"I don't trust the polls. People told me they were skewed" - Dave Weigl
Something happened the other day that could have had profound implications on the way that people see the race this week, but it wasn't to be. Gallup releasing a three day culling of its seven day tracking poll, showing Mitt Romney and President Obama tied at 47% for after the debate, was far more insightful after Obama regained his five point lead in the seven day tracker. It probably meant that after news of unemployment dropping had spread around and hysteria over Romney's good (but hardly earth shattering -- folks, you don't win a real debate by lying your ass off and steam rolling a poor moderator) performance Wednesday had setteled, that Obama had not just regained his pre-debate lead, but actually increased yet.
The problem with that is two-fold. First, we couldn't ever know that for sure because unlike the stupid game Gallup just played with the three day debate sample, they don't release actual daily results. Second, Gallup just switched from its registered voter model to a historically bad likely voter model. Obama went from a five point lead with registered voters on Monday to a two point deficit with likely voters today.
There's always a gap between the two models and likely voter models always favor Republicans. The difference between these models is not skew or bias, there are good reasons for it. But there's simply no way to justify a seven point gap here. To illustrate how conservative and useless Gallup's LV model is, Gallup has Mitt Romney up by two points today while Rasmussen, the most conservative and inaccurate pollster out there which was the only pollster to show Romney with a lead in the entire month of September, has the race tied. And will probably return to showing Obama with a lead by tomorrow or Thursday at the latest.
Now, voting models can be applied post hoc. So Gallup is still releasing its registered voter model results. Just look at the difference here:
49-46, Obama +3 (RV)
49-47, Romney +2 (LV)
Even with the two point drop amongst registered voters, that's still a five point gap.
Meanwhile, look at the results of the other two tracking polls that I... track:
49-45, Obama +3.83 (RAND)
45-45, Tie (Ipsos/Reuters) (LV)
45-42, Obama +3 (Ipsos/Reuters) (RV)
What to take away from all of this?
I've relied heavily on these four tracking polls in the past few days because it's all we've had to go on. But tracking polls are notoriously noisy. Ideally, I'd tell everyone to ignore all polls until the third and fourth weeks of October, once the big non-tracking dogs start barking. We need to see data from CBS, NBC News, New York Times, Wall Street Journal, Washington Post, CNN, FOX News, and Associated Press before we know what we've got, and we haven't had a single poll from any of those organizations in over two weeks.
The numbers have been all over the place. Before the debate, Rasmussen had Obama up two, after the debate down two, and now tied for two straight days. Gallup had Obama up by 6, but no worse than up by 3 with its RV model, and now down by 2 with its LV model. RAND had Obama up by 4-5 before the debate, up by no less than 3 in the days covering the debate, and between +3 and +5 in the past two days. Ipsos/Reuters had Obama up by 5 before the debate, up no less than 2 after, and a tie now.
That's pretty unstable. The average Obama lead before the debate appears to have been 5.45 points, but that includes an outlier +9 for Obama from Ipsos that was +5 the day before, and +2 the day after. The average was probably closer to 4.45 points. After the debate, like say yesterday (it's hard to do much with Gallup today because of the drastic change in voting models), it was 3.5 points. (I'm projecting Ipsos at Obama +2 for the 7th because they didn't release any results that day, and +2 was what they released on the 6th.)
The average today, with Gallup's model switch in mind, is +0.9 for Obama.
This graph may clear things up a bit:
Rasmussen (3 days) flipped and is probably in the process of flipping back. Ipsos, with a longer sample (5 days) is still tightening, but the overall picture says it'll probably come back to Obama in the next few days, probably by two points. Gallup flipped the fuck out and has such a conservative LV model that it may actually be less useful that Rasmussen, but it's RV model shows no real change. And RAND is rock solid stable.
I think the takeaway today is the same as it has been all weekend. Romney clearly gained something, but how much and where depends on the pollster. He hasn't gained enough by a long shot to alter the election outcome based on the electoral college. That's certain thus far. It's relatively easy to change a tight national poll -- simple statistical noise can do that -- but much more difficult to move multiple states from a deficit to a lead. As I last wrote, Romney's state gains appear to have been even more temporary than his national gains.
Nate Silver's models are showing a 302-236 electoral victory for Barack Obama, and a 50.8-48.3 popular vote win. Obama is still a 4-to-1 favorite (74.8%). And you must understand that that model is predicting things to tighten between now and November 7th. If the election were held today, his models project a 305-233/50.7-48.1/82.5 for Obama.
People are still ignoring the importance of the electoral college. Perhaps other than Colorado, which is debatable, no other state has changed hands as a result of Romney's "historic" debate win. You need 270 electoral votes to win and no model has Romney with more than 230. Most models have Obama with 310-330.
So here are your two bottom lines:
1: The national public polling race is tighter after the first debate, and very unstable.
2. The electoral math has barely changed at all, with Obama still a big favorite to win.
If you don't care for my analysis, I don't really see Nate Silver saying much different. But I recommend you read him, even if you have no problem with what I'm doing over here. This last bit is something I was just writing about that nobody (when I say nobody, I mean the media you see on TV and read in the papers) is talking about:
In some ways, then, the election might not be quite so unpredictable as it appears. There was reason to believe that Mr. Obama’s numbers would fade some after his convention — and the first debate has quite often been a time when the challenger drew the race closer.
We still don't know if the change in polling was Obama coming down off his convention bounce, Romney having a debate bounce, or both. Since a bounce is a bounce because it has to come back down, it's possible that we could have any of those three scenarios and still end the week with Obama leading most polls.
We'll just have to wait and see.