Trump: I just can't believe this "dirty" WaPo/ABC poll

Donald Trump may have some legitimate polling issues, but he’s also got a legitimate gripe in this instance. On Friday, the new Washington Post/ABC national poll showed Trump trailing by twelve points — which also happened to be the gap between Democrats and Republicans in their sample of adults. For registered voters, the gap was D+10. That led the Republican presumptive nominee to take his objections to Twitter:

https://twitter.com/realDonaldTrump/status/747102432862273536

https://twitter.com/realDonaldTrump/status/747206160806219776

In response, a number of people accused Trump and his supporters of trying a new version of “unskewing the polls,” the 2012 project that attempted to prove bias in presidential-race polling. That project, however, was much more than just criticizing sample skew. Four years ago, people on the Right were so convinced that the turnout model would be more like 2010 than 2008 that we recalculated polls based on an even-up model — even those with reasonable samples, like D+7 or so. It contributed to significant self-delusion about Mitt Romney’s prospects on the eve of the election, and is a project that should be permanently buried no matter how many problems we see with polling.

This complaint doesn’t resurrect it, though. While the WaPo/ABC sample and poll results aren’t “dirty” or dishonest, the presentation seemed a bit misleading. Here’s the second paragraph:

Following a month of self­inflicted [sic] controversies, the survey shows that support for Trump is plunging, including among fellow Republicans, propelling Democrat Hillary Clinton to a double-digit lead nationally. The poll reveals fresh doubts about Trump within his own party just three weeks before Republicans convene in Cleveland for their national convention.

Trump’s performance might have had some impact on those changes, but the Post neglected to note [see update] that the difference might also have come from a change in its sample composition. The previous poll, in which Hillary led 48/42, had a D+8 sample, while this one had a D+12 — and that might account for most of the change, with Hillary gaining three points and Trump losing the same amount. It claimed significant motion where perhaps little really existed.

There are a couple of points to note, though, about this series. First, the Post and ABC do disclose the historical trend of their sample compositions in each polling document, even if their reporters tend to ignore it, so it’s pretty easy to keep that context in mind. Second, inconsistent sample composition has always been a problem in the WaPo/ABC series, and that’s why any polling analysis has to keep that context in mind in the first place. It’s not dirty or dishonest; it’s just not necessarily reliable, especially as spun last week.

Third, and most importantly, national polling (especially among general-population adults) is only good for broad measurement of trends. State-by-state polling is what matters in presidential elections, and the news is more mixed on that front — but still not terribly good for Trump. The new CBS battleground polls show Hillary Clinton maintaining an edge that could shut the GOP out of the White House:

Hillary Clinton holds narrow leads over Donald Trump across a number of key states of Florida (up three points, 44 to 41 percent); Colorado (Clinton 40 percent, Trump 39 percent); Wisconsin (Clinton up 41 percent to 36 percent) and North Carolina, which has flipped back and forth between the parties in the last two elections, where it’s Clinton 44 percent and Trump 42 percent. …

And much of the vote appears locked in already: the bulk of those not voting for Clinton say they will not consider her, and the bulk of those not voting for Trump say they will not consider him. In Florida, sizable numbers of voters are voting in opposition to a candidate they don’t like: Forty-eight percent of Trump’s voters are backing him mainly to oppose Hillary Clinton, and 32 percent of Clinton’s voters are with Clinton in order to oppose Trump. That opposition effect works for both candidates, but voters say each party may have lost opportunities. Fifty percent of those not backing Clinton say they might have considered a Democrat this year had the party not been selecting Clinton as its nominee, and 47 percent of those not for Trump say they might have considered a Republican, but won’t support Trump.

All four of these states are featured in my book Going Red, and the Wisconsin results might be the most disturbing here. Wisconsin should give Trump his biggest opening, with its heavy population of working-class white voters and the bitter divide in politics in the state over the last five years. If Trump only gets 36% of the vote at this stage, that might be a big alert to the lack of traction among those demographics outside of the Republican base — and without any outreach to other groups, the Rust Belt strategy looks as though it’s at serious risk of failure, at least at this stage.

CBS’ Anthony Salvanto makes a similar point, although he notes it’s a problem for both parties:

While some Republican leaders are at best lukewarm about how Trump is running his campaign, rank-and-file Republicans in these states are largely okay with it, and many don’t care whether or not Trump listens to party leaders. This was an anti-establishment sentiment we saw repeatedly through the primary season, too. However for the larger electorate, and especially among independents, this campaign is leaving them a bit distant. Independents say that the Democrats nominating Clinton hasn’t made them think better of the Democratic party, and that the Republicans nominating Trump hasn’t made them think better of the Republicans.

The WaPo/ABC poll isn’t really Trump’s biggest problem, even if he does have a legitimate gripe about it. It’s his low numbers in these key states.

Update: As Karl notes, the Post actually did address this point, but later in the article:

https://twitter.com/justkarl/status/747469273186992129

So yes, they did address it — but quite a way down from their presentation of this as significant movement in the second paragraph. It’s also a little difficult to buy the idea that a four-point shift in the sample towards Democrats accounted for “less than half” of Hillary’s three-point pickup, especially since Hillary holds Democrats a bit better than Trump holds Republicans. The point here is that this looks like an artificial movement on the basis of sampling differences, not an actual trend based on political movement — and most of the other polling series during this period suggests the same.