There are other kinds of bias - importantly party bias - that defy the null every election. In 1994 the year of the Gingrich revolution, there was Democratic bias because most polls understated GOPs in those elections. In 1998, there was GOP bias when in October the Judiciary Committee sent the articles of impeachment to the House floor which apparently motivated Democratic voters.

Even late national presidential polls hardly ever show anything close to normal distributions of party bias. The last time they did was in 1984.

My comments: 1. The 1989 paper was based polls on taken in the 70s and the 80s. As you can tell, poll result data were harder to find back them. The source document for my paper at the Polling Report site is attached. [I sent him mine as well] 2. Analysis of polls in the 90s showed less extreme distributions, closer to 70%/25% or 65%/30%. See the next four attachments. Tom Gruca at U. of Iowa wrote on the subject covering 1990-1992 and data that George Terhanian of Harris complied for another project I was working on.

3. In a 1996 Public Perspective article I refer to this distribution in state polls, a paper discussing the pattern in final national polls conducted just a few days out from election day. The national poll analysis showed "In 26 of 36 cases or 72%, more undecideds appear to vote for the challenger". See attached.

Regarding you analysis of over 220 polls that were taken in Governor, Senate and House races that involved an incumbent since 1998 taken within two weeks of Election Day.

4. I did not do 2000 elections. I did analyze the Terhanian data from 1998 above. They showed a distribution of:
Incumbent Bias 65%, Challenger Bias 31%, No bias 4%, still strong enough to reject the null.

5. In 2002 I analyzed 159 Senate and Gov races for the NCPP. The original spreadsheet is attached. See link below for my reason for doing this analysis<<br>. http://www.pollingreport.com/ncpp1.htm

In 2002, although incumbents are indicated in the spreadsheet, not evident is the distribution of 86 incumbent polls. I did an incumbent analysis in late 2002. The results were surprising, 38 were biased toward the incumbent *41* were biased for the challenger, and 7 had no bias.

At the time I figured that that after year after year of finding incumbent bias, that 2002 may have been an exception. 2002 may have been the exception in our data. That happens. Please send me your data. As for me, I was not prepared to conclude that the rule was suddenly out of date.

I will be doing the NCPP analysis after November. Let's wait to see what those data say. Again, please send me your data.

Nick