Three weeks ago, I pointed out that one of the favorite parlor games of political junkies like myself is to heap praise or scorn on the latest polling data based on the demographic assumptions buried within the poll.
This is often, of course, a defense mechanism. Confronted with a poll whose toplines are giving one side or the other a bout of anxiety, nothing calms the stomach more than to dismiss the poll because of its ridiculous split among conservatives and liberals, or how far it deviates from this exit poll or that one.
It is a practice as old as the polling game itself, and I freely confess to doing it myself often. Indeed, I have done it as recently as this month, when something in the Quinnipiac polls in Ohio and Pennsylvania caught my eye.
And it is not an exclusively left-of-center tendency, as evidenced last night by this tweet from Meg Whitman campaign advisor Mike Murphy:
Polls: Party ID numbers in sample: Field/reality: 44D 35R 21DTS. LAT: 55D 35R 9DTS. Trust Field. LAT poll is way, way off. 10 pts too Dem
All too often, "reality" in those cases is defined by whatever creates the data that is most amenable to your candidate. Confronted with a bad poll (something the Whitman campaign has not been used to in recent months), Murphy went with the simplest reply--he refused to accept the premise on which the poll was built.
Given that they have been on the receiving end of more crappy polling this cycle than Republicans, we have seen this technique employed far more often by Democrats.
And, while I proudly count myself within their number, there is a lot of wishful thinking in their complaints.
A lot of the comments I see deride the numbers in recent polling compare the partisan or ideological breakdowns to those in the most recent election.
Comparing current polling demographics to 2008 exit poll demographics is, it must be said, optimistic to the point of being unrealistic. It is fair to say at this point, I would think, that this is not 2008.
For one thing, and this can't be pointed out enough, even the most optimistic assumptions about 2010 turnout will create an electorate that has 30-40 million fewer voters than showed up at the polls to elect Barack Obama in 2008. By that standard alone, it is awfully tough to make any direct comparisons between 2008 demographics and 2010 demographics. Large shifts are practically inevitable when you are chalking off one quarter of the previous electorate.
Furthermore, motivation is intrinsically different, even if Democrats get more fired up between now and November. There is a gulf of difference between smelling blood in the water (as the Democrats did when they chased the control of Congress in 2006 and the White House in 2008) and being the body bleeding into the water, which comes with the territory of being in the majority.
It's not 2008. Nor, we must say, is it 2006.
However, that doesn't mean that the assumptions being made by the pollsters during the 2010 cycle (particularly in their likely voters screens) aren't way, way off.
My colleague DemFromCT made a pretty compelling case to that end yesterday. Looking at exit polling data, it does not seem unreasonable to question those likely voter screens, at least to some extent.
For the obvious reasons stated above, 2006 or 2008 exit polling were not used to draw that conclusion. They are tempting because they are the most recent numbers (and demographics in states do change over time), but they were cycles where the terrain tilted far too palpably to the Democrats. 2004 wouldn't be a bad exemplar, because it was arguably the best election cycle for Republicans in recent years (remember, 2002 had no exit polling). However, as a presidential election cycle where 120+ million voters participated, it seemed incongruent, as well. Therefore, the decision was made to go into the wayback machine and pull exit polling data from one neutral midterm election (1998) and one historically GOP midterm election (1994).
EXAMPLE #1: SurveyUSA poll--New York--September 21
The SUSA poll's most eye-opening piece of data was not its partisan breakdown. That (Democrats +9) was roughly in line with both the 1994 and 1998 exit polls. Where this poll drops jaws in it the ideological breakdown. The SUSA poll is presuming an electorate that is 35% conservative and 20% liberal. Now, there is nothing earth shattering in finding more self-described conservatives than liberals, but in a state as blue as New York, that gap is much, much closer. Even in the most Republican-friendly electorate in recent history (1994), the Con-Lib gap in New York was smaller on Election Day (34 C/22 L). In 1998, the gap was considerably less (26 C/23 L). If a relatively neutral midterm electorate turns out (1998), then the poll goes from 49-40 Cuomo to 52-38 Cuomo. Even if the midterm electorate skews a bit to the R's (say...a midpoint between 1994 and 1998), the result still moves to the double-digits (50-39 Cuomo). The only way you get a single-digit result is if you presume that the electorate will be even more conservative than the best election for Republicans in recent memory.
EXAMPLE #2--Quinnipiac Poll--Ohio--September 16
SurveyUSA could have been used here again, since pretty pessimistic poll on September 13th had both samples whose partisan and ideological makeup created an electorate further to the right than either 1994 or 1998. But, in this state, Quinnipiac is the better bet. While Quinnipiac does not break down their samples by these characteristics the way that SUSA does, we can look at their breakdowns of the trial heat and figure a few things out. And what one can figure out quickly is that the Q poll is assuming an electorate that is considerably more Republican than any election in history. Even if you apply the 1994 voter turnout (37 D/37 R/26 I) to the partisan breakdowns of their mid-September poll, you get dramatically different numbers in the two marquee races. What Quinnipiac forecasts as a seventeen point lead for John Kasich in the gubernatorial race becomes a considerably less lofty ten points (51-41), even by the depressed standards of the 1994 electorate. The effect is even stronger on the poll in the U.S. Senate race. Recalibrating that by the 1994 electorate's party ID, you see a twenty-point GOP lead whittled down to twelve points (53-41).
Double-digit deficits, even modestly reduced ones, aren't exactly anything to write home about. But there is a pretty big psychic difference for the candidate and his supporters between a ten-point race and a twenty-point race. And, remember, that is assuming the worst case scenario in recent history. Bear in mind that Quinnipiac seems to be assuming a decidedly more pessimistic electorate for Democrats than even showed up at the polls in 1994.
Something they were assuming in another state, as well.
EXAMPLE #3--Quinnipiac Poll--Pennsylvania--September 21
As in neighboring Ohio, Quinnipiac's most recent polling in the Keystone State seems to be portending an electorate more slanted to the GOP on Election Day than any we have ever seen.
Consider: typically, more Democrats show up to the polls than Republicans in Pennsylvania. The lone exception to that was in the tsunami year of 1994, where the electorate tilted incrementally to the GOP (41 R/39 D/20 I). If one were to break down the most recent Q poll in the state, and crunch the numbers with the 1994 partisan breakdowns in place, the numbers become dramatically different. In the gubernatorial race, where Quinnipiac saw a blowout in the midst (Republican Tom Corbett +15), a recalculation with the '94 party ID numbers gives a pretty different assessment of the race. Where Corbett led 54-39 in the Q poll, he leads by just 51-42 with the '94 numbers in place. While, on a pure percentage level, the effect was more muted (a seven-point Toomey lead drops to four points), there is a tangible difference in how seven-point races are covered in the press and how four-point races are covered (even if float within the nargin of error could account for all of that difference).
Of course, both pollsters (and others, like Rasmussen, who seem to be assuming a historic R/D spread in November) could be correct. This electoral environment could be worse for the Democrats than even 1994. Gallup released numbers this week that indicated that the gap between how voters feel the political parties "represents their values" leans Republican for the first time since...well...1994. However, as a compendium of polls over at Polling Report makes clear, the GOP is not necessarily seen as a suitable alternative. The majority of recent polls still finds that voters are more likely to say that Democrats would do a better job of solving the country's problems than Republicans. This would seem to make a 1994-esque tsunami a bit more difficult to achieve for the GOP. In typical wave elections, as this excellent piece from DemFromCT pointed out last week, the out-party is more liked than the party in power. In 2010, that simply isn't the case.
Which might challenge the partisan assumptions being made by pollsters throughout this cycle. The post-election postmortems, no matter the outcome, are bound to be fascinating.