« A little gratitude would be nice |
Blog Home Page
| Polls changing »
November 5, 2006
A premature election orgasm
As a scientist with thirty years of experience working with data just like election polls, I think the Left is having its victory orgasm way too soon. The pollsters are using the same assumptions since the famous Truman upset election of 1948. They are now questionable or plain wrong. The biggest question is: How wrong are they?
The key to scientific polls is random sampling of target groups, in this case (a) actual voters (b) Democrats vs. Republicans. If you're not properly sampling those groups your poll results are going to be wrong.
For fifty years or so, poll takers could find enough people in the the target groups to make reasonably good prediction.
What has changed today is a massive revulsion among conservatives and Republicans against the media and their handmaidens, the pollsters. The polls now have to guess at (a) and (b) among the conservatives. You can't check on your sample if your target group won't answer. Twenty years ago Black opinion would be consistently off, because Blacks were suspicious of White pollsters. Conservatives are the Blacks of today, as far as alienation from the media is concerned.
In technical sampling language, the drop—out rate or sample mortality is wildly different on the Left and Right. There are no statistical techniques that can make up for a lousy sample.
The Democrats are wildly overconfident about their coming revenge. Headlines here and in Europe are celebrating a bloody defeat for George W. Bush. Leftwingers are bubbling with eagerness to talk to pollsters.
On the Right, Republicans can't stand the pollsters, who are blamed for their constant push—polls on behalf of the Left. Their predictable reaction is to avoid pollsters like the bird flu that never quite met the media hype. Today you would almost have to waterboard a lot of conservatives, one by one, to get them to talk to pollsters. You can't get a random sample if one of your target groups won't answer. So the data will be wildly skewed.
Sampling bias would be serious if the pollsters missed even ten percent of the actual GOP voters. My unscientific guess is that they are missing much more than ten percent of conservatives, maybe 20 or 30 percent.
Any scientific paper based on such wildly unlikely sampling assumptions would be rejected by journal editors, and never appear in print. The media just don't have anything approaching scientific standards. They are run by "activists," not scientists. They're already partying all over the international Left. The Islamist world is probably celebrating right along.
My advice: Watch the premature election orgasm as it happens.
You might enjoy the vote count.
James Lewis 11 5 06
Update: Rosslyn Smiths writes:
I agree with James Lewis, but I also think another factor has been at work with the polls.
Political polling has become a very big business and like every other type of service industry, pollsters thrive by keeping their customers happy. The single biggest customer for the political polls that get made public is the news media.
Keeping difficult customers happy has been known to challenge the integrity of even the most consummate professional. Consider the appallingly overstated assessments of Confederate strength that noted investigator Allan Pinkerton gave his customer, the ever timid General George McClellan, during the Civil War.
In my own experience as an accountant, I once uncovered significant misappropriation of partnership assets by a developer who had just received an opinion that the venture had been following proper accounting procedures from the reputable CPA firm he had hired. That opinion was not in error. Those accountants confined their opinion to the existence of documentation while I looked behind the documentation. I suspect the other accountants knew something was amiss, but they had not been paid to label their client a crook.
Increasingly pollsters have been hired to generate political news stories of the 'horserace" variety in lieu of the more traditional issues oriented election coverage. These pollsters often report a last minute surge to the Republicans. I suspect such surges have less to do with changes within the electorate than with pollsters protecting their reputation with the general public by adjusting their final numbers away from the results their media customers' desire.