Kevin Drum notices something interesting in the polling averages:

According to both Pollster and Real Clear Politics, Mitt Romney began his big surge well before last week’s debate. In the ten days before the debate, Pollster shows Romney gaining 2.4 points and RCP shows Romney gaining 1.8 points.

Okay, I’m going to look at all of this in some detail, but just to skip to the point: I’ll tentatively agree with Drum, but a lot depends on how the averages are being compiled, which gets into technical issues that go beyond my expertise.

But before I get to that, Drum asks “Anyone have any good guesses about what happened during that last week of September?” My guess is that the answer is: nothing. Which means that instead of news that was one-sided in Barack Obama’s favor, which it had been since the Democratic convention started, the news resumed being balanced, thus finally deflating a bounce that was artificially prolonged by September 11 (which likely plays for Obama, bin Laden killer), an initial rally effect from the Libya attack, the Romney response to the Libya attack, and 47%. During periods of one-sided information flows, we would expect undecided voters to temporarily park with the candidate getting good news or be less likely to park with the one getting bad news; we would also expect likely voter screens to pick up more of those who support the favored candidate and fewer of those who support the disadvantaged one — even if no one actually makes new decisions.So if you ask a marginal Romney voter in early September who she’s going to vote for and all she’s heard lately is that Romney is in trouble, she might say that she’s undecided. So what might be happening in the ten days or so before the debate is that the news environment goes back to even, and all that temporary stuff washes out, revealing a 2-3 point lead for Obama rather than the 4-5 point “bounce” version.

As for whether that happened? Looking closer at Pollster (and if you don’t care about the details, skip down to the last paragraph for my guess at an overall story that fits)…

The “HuffPoll model” version has Obama’s lead peaking on September 23-24 at 48.4 to 44.2 (with Obama’s support going up to 48.5 on September 25), and then falling steadily and rapidly after that. The debate was on October 3; Obama’s lead was down to 2.2% on October 2, 1.7% on October 3, 0.9% on October 4, and 0.4% on October 5, reaching just 0.3% on October 7 before heading back up — it’s not at 0.7%. But remember that Pollster is drawing trend lines, not simply averaging the polls on a particular day. So it’s possible, I think and if I’m understanding the way the model works correctly, that a real sudden drop might get smoothed out by the computer to look as if it happened over time (however, and again if I understand correctly, that would not happen with the RCP average that Drum also noted).

Ready for it to get more complicated?

Switch off Pollster’s model, and go to their “create your own” trend line, and now Obama’s lead peaks at 4.0%…but now that peak is September 17-20! The lead, by that measure, erodes a bit over the next week, but then falls quickly, and in fact that version now has Romney in the lead by 0.8%.

But that’s with “moderate” smoothing. That is, the trend line is set to not overreact to small fluctuations in the polls, which most of the time are picking up nothing but random variation. Use Pollster’s “more smoothing” option, for that matter, and the whole Romney October surge never happened; the trend line treats it as random movement and smooths it over, and Obama still has the same lead of about two and a half points that he’s had all year.

Assuming that’s not correct, what if we use Pollster’s “less smoothing” option? Call that one the Andrew Sullivan option — it’s designed to (over?) react to any hint of change in the polls. That one has Obama 49.0, Romney 45.3 on October 1, with the Romney surge coming just after that. Or, to be more technical, Romney starts moving forward after September 26, while Obama doesn’t start falling until October 2.

Now, remember, even a little “smoothing” of the trend line is going to make the original lead smaller, as the computer thinks that some of that lead was just random movement to Obama. Which might be true! On the other hand, it might not be. And, to tell the truth, I don’t know whether the settings they’re giving us are even capable of picking up an overnight 5 points swing even if it happened, or if it would automatically distribute it over several days.

I’d like to see Mark Blumenthal (or Simon Jackman) comment on all of that, but overall I think that there’s at least something to what Drum said: the debate happened in the context of a falling Obama lead. On the other hand, Sam Wang, who spends a lot more time looking at the polls than I do, thinks that it was really an immediate 5.5% post-debate bounce…I just don’t see that in the Pollster numbers.

My guess — and it’s really only a guess, but it fits pretty well with the numbers and with what we general know about voters, is this: Obama had a small lead before the conventions. During the conventions, Obama added to that lead, giving him about a 2-3 point edge, mostly by getting straying or low-information likely Obama voters to shift to him…and he also had a temporary bounce which then stayed in place for a couple of weeks thanks to a series of campaign and world events. The debate gave Romney a 3 point or so bounce, bringing him to a very slim lead, which is now receding. If nothing else happened, we would eventually see Obama’s lead settle back at 2-3 points — if the debate bounce was purely temporary — or about 1 point, in the relatively unlikely event that Romney really did permanently grab some straying or low information likely Romney voters. But of course events and campaigning keep happening, so it’s always very difficult to know for sure.

[Cross-posted at A plain blog about politics]

Jonathan Bernstein

Jonathan Bernstein is a political scientist who writes about American politics, especially the presidency, Congress, parties, and elections.