A look back at how bad polls were last time--

[mistertrickster]

In final presidential trial heats compiled one to four days before the last presidential election, only four polls out of 18 showed Gore winning or tying. The majority—78 percent—showed Bush winning the popular vote in the latest polls before the election. The average bias toward Bush was three points, with some polls putting Bush as high as six points ahead.

Two days before the election, the widely quoted Gallup poll predicted Bush up by five. Three days before showed Bush ahead in the Washington Post poll by three points, the ABC poll showing Bush with a four point lead and the CNN/USA Today poll putting Bush six points in front.

According to data I analyzed from the National Council on Public Polls website http://www.ncpp.org/, national polls have been remarkably accurate from 1968 to 1996. In ’96 for instance, every poll showed Clinton winning over Dole, and Zogby / Reuters even had the percentages exactly right. In the three-way ’92 race of Clinton vs. Bush vs. Perot, all the polls showed the eventual winner as winning. While a very few individual surveys in tight races over the decades have bet on the wrong horse (Gallup in 1976 when they showed Ford beating Carter, for example), looking at the overall consensus of the polls, the winner had been accurately predicted every time.

Why the 2000 elections so confounded the pollsters remains a mystery that still defies clear explanation. One could argue that the race was too close for anyone to accurately predict—but that still doesn’t explain why the polls skewed so heavily in favor of the Republican, who lost by the way. If the race was too close to call accurately, then half the polls should have predicted that Gore would win; instead only two out of 18 showed him winning.

A kid flipping a nickel would have done better.

************

Two caveats-- 1. pollsters would say they DID WELL because they were very close to the final percentages that the candidates earned. True, but they failed miserably in what people really wanted to know--who the winner would be.

2. You can also see from the data that undecided voters clearly swing to the challenger (not the incumbent) when voting takes place. That's a big plus for our man, Kerry.

[unfrigginreal]

1. Here's a link that shows all of the polls from 2000

http://www.pollingreport.com/wh2gen1.htm

[pmbryant]

2. Polls were not particularly accurate in 1996, or in 1980 for that matter

The polls in 1996 generally showed Clinton winning by a far larger margin than he actually won by (Zogby was a rare exception).

The polls in 1980 indicated a tight race, but Reagan won by 10 points.

In 1992, all polls suggested Clinton would win, but almost all gave him a margin significantly more than he eventually won by.

The track record of the polls in recent Presidential elections is rather poor if you want accuracy to be better than 5 percentage points in the margin between the candidates.

--Peter

[mistertrickster]

3. Good point--that could partly be explained by undecideds switching

to support the challenger, couldn't it?

Reagan in '80 and Dole in '96.

[ISUGRADIA]

4. The polls in 2000 were close enough that

a "winner" could not be predicted given the margain of error. Most of the polls were within the margain of error when compared to the final result.

Unless a candidate is beyond that MOE rate, the race should be called by the pollster or media as a tossup statistcally. Factors that favor a candidate should be mentioned. In 2000, Gore was making a steady improvement in the polls in the last week, so a final peak was not surprising looking at the series of polls.

http://www.ncpp.org/1936-2000.htm - final polls of 2000