George,

After the recent flurry of comments on your site regarding post-convention polls and “bounce”, perhaps I can shed some light. (I taught for many years in a Master’s degree program that trained students for a job in the polling “industry”.)

First, not all polls are created equal. The sampling process must be totally random, but some pollsters will cut corners to get their results out before their competitors. It takes at least 3 days to conduct a valid random poll. (See Note at end for an explanation of this 3 day requirement.) The post-convention CNN/USA Today/Gallup Poll was done in 2 days. Bad data should not be considered.

Of the reputable polling organizations which used the proper 3 day procedure, this leaves the CBS poll and the ABC/Washington Post poll. The CBS poll showed no bounce for Kerry; the ABC/Washington Post poll did. Why the different results? I e-mailed a colleague who is the director of the CBS poll and asked her. The answer is “very easy”, she said. You have to look at the PRE-CONVENTION poll which was used as the base line for measuring the “bounce”. The CBS pre-convention poll was done back in July 11-15 just after the announcement that John Edwards would be Kerry’s running mate. That announcement produced a pro-Kerry bounce at that time. Thus, a comparison of the CBS pre-convention poll and its post-convention poll is a comparison between two peaks – both high points for Kerry. Measuring from one high point to another shows little change and thus no convention effect, but both polls showed Kerry with a 5 to 6 point lead over Bush.

So this leaves the ABC/Washington Post poll. They did a survey just before the convention and another one just after – a true pre-post test. They found Bush 2 points ahead of Kerry (48 to 44) prior to convention. The post-convention poll found Kerry 6 points ahead of Bush (50 to 44). This is an 8 point swing! Not a bad “bounce”.

Most media commentators know nothing about polling, yet they use the results all the time. They mix quality, scientific poll results with corner-cutting crap. It’s all the same to them, especially if it “proves” their point.

David RePass

Note: The reason that at least 3 days are needed to compete a valid poll is to ensure that each person who was drawn in the sample is given every opportunity to be reached. Not everyone is sitting by their phone waiting for a call from a pollster! People go out for the evening, they may be at work, they may be running an errand, etc. when the pollster calls. Pollsters must often make many call-backs over several days before finally reaching the person who has been randomly selected.) (electronic mail, August 9, 2004)