Notes on the Atrocities
Like a 100-watt radio station, broadcasting to the dozens...

Thursday, June 12, 2003  

An astute reader pointed out a serious flaw in the polling data I reported on a coupla days ago. I'm talking about the NPR poll (.pdf file) I described as dispiriting. Well, turns out there's a durn good reason why: it polled mostly conservatives.

If you read down into the demographic data, it becomes pretty interesting. NPR asked the question, "Do you consider yourself to be a conservative, moderate, or liberal?" Only 16% fessed up to being liberal. Forty-three percent called themselves conservative. That's nearly three times as many, for those of you scoring at home. Asked whom they voted for in the last election, 51% voted Bush, 41% for Gore and Nader. (And of course, no one's forgetting Gore got more votes than Bush in the election.) So, given that the deck was heartily stacked in favor of King George, it's not quite as surprising--or dispiriting--that the numbers looked so bad for Dems.

On a related note, this is further evidence that polling data are no longer particularly reliable. In the past two elections, the polls (both before for and, notoriously, after the election) proved to be inaccurate. Why? Well, it's not a statistical issue--that is, no one believes that someone ran the numbers wrong. But let's look at the NPR poll. According to pollsters, it had a plus or minus of 3.64%. This is a mathematical over/under, based mainly on the number of warm bodies they found to talk to. But a quick glance at that demographic data point out that the sample doesn't look much like the US (and fortunately, we have a reliable sample with whom to compare NPR's--actual voters).

It's a question of sampling. I've heard a couple theories that are convincing logically (though I haven't seen any studies). 1) Cell phones. They're unlisted, and a huge number of Americans use them, meaning that although they're part of the population, they're not in any phone sample. 2) Only a small part of the population answers the phone, or (if they don't recognize the person on the other end) stay on the phone. Telemarketers have made pariahs of anyone non-personal callers, and so only a few people are willing to be surveyed. I think Americans are increasingly suspicious of pollsters, and they may not be as truthful as they once were.

In any case, for a country that conducts politics by way of polling data, we sure have an imprecise way of measuring voters' positions. In fact, so skewed is that NPR study that I wouldn't even call it valid.

posted by Jeff | 12:59 PM |
Blogroll and Links