Reply #195: OK, then what is your problem [View All]

[Febble]

with the NES survey as evidence, as cited by OTOH here?:

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=379573&mesg_id=390057

In the 2000-04 NES panel study, among the folks who reported a 2000 vote in both the 2000 postelection and the 2004 postelection survey:

Among the 308 who had voted for Gore as of 2000, 29 of them said in 2004 that they had voted for someone else in 2000 (23 Bush, 3 Nader, 2 other, 1 DK -- you can skip that one).

Among the 309 who had voted for Bush as of 2000, 6 of them said in 2004 that they had voted for someone else in 2000 (5 Gore, 1 Nader).

I make it that 6% of participants in the survey misrecalled (we do not know whether they forgot or deliberately misreported) their 2000 vote when asked about it in 2004. 9.4% of Gore