because (a) it may not be "off" at all, given shifting moods within the electorate and (b) the relatively small impact to the overall numbers that a few points of movement in a small slice of the electorate will create.
In other polling news ...
- Two bits of catnip for polling junkies, courtesy of Mark Blumenthal and the crew over at HuffPo/Pollster. They have revamped their 2012 presidential election polling model. The explanation is at the link, but the new model takes a lot of the uncertainty about state projections. Ergo, what had been a projected map of Obama 247, Romney 191 (and a bunch of tossups) is now a model that projects Obama 313, Romney 206.
- In another welcome development, you also now have the ability to download their polling database, and make adjustments to the data as you see fit (eliminating RV samples or LV samples and the like).