Wednesday, October 19, 2016

Poll Vaults

     Yes, yes, I’m sure that title has been used before. I like it anyway.

     I’ve been concentrating on matters fictional for the past few weeks, as my regular Gentle Readers could tell from my recent emissions. That choice of focus has had two conspicuous results:

  • For the first time in far too long, I’ve actually finished something substantial;
  • My overall mood and outlook on life are better than they were previously.

     So I’m dubious about whether I should pay much attention to the election or any of its peripheral phenomena. But duty calls in a voice of iron.

     Political pollsters are a lot like other kinds of consultants: They do their level best to produce “results” that will please the man who pays them. In the case of opinion surveyors, the end to be attained is one that will persuade the client that what he’s saying or doing is popular. In the great Robert C. Townsend’s memorable phrase, they borrow your watch to tell you what time it is...and then walk off with it.

     Polls about the upcoming presidential election are “all over the place,” specifically with regard to short-term opinion swings. They disagree as wildly as I’ve seen surveys disagree in decades. The ones being conducted by supposedly disinterested professional polling organizations have exhibited no more stability than the ones openly paid for by the candidates and their respective parties.

     Is there anything to be gleaned from this? I’m not sure. The only certainty is that for as long as political conflicts will continue, the pollsters will do their damnedest to persuade their audiences that they’ve got their most sensitive digits on its pulse. Oh, and their fees will go up, too.

     Among Donald Trump’s allegiants are many who like him for his bluntness. A penchant for bluntness always comes at a price, of course: the likelihood that one will at some point say something he really should have kept to himself. The “Grab her by the pussy” foofaurauw is a perfect example.

     I have a little of that myself. It’s been the cause of quite a high percentage of the conflicts in my life. But I’d rather be known as someone who speaks his mind plainly and somewhat fearlessly than for a milquetoast afraid of ever “offending” if never giving “offense” were possible in our hypersensitive age.

     The harping of the Democrat-friendly media, and their affiliated polling organizations, on the “Grab her by the pussy” episode and on the subsequent flood of women “coming forward” to accuse trump of Neanderthal behavior toward them, powerfully illustrates the dynamic I mentioned above. People – except for Establishment figures determined to see Trump defeated – weren’t reacting to the “Grab her by the pussy” schtick the way those media wanted. Their paid pollsters knew it, and could see the probable consequences for their future employability. So they beat the bushes in search of incidents that would substantiate their “Trump treats women with contempt” message. A few women with less sense than is good for them have agreed to assist.

     However, due to objective refutations of the majority of the accusers and the outpouring of testimony from women of great beauty to the effect that “Trump was always a perfect gentleman toward me,” those attempts to prick popular opinion have failed as well. The opinion swings the pollsters wanted to see have been far smaller than they liked. So they knew they needed another tactic.

     Opinion polling depends on sample selection and the design of survey questions. Not enough attention is paid to either consideration, but at the moment the more important one is sampling.

     Let’s imagine that a pollster wants to produce a result that his Democrat-aligned client will look upon with pleasure. (Not hard, is it?) How should he engineer his sample toward that end?

  1. He should attempt to over-represent:
    • Unmarried women;
    • White-collar workers;
    • Government employees;
    • Atheists and agnostics;
  2. ...but under-represent:
    • Married women, especially married mothers;
    • Blue-collar workers;
    • The self-employed;
    • The religious, especially Christians.

     The resulting demographic balance will almost always favor the Democrats. But we’re not quite done yet. Will the persons in the sample vote? A few further questions are required:

  1. “Are you registered to vote?”
  2. “Have you voted in the last N elections?”
  3. “Do you expect to vote in the upcoming election?”

     These allow the pollster to make certain representations about “registered voters” and “likely voters.” To reinforce the seeming probity of the result, the pollster will attempt to contrive a sample whose party-affiliation proportions replicate those in the general populace.

     The opportunities for a palmed card here are several:

  • Does the pollster use voter registration lists to select his sample, or does he rely on what the persons surveyed tell him?
  • What is the “best” value of N for the pollster’s purposes, and should the answers to the question be treated as sincere and accurate?
  • The answers to the third question are usually probabilistic: “Certain,” “Very likely,” “Somewhat likely,” etc. Which groups should the pollster aggregate into “likely voters?”

     Moreover, we encounter the “toothbrushing trap” here. Our dentists tell us we should – must, really – brush at least twice a day, and preferably after every meal. Surveys of brushing habits regularly suggest that that we’re doing as we’re told...even though most Americans only brush once a day, shortly after rising from their slumbers. Despite the absence of consequences from that “disfavored” answer, people are loath to “look bad” by giving it. (Of course, when there are possible consequences – consider the polls that preceded the 1990 Nicaraguan election that elevated Violeta Chamorro over Daniel Ortega – the answers will be biased heavily in favor of the ones likely to reap the result the customer would prefer.)

     Pollsters know all of this – and use it.

     There’s much more to be said about polling and why one should always approach the result of an opinion poll with caution, but that’s enough for the moment. The most important message is plain: There’s a variety of techniques by which pollsters can engineer the result they desire. In the usual case, the desired result is dictated by the source of the pollster’s funding. So be on your guard: not too ready to believe, and ever on the alert for indications that all is not what it seems.

     “Believe nothing, no matter where you read it, or who said it – no matter if I have said it! – except it agree with your own reason and your own common sense.” – Siddhartha Gautama, a.k.a. the Buddha


Malatrope said...

It's even simpler to produce a fake-yet-true poll by just throwing out the samples that don't fit what you want to end up with. If the pollster doesn't tell anyone that they had to call 600 people to get the 400 sample population they used, there isn't any way to refute them!

Unless the pollsters are honest (I'm not going to go into a Diogenes riff here) the published polls are completely fake.

Remus said...

Related: betting odds heavily favor Hillary Clinton to win, if stated in dollar totals for each side. If stated by number of bettors on each side, Donald Trump is heavily favored, 65% if memory serves. You needn't guess which measure the media uses.

daniel_day said...

Sean Hannity just said on his radio show that UK bookies have switched from favoring Clinton to favoring Trump. He also said that the UK bookies have usually called US elections accurately.

Anonymous said...

Since the polls are a joke anyway (political or otherwise) I treat them as time to widen the bell curve a bit. It can be great fun.