SKEWING WITH THE POLLS

Skewing with the polls.  That double-entendre spelling says a lot about what some mainstream media venues are doing in an effort to “prove” how disliked Donald Trump is.

From Christian Datoc’s article at dailycaller.com:

Two polls released Tuesday — one from ABC and a second from CNN — tout Donald Trump as being the most unfavorable incoming president in modern history — yet on second look, the data is clearly boosted by the pollers’ decision to oversample Democrats.

According to Gallup, 28 percent of Americans identify themselves as a Republican, while 25 percent identify as a Democrat.

Despite this information, eight more percent of participants in both the ABC and CNN polls identified as Democrats, leading to an 11 point swing in partisanship breakdown off the national average.

FYI:  here are the Gallup party affiliation data  – not just for its most recent wave of interviewing, but for every wave since the election:

Republicans Independents Democrats
  % % %
2017 Jan 4-8 28 44 25
2016 Dec 7-11 28 39 29
2016 Nov 9-13 27 40 30

Does it make sense to use only the latest data, showing Republicans with a 3% advantage?  Sure it does:  that is the closest to when ABC and CNN did their polls.

But, just to give every benefit of the doubt, let’s average all three post- election waves.  Doing that gives the two parties a virtual tie (Democrats “win” by 0.3%).

Knowing this, can you give me any basis at all for ABC and CNN polling that gives Democrats an 8% advantage – which, of course, materially affects the final polling data?

And, speaking as someone with many years in both quantitative and qualitative research, let me assure you that this is not the only way they can, er, “skew” with these data.

If, for example, interviewing was disproportionately conducted in major cities, the percentage of Democrats and Republicans might be correct, but there would be an over-representation of more liberal/leftward members of each party, versus the more moderate/conservative people in smaller cities and rural areas.

Given what was done with the Democrat/Republican mix, do you have any confidence the geographic mix was any better?  If so, why?

And if this is what our so-called “neutral” media do with polling data, do you have any confidence that the neutrality of their news reportage is any better?  Again – if so, why?

Little wonder that so few people trust them anymore.

Why should they?

9 Comments

  • Just heard on the radio:

    Polls show Obama ends his term with the highest popularity of any previous President.
    And Trump begins his Presidency with the lowest popularity.

    Could these also be skewed ?

    • Donald Trump won an electoral landslide over Hillary Clinton. Barack Obama campaigned heavily, and specifically, on the premise that Clinton’s presidency would be an affirmation, and continuation, of his legacy. That should give you an idea of the regard I have for these polls.

    • Grave concern or derisive laughter: which of those two do you figure as Donald Trump’s reaction to the poll data?

  • 538 politics, a website I respect deeply on political polling, published some interesting analysis on this topic (note that they assigned a 29% chance of Trump winning the electoral college the night before the election, much higher than I saw in many other places). They place blame not on the pollsters but instead on the media who have not done a good enough job explaining margins of error. See full article here, I think you’ll find it interesting:

    https://fivethirtyeight.com/features/can-you-trust-polling-in-the-age-of-trump/

    • I think Ken explained it rather well when he wrote about the huge over sampling of D’s and not knowing the geographic area polled.

    • Mike – Thanks for the heads-up. I’m very familiar with Nate Silver and 538 politics. And, having read the article, I see where you’re going. But let’s be honest and acknowledge that a 29% probability – which dropped from a higher percent in the last day of the campaign, thus the momentum was downward – is pretty thin gruel to declare polling superiority. That’s a little like saying a guy who is batting .227, down from .231 a couple of days ago, is a good hitter because the other guys are batting .214. I do agree that if you take into account the statistical margin of error and assume the error was in Hillary Clinton’s favor (therefore Trump, in actuality, was doing better than the data indicated) a case can be made that several “wrong” polls were, in fact, more accurate than they looked. Not exactly something I’d want as my research epitaph, though…

      • Agree- many of the traditional pollsters got something wrong. Because 538 doesn’t actually conduct polls of their own, I view them as a lot more objective- they weight various polls based on sample size and statistical soundness. Having listened to so many of their podcasts throughout the election season, I can tell you that they did a much better job highlighting the very real chance of Trump winning than I saw in most other outlets (I guess that’s a function of their expertise on the topic). Because their probabilities were based solely on a large variety of public polls that had Clinton winning, I don’t think the model could have possibly predicted a Trump victory at that point without screwing with the model. Interestingly, something that did worry me as a Clinton voter was that the IBD/TIPP poll did get a very high grade from 538, so that definitely made me think a Trump victory was perhaps more likely than 29%.

Leave a Reply

Your email address will not be published. Required fields are marked *