Zehr on polls:
The Game of Poll-O
It is necessary to understand quantum physics in order to grasp the mechanics of modern electoral prognostication. See, during the campaign each pollster lives in a little universe all his own, where he gallivants about with boldest abandon, making up the rules as he goes. Oh yes, polling is really more of an art than a science. Forget all that "margin of error" folderol with which the prognosticating prestidigitator engages the attention of the audience to distract them from the legerdemain he is secretly working. On those rare occasions when the results fall within the margin of error it is usually due to pure, dumb luck. The outcome is determined by the input, that is to say, by the makeup of the sample population polled. And the method for selecting the sample is a closely-guarded trade secret, just like the ingredients of Coca-Cola. The situation is very much as that stated in "The Space Child's Mother Goose," a classic of cynical wisdom from the early, idealistic 1960s (not the later, decadent part of that decade):
"Hey, diddle, diddle,
Dissemble the middle,
The premise controls the conclusion ..."
Of course it is theoretically possible to predict the outcome of an election with almost diabolical precision, provided one uses a representative sample. And how, pray, does one obtain a representative sample? Well, that's the tricky part. If you happen to be the typical mainstream pollster you start by asking your wife whom she intends to vote for, then you ask her to ask her mother, and go on from there. For example, you might pause to query the wino who sleeps in the cardboard box in the alley down the street and hand him a few questionnaire's to pass around to his pals (those who can read, anyway).
Okay, they do it a little more systematically than that, but the results they obtain are almost as fluky as those one might expect from such a procedure. For example, they persist in polling on weekends, which anybody who has been paying attention must know by now is a sure-fire method for over-representing Democrats. (It seems that Republicans are more inclined to do family things on weekends, hence they are less likely to be home when the pollster calls). So is daytime polling (it overestimates the layabout vote as well as the hormone-crazed Oprah watchers) and failure to make follow-up calls.
If the pollsters all know that such procedures are flawed why do they persist in using them? Because the flaky procedures are usually cheaper and, more important, the phony results they obtain thereby fit their none-too-well hidden agenda. Just consider the effect they have had on many of the less perceptive conservatives: [Tap, tap] "Barman, a double hemlock, please -- with a twist."
Mind you, most pollsters do this sort of thing for a living, so they must consider their reputations. Have you ever wondered why the polls tighten so noticeably in the final week or so? That is when the poll takers begin to get serious -- the final one is for the record. Until election day all those poll figures are mere theoretical possibilities, but then, in that defining moment, "Tse vafe eqvation collapses, und ..." the alternate universes vanish in a puff of smoke [*paff*] leaving only the one we like to think of as "the real world."
"Nice try," say the skeptics, "but what about the recent Zogby poll that showed Bush trailing by 8?" The subtext is that Zogby has been spot-on in calling the results of the last few races. That may have been true in '96, but Zogby had a lean and hungry look in those days. Nobody had heard of him. That is to say, the mainstream newsies all pretended that they hadn't heard of him because their job requires that they push the poll commissioned (at great expense) by the company that employs them. Oh yes, there is a business aspect to reporting the news, you know. Take away all the high-flown imagery with which they glorify themselves, and what journalists are actually seen to be doing is selling soap (or whatever). That is why it is rare for a mainstream pundit to hang his sage assessment on other than a single poll. If your company is Ford it isn't smart to drive a Chevy.
Zogby made his bones in '96 using methods that were quite well known to others in the trade who eschewed them, but now that he is one of the boys, quoted for pay by mainstream news outlets, he seems content to use the same tricks as the others -- polling on weekends, for example. One must be competitive -- it's a business, after all. Besides, Zogby is a Democrat. There is no reason to suppose that his agenda differs from that of the others. What if he did show Dole a mere 8 points behind instead of 20? Where's the harm in that?
