Oh man. She has a duty to spin for her boss, sure, but a professional pollster resorting to “silent majority” arguments? Yeesh. Why not just stick to the old chestnuts like “the race will tighten in September” (which is likely true) and “the only poll that counts is the one on Election Day”?

She’s making a different argument here than Rush Limbaugh did yesterday. Rush’s claim was that pollsters aren’t even contacting Americans who haven’t voted recently but are quietly preparing to turn out for Trump this fall. Why he thinks they’re not contacting them is unclear. It’s more likely that they are contacting them but that some people who intend to register and vote for Trump haven’t done so yet, which would exclude them from tabulations of registered voters. But as I said last night, if that’s true then there’s peril in it for Trump. He doesn’t have much of a ground game and unlikely voters, almost by definition, might not be aware of registration deadlines. If there’s a hidden cohort of not-yet-registered Trump voters, they’d better get cracking on registering soon or else, to their great surprise, they’re going to be turned away at their polling place on November 8th. We’ll know one way or another by October, at which point most of the current not-yet-registereds should have registered and will start being counted in polls.

Conway’s not making a claim about unlikely voters, though. She’s suggesting there’s a “Bradley effect” happening, which she calls the “undercover Trump voter,” in which lots of college-educated Americans are lying to pollsters by telling them they support Hillary because they’re too embarrassed to admit they really back Trump. (The idea that vast numbers of college grads might feel so ashamed of their Trump support as to not want to confess it to a stranger suggests a problem in its own right.) I took a shot yesterday at explaining why that’s implausible, noting that the evidence is thin that the “Bradley effect” has ever actually existed. But let’s ask the pros. Harry Enten of FiveThirtyEight compared Republican primary polling to the results of 34 primaries and caucuses this year looking for an “undercover Trump voter.” After all, presumably there are also college-educated Republicans who felt too embarrassed to admit their Trump support to pollsters this past spring. If so, it should be the case that Trump consistently did better on Election Day than he did in the primary polls as the “undercover Trump voters” pulled the lever for him. Enten’s conclusion: He didn’t.

Trump did worse than the polling forecast in 19 states; he did better in 15 states. That hardly suggests that Trump outperforms his polling. Still, the difference isn’t so great that we can say Trump usually underperforms his polling. It’s a fairly even split, with Trump missing his average poll by just 1 percentage point in the median state.

By and large, Trump performed about as well as the polling suggested he would. The exception to that was the mid-Atlantic primaries in April, when he performed several points better than the polling indicated he would, but Enten reasons that that wasn’t due to “undercover Trump voters,” it was due to the fact that Cruz was unpopular in that region and Trump was picking up momentum as the likely nominee. He wasn’t winning “secret” voters, he was winning plain ol’ undecideds. Enten then went further and looked at whether Trump did any better in polls that used live interviews with a human pollster versus automated polls that didn’t, reasoning that “undercover Trump voters” would be more willing to confess their true preference to an automated system. Result: Again, no real difference. In fact, automated online polls tended to overestimate Trump’s performance on Election Day. Ted Cruz’s team, which had an obvious interest in measuring Trump’s support accurately, told Politico in June that they saw no evidence of an “undercover Trump voter” in their own surveys during the primaries.

What about the possibility, though, that general-election voters are more inclined to lie about their true preference than primary voters were? There may be moderate Republicans and right-leaning independents and Democrats who preferred other candidates in the primaries but are quietly and reluctantly prepared to pull the lever for Trump over Hillary now as the last remaining conservative-ish option. Those people might lie to a pollster by telling them they’re backing Hillary because they’re embarrassed by Trump, no? (The idea of people with a reluctant Trump preference turning out en masse for him doesn’t sit easily with Rush’s theory of an ardently pro-Trump silent majority preparing to storm the polling places to elect him.) HuffPo tried to answer that, once again by looking at live interviews versus automated polls — and via another metric:

“[D]ata from HuffPost Pollster’s tracking of the presidential race suggest that support for Trump varies little between live-interviewer surveys and those conducted online or using automated calls.

[Democratic pollster Anna] Greenberg noted one other way of testing whether Trump is suffering from voter embarrassment. And here too, she said, the phenomenon seems overstated or non-existent.

“I also look at how the Trump vote lines up with partisanship,” she said in an email before the conventions. “In many of the places I’m polling, Trump’s vote is lining up with the percent who call themselves Republican.”

If anything, Greenberg said at the time, it was Hillary Clinton who was “underperforming relative to partisanship.”

That last bit is interesting. Could there be an “undercover Hillary voter” too? Her favorable rating is smoldering garbage, just like Trump’s is. There may be moderate Republicans and far-left Sanders fans out there who simply can’t admit to a pollster that they’re going to support a corrupt establishment Democrat yet suspect that, once they’re in the booth and feel compelled to choose between her and Trump, they’ll pull the lever for her. How many of those voters are there compared to undercover Trump voters?

One more point from Nate Silver. If there’s an “undercover Trump effect” now, presumably there have been “undercover effects” in other recent elections. Trump may be sui generis as a presidential candidate but there have been candidates at the state level like Christine O’Donnell or Todd Akin whom educated people were “supposed to” be ashamed of supporting and might lie about backing when asked by a pollster. (The same goes for Anthony Weiner in New York’s last Democratic mayoral primary.) Is there any evidence that O’Donnell or Akin, both of whom lost badly, significantly outperformed their polling in Delaware and Missouri, respectively, on Election Day? Silver himself notes that Democrats made this same argument two years ago about the impending GOP midterm landslide: Supposedly, pollsters weren’t capturing the “silent majority” of Democratic voters because they were somehow overlooking minorities’ eagerness to vote that fall. You know how that turned out. Silver also notes that Michael Dukakis and Walter Mondale were known to point to the size of their crowds (and a “hidden women’s vote” in Mondale’s case) down the stretch to counter claims that they were headed for blowout defeats. You know how that turned out too. The only thing that elevates Conway’s spin here to something worth taking semi-seriously is that she’s a pollster herself and claims that her own polling is able to detect the “undercover Trump voter” — but she won’t say how. If that’s true, the best thing she could do for her boss would be to share her methodology with the public. Nothing would give him and his fans a shot in the arm like watching public pollsters adjust their own models en masse to account for these “hidden votes.”

Exit quotation from Trump Super PAC chair Ed Rollins: “If we’re sitting here three weeks from now after Labor Day and it’s in the same position, we’re gonna have a hard uphill battle. He’d lose badly today.”