Posted 10:36 AM by Robert

Smaller Bounce?

Sunday September 05, 2004:

The Rasmussen Reports Presidential Tracking Poll shows President George W. Bush with 48% of the vote and Senator John Kerry with 46%. The Tracking Poll is updated daily by noon Eastern. Two-thirds of the interview for today's report were completed after the President's speech on Thursday night.

That's not much bounce - even less than yesterday. But so far Rasmussen seems to be a bit anomalous. Zogby, TIME and Newsweek all show a bounce of at least 9%. But some of those polls may have sampling problems exaggerating the bounce.

By the way: "Oversampling" of one group or another does not by itself invalidate a poll or make it misleading. One of the black arts of polling is "adjusting" the final results from the raw data for exactly over and under sampling. Almost all major pollsters make such adjustments - but most are very discrete about revealing how they adjust. For example, pollsters are aware that under certain conditions and at certain times one group or another tends to be home to answer the telephone, and that some people just won't respond to surveys very often even when they do answer the telephone - and the pollsters adjust for all that.

The problems arise when the "adjustments" are bad. For example, the infamous Los Angeles Times polls purporting to show Gray Davis not doing too badly virtually on the eve of his landslide recall seems to have "adjusted" its raw data by assuming voters favorably disposed to Mr. Davis would turn out fairly massively. They didn't. A recent Los Angeles Times presidential preference poll unfavorable to Bush-Cheney was widely criticized because it clearly oversampled Democrats - and from the final results it was just as clear that the pollsters didn't adjust correctly for the oversampling.

I cannot determine from the TIME, Newsweek and Zogby Polls whether they have been adjusted at all, or at all appropriately.