Democrat-heavy sample nets better news for Trump
- Among the same Democrat-heavy sample: 26% say they voted for Romney in 2012. 42% say they are leaning toward Trump in 2016.
The following is a media news analysis[hr]
Another poll; another way to spin.
Earlier this week, I showed how the reporting on a Bloomberg poll could be skewed to make results look more or less positive for a given candidate.
Today, we look at a Washington Post/ABC News poll that also purports to show a widening Clinton lead over Trump – by 8 points: 50% to 42%. This may well be the case. However, looking at the poll sample numbers, there's some relevant context not reported in news stories.
The poll interviewed 10% more people who identify as Democrats (33%) than Republicans (23%), with the largest group (36%) calling themselves independent. So with 10% more Democrats than Republicans questioned, Clinton leads Trump by 8-points.
Even more interesting, the same Democrat-heavy sample favored Obama by a larger 10-point margin over Romney in 2012: 36% Obama to 26% Romney (with 32% saying they didn’t vote). We know this because the poll asked respondents how they voted in 2012. So today, Trump is outperforming Romney with the exact same Democrat-heavy sample of voters.
[quote]In other words, the same Democrat-heavy sample of Americans that gave Obama a 10-point edge in 2012, gives Clinton a slightly smaller lead, 8-points, in the latest Washington Post/ABC News poll.[/quote]
Further, this particular sample has not proven to be representative in the past. Of those who said they voted in 2012, they gave Obama a hefty 15-point edge over Romney: 54% to 39%. But the actual general election was a much tighter 4-point race: 51% Obama, 47% Romney. So Romney ended up performing 8-points better and Obama 3-points worse than this Democrat-heavy sample group reflected.
[quote]Among the same Democrat-heavy sample: 26% say they voted for Romney in 2012, 42% say they are leaning toward Trump in 2016.[/quote]
One polling expert told me there's typically no disclosure or adjustment made when random sampling turns up significantly more respondents identifying with one party over another. There's no way to know how that will match up with the population that actually turns out to vote. "It's a judgement call," says the expert. Finding substantially more respondents identifying with one party over another could be an indication that the makeup of the electorate is changing, she adds.
The Washington Post/ABC News poll does what the Bloomberg poll did in pressing respondents to pick a candidate even if they initially stated they didn't know if they were going to vote or who they would vote for. [Bloomberg added in the "leaners" when reporting the totals in an article, even though the respondents were answering a different question than "for whom would you vote." This gave Clinton the appearance of a slightly larger lead than she actually had in the Bloomberg poll.] The Washington Post/ABC News poll seems to take this a step further: they represent the two questions "for whom would you vote" and for whom would you "lean" as if they were a single question, though they were undoubtedly asked as two separate questions. See question #2. For some reason, they chose not to separately publish both answers, and only provided the combined total. Does that favor Clinton, as in the Bloomberg poll, whereas without the " leaners," Trump is closer? A query to the Washington Post polling department was not answered by publication time.
There’s another point worth noting. The pollsters asked a series of four questions raising negatives about Trump: "goes too far in criticizing," "a problem with respect for for people with whom he disagrees," "criticism of Muslim-American family whose son was killed while a U.S. Army captain in Iraq," "biased against women and minorities." But they asked just one question raising a negative about Clinton: "too willing to break the rules." One could envision other questions more comparable to the Trump questions such as: "considers herself above the law in light of the FBI findings about her email servers," "committed perjury giving incorrect testimony to Congress," "demonstrates hypocrisy on women's rights considering her husband's background and her response to it," and "jeopardized national security with conduct the FBI called extremely careless." But these questions weren’t asked. This means there are a number of potential negative Trump points to highlight when reporting on the poll, but fewer potential negative Clinton points available.
None of this is to suggest the headline of this poll won’t prove to be entirely accurate in the general election. Poll trends over time are typically fairly accurate predictors. But this poll is most likely to be an accurate predictor, it seems, in a race where 10% more Democrats vote than Republicans… and that remains to be seen.