Every year we see the same pattern. Way out there when nobody can prove them wrong, all the polls show the Democrats with a wide lead over Republican candidates. And lo and behold, in the last week of the campaign, everything changes around, and the Republicans narrow the gap and surprise, surprise, Ronald Reagan is President, or Newt Gingrich is Speaker of the House.

How many times do the polls swing the other way? They do, but it takes a DUI charge from a couple decades ago hits the news in the last weekend before the election or some such.

My point is that polls tend to be biased estimators. ERGO, I'd like to propose that whenever any pollster makes any prediction about any election, the closest comparable election should be found, and the pollster's closest comparable prediction should be compared with the final result. Write down the difference. Then subtract that difference from his current prediction.

It would be amusing to gauge the accuracy of this statistic. Anybody got some historical data and a spreadsheet?