Nate Silver wrote an interesting column this week, arguing that the generic ballot might be underestimating the Democratic position right now. Silver writes:
In the last couple of posts on our House forecasts, I’ve noted that the generic Congressional ballot tends to show worse results for Democrats than other indicators like polls of individual House races. The generic ballot, as of last week, pointed toward a Republican lead of 7 or 8 percentage points among likely voters, according to our estimates. Local polls are a little bit less straightforward to read because you have to make a lot of adjustments for polls that are (i) partisan, (ii) bad, or (iii) both. But in general — while there is a lot of variance from district to district — the local polls seem to point toward a House vote that would be about evenly divided.
Color me a skeptic. Historically, the generic ballot has overestimated, not underestimated, the Democratic position. I’m open to such a change as a theoretical possibility, but there is a significant evidentiary burden to be met. And I don’t think Silver’s analysis meets it. For starters, any analysis that uses House polls is going to be problematic for all the reasons that Silver cites. Beyond that, Senate polling is certainly consistent with an R+7 national year, and you don't find John Dingell under 50 percent in anything short of a GOP tidal wave.
Yet Silver uses the polls from the conservative American Action Forum (AAF) to make the case that the generic ballot is suddenly tilted toward the GOP. The AAF has been polling Democratic-held districts and asking voters for their generic preferences (Republican or Democrat) as well as their preferences for specific candidates. Silver notes that actual Democratic candidates do better than the generic Democrat in the AAF polls, and from this he concludes that such a trend might be in play nationwide.
Yet the AAF poll is a bad data set to make the claim that Silver is making. The reason is that an unusually high percentage of respondents to the AAF answered “Depends.” The result is that the AAF generic ballot polls have a much larger undecided/depends bloc than the national polls. The average of the American Action Forum House polls shows a whopping 28 percent of respondents claiming either they are undecided or their vote depends on the candidates. Compare that to the 12.1 percent undecided in the RealClearPolitics generic ballot average.
This is a problem for Silver's conclusion. There is something unique about the AAF methodology, something that pushed an unusually large number of respondents to an uncertain position. I don’t know what the reason is, but frankly it doesn’t matter here. What matters is that there is a difference between the AAF generic ballot polls and the national generic ballot polls. That’s a problem for Silver because he is using the AAF polls to make a point about the national numbers, which means he is basically comparing apples to oranges.
There’s more. Here is how Silver explains the difference in generic versus actual Democratic support in the AAF polls:
Still, the notion that specific Democratic candidates do slightly better than generic ones would square with what we’re seeing elsewhere in the data. It also arguably squares with the respective strategies of the two parties, as Republicans are generally trying to nationalize the race, while Democrats — lacking much in the way of a coherent national message — are trying to localize it…
The problem with this is that in the AAF polls, specific Republican candidates are doing better than generic ones, too. The generic Republican pulls in on average 39 percent of the vote. Actual Republicans pull in 45 percent. Meanwhile, the generic Democrat hauls 33 percent while the real ones get 44 percent on average. So, the numbers for both sides improve when AAF moves from generic to real. The R-D spread decreases because the Democrats improve slightly more on average than Republicans.
What accounts for the fact that the Democrats do slightly better than Republicans? It might be Silver’s suggestion that the Democrats are successfully localizing the election.
But I doubt it.
Campaigns spin about how they are going to “localize” or “nationalize” elections, but the reality is that parties and candidates have a very limited role in setting the agenda. Instead, the competitive nature of the campaign forces candidates to talk about what the voters want to talk about, just as the competition between Wal-Mart and Target forces both stores to stock what their customers actually want. If voters want to talk about national issues, then the campaign will be about national issues. So, I don’t think Silver's explanation cuts it.
A much more likely account is that AAF polled incumbent-held districts where the Democratic incumbent has substantially better name recognition than his/her challenger. After all, these AAF polls were taken in mid-August when challengers were still essentially unknown. In some of these districts, primaries hadn't even been held, and GOP challengers had not yet been determined!
I think what probably happened is that soft Democratic leaners answered “Depends” to the generic ballot question for whatever reason, then gave support to the incumbent Democrat whose name they recognized. Meanwhile, some Republican leaners and pure independents (who will likely end up voting Republican nationwide this year), answered “Depends” as well, but they did not recognize the name of the GOP candidate, so they were undecided when the choice came down to two actual people.
In support of this proposition, consider the following: 7 of the 31 districts AAF polled show a larger break toward the specific Republican candidate; in 5 of those 7, the Republican running has already run for office before, and thus has better name recognition. The other two that break toward the GOP are PA-3 and CO-4, districts where two freshmen Democrats announced late-stage, high-profile support of the health care bill.
This explanation has the added virtue of resolving the tension that Silver’s analysis leaves behind: Historically speaking, the generic ballot has favored Democrats, sometimes by overwhelming margins, yet this year it is suddenly undercounting the Democratic strength. The alternative account I offer – wherein methodology and differences in name recognition explains the peculiarities of the AAF polling – does not require us to stipulate a historic change in the tilt of the generic ballot.