Why Some Pollsters' Findings Were So Wrong on Election Day

Wall Street Journal ^ | JOHN HARWOOD and SHIRLEY LEUNG

[rs79bm]

Just before Election Day, the St. Louis Post-Dispatch published a stunning poll about the governor's race in Illinois: GOP candidate Jim Ryan was ahead.

The Republican's 43.5%-43.2% edge over Democrat Rod Blagojevich, though tiny and statistically insignificant, landed with a bang in the world of Illinois politics because many earlier surveys had shown Mr. Ryan to be far behind. Moreover, it was conducted by nationally known pollster John Zogby, who had been acclaimed for his accuracy in the last two presidential campaigns. Mr. Zogby told the Post-Dispatch that he had personally reviewed the result and had affirmed its accuracy.

[SAJ]

Why pollsters got it wrong? Easiest question in the world.

Remember LTCM, the single most successful investment/arbitrage group in the history of the world from 1994-97? They ultimately got it wrong, too, and for the same reason.

Statistical analysis of past events (even assuming it is conducted honestly, a LARGE assumption in the case of pollsters) DOES BOT AND CAN NOT 'predict' the NEXT outcome well. The assumption of these people is that events are rendom, distributed along a bell curve or its first cousin, a log-normal curve.

This is cr*p, and false-to-fact, and why ANYONE (no matter their political persuasion) puts any credence in 'polls' is absolutely beyond me.

[Always Right]

Polls are black magic. There are way too many assumptions in them and manipulations in them. It is a joke that polls claim to have a +/-2% error margin. If you look at polls vs. actual election outcomes, I would bet the errors are closer to +/-8% at a 90% confidence level.

[AndyTheBear]

Yes that "error" margin--which applies to their distribution curve, and might be a good predictor of what another poll with the same methodology might predict if taken around the same time with similar demographics. The problem is it does not apply to the actual election.

Although I do think the polls have some value, the vain inclusion of the margin of error is just a bogus attempt at appearing "scientific".