Buried in New York Times polling guru Nate Silver's latest blog post is a chart that should have Mitt Romney ordering out for some Xanax. The chart depicts Silver's statistical inferences from several multi-day tracking polls that include Thursday and Friday, the final day of the Democratic convention and the day after--in other words, the two days after Bill Clinton's barnburner of a speech. I'll explain below why I'd add one grain of salt to Silver's inferences, but first, here they are, in this chart's final column, which depicts the estimated swing in Obama's favor since Clinton's speech:
My two reasons for mild skepticism:
 I support Obama, and I'm a pessimist. OK, I admit that this reason isn't a paragon of rigor. So on to the second reason:
 As Silver himself emphasizes, there's a big challenge in interpreting multi-day tracking polls, because they don't release their day-by-day figures. So when, say, Obama's seven-day Gallup tracking number goes up for Friday (as it did), what that means is that the average Obama number for Sep. 1-7 is higher than the average Obama number for Aug. 31-Sep. 6. And the only highly granular thing you can infer for sure from that (if I understand this correctly, something that has a 78-percent chance of being the case) is that the Sep. 7 number was higher than the Aug. 31 number. And we can't say for sure what the Aug 31 number was, because, again, these polls don't release day-by-day numbers.
Here's the reason this matters: Obama's multi-day tracking numbers rose for Thursday, the day after Clinton's much-praised speech, and then reached an even higher level for Friday, the day after Obama's not-so-praised speech, and the day a weak jobs report came out. (These are the numbers released by the polling companies on Friday and Saturday, respectively.) It's tempting to look at those numbers and think, "Wow, Obama's numbers continued to rise immediately after his speech, and notwithstanding the weak jobs numbers." But, again, all we can say for sure is that his post-speech numbers were higher than his numbers had been a week earlier. (That's with the Gallup poll. Not all the tracking polls are 7-day--some are just 3-day--but the same type of uncertainty applies to all of them.)
So here's a scenario that, if I'm understanding this correctly, is entirely consistent with the data: Obama's numbers rose a lot on Thursday, after the Clinton speech, but then fell on Friday, after the Obama speech and the weak jobs numbers, though they didn't fall as low as they'd been some days earlier (7 days in the case of the Gallup poll, other periods for other polls). If that's what happened, then the convention bounce, as it currently stands, could be significantly less than the eight points suggested by that final column in Silver's chart.
So I would add my one grain of salt to this sentence of Silver's: "Despite a mediocre jobs report on Friday, there were no signs in the polls that Mr. Obama's bounce had immediately receded, as he gained further ground in the surveys that were released on Saturday [i.e., the surveys conducted through Friday]." True, there were no visible signs of slippage, but it's in the nature of these multi-day tracking polls that actual slippage could be concealed by numbers that keep rising.
All that said, there's no denying that these tracking poll numbers will get filed under "bad news" by the Romney campaign, and deservedly so. Even the most pessimistic of pro-Obama pessimists should feel a mild lightening of spirits.
[Update, 9/9, 11:20 a.m.: Adding strength to Silver's analysis--and sapping strength from my grain of salt--is the latest Rasmussen tracking poll, which came out this morning, after Silver's post, and includes Saturday polling numbers. It has Obama up by four percentage points, higher than the two percentage points of the previous day's Rasmussen poll. (And Rasmussen is a three-day tracking poll, so this latest number reflects entirely post-Clinton-speech findings.)]
We want to hear what you think about this article. Submit a letter to the editor or write to email@example.com.