IT ZOGGLES THE MIND:

The Devil in the Data: All Poll Numbers Are Not Created Equal (Stuart Rothenberg, October 17, 2002, Roll Call)

Zogby's September poll in Tennessee found Lamar Alexander (R) leading Bob Clement (D), a reasonable conclusion. What wasn't reasonable was Zogby's finding that Clement was leading by 9 points among men, while Alexander was leading by 25 points among women.

In his release, Zogby referred to the "reverse gender gap" and asked, "What's going on in Tennessee?" I know the problem, and it had nothing to do with the electorate in the Volunteer State. It was the data. Zogby's October data were more reasonable, showing Alexander with a big lead among men and running even with Clement among women. The September cross-tabs were simply wrong.

Want another example? Zogby's Sept. 16-17 New Jersey poll showed Sen. Robert Torricelli (D) leading Doug Forrester (R) 39 percent to 34 percent. Whatever you think about those numbers, it's awfully hard to swallow Zogby's finding that twice as many Republicans were planning to vote for Torricelli as were Democrats for Forrester.

Sure, once you take into account the margin of error in each of these cells, the opposite result could be (and certainly was) true. But these dopey, small-sample, sub-sample results only demonstrate that cross-tabs are of limited utility when trying to monitor Senate races.

How can you even release a poll that shows the Republican leading among women and the Democrat leading among men? Posted by Orrin Judd at October 17, 2002 8:14 PM