The bottom line would seem to be, if it's accuracy you want, go with the new kid on the block. That would direct us to such surveys as Rasmussen's Portrait of America (POA) poll and the Battleground poll (or Bat-poll, as I prefer to call it). But both of these surveys have been around for a while, what's new about them? They appear to be virtually unknown to the mainstream press, for one thing. One might say that they are getting the treatment meted out to Zogby the last time around. That might seem paradoxical, since both surveys have demonstrated their accuracy in past elections. Also, both are tracking polls -- one can follow their progress from day to day and form one's own opinion as to whether they are measuring actual trends or merely playing shuffleboard with their samples. They also use larger samples than the polls preferred by the major media which affords a higher degree of confidence in their accuracy.
"You're only saying that because both polls have shown Bush leading most of the time," scoff my critics. Believe that if you wish -- the truth is, both surveys use greater care in selecting their sample and that can make all the difference. Besides, they have that lean and hungry look. To a true-believing Democrat, such polls are dangerous.
At midweek the tracking poll in Gallup's alternate universe showed "Vice President Al Gore's lead" shrinking. Meanwhile, off in another universe, the Bat-poll showed "Texas Gov. George W. Bush's lead" continuing to grow, according to the Washington Times. But, according to Gallup, Gore now leads 44-to-48 percent, half the 8- point lead they gave him the preceding Saturday, while the Bat-poll favors Bush 41-to-37 percent, twice the 2-point lead they gave him on Monday. At least the trends agree, even though they seem to apply to different worlds.
As Greg Pierce noted sarcastically in the Times, "Bush's recent surge in the polls came just after news reports over the weekend that actor Alec Baldwin said he would leave the country if the Republican presidential candidate should prove victorious." Could this explain the shift in the big-mo? Why not? It makes about as much sense as anything else I've heard lately about this election. (By the way, who is Alec Baldwin and why am I supposed to care where he lives? I stopped keeping track of Tinseltown toads years ago. Does Alec Baldwin, whoever he is, care where I live?)
The POA poll, based on a three-day rolling average, taken on Sept. 13, 14 and 16 showed Bush ahead with 44.1 percent to Gore's 41.4 percent. As for the trend, a Bat-poll released at about the same time showed Bush with a "positive" rating of 59 percent and creeping upward, while his negatives were inching down to 32 percent. Over the same 5-day period, Gore's positive rating dropped from 57 to 55 percent, while his negatives increased 2 percent to 34 percent. All of which suggests that the race is pretty tight at the moment, with Bush holding a slight edge. This is about what the candidate's managers had expected, based on statements made by them before the convention. Of course, a lot of Bush supporters, dazzled by the fool's gold of phony polling figures that showed their man 17 points ahead, discounted these cautious appraisals at the time, only to fall into deep depression when they turned out to be accurate.
Meanwhile, CNN was touting the house poll (CNN/USA Today/Gallup) that showed Gore pulling 49 percent to Bush's 42 percent. A later poll released on the 19th showed Gore with 48 percent to Bush's 43 percent. The sample was taken among 690 "likely voters" (as contrasted with 1,000 likely voters sampled by the Bat-poll) giving it a 4 percent margin of error. Why they are content with such a large MOE when most mainstream journalists seem convinced that this is some kind of magic margin below which the poll results are invalid is another of life's little mysteries. Perhaps they just don't wish to waste too much money on poll figures they know to be garbage.
Actually, the dirtiest little secret about the CNN/USA Today/Gallup survey was revealed by Rush Limbaugh last Thursday. Their sample is selected to contain 39 percent Democrats and 29 percent Republicans. The rationale for this (apart from showing Gore to be "leading") has not been explained. The percentage of Republicans and Democrats among actual voters is usually about the same, with Republican voters having a slight edge. In other words, the voters are split about evenly among Republicans, Democrats and independents. Thus CNN, et al., would appear to be over-polling Democrats by about 5 percent. Could that be right? Try this: subtract 5 from Gore's 49 percent and add it to Bush's 42 percent. The result is, Bush 47 percent, Gore 44 percent. Bush 3 points ahead? Why that looks kind of like the results the POA poll cited above. Results of the Bat-poll released Friday show Bush continuing to pick up support. According to them, Bush now leads Gore by 5 points, 43 percent to 38 percent. |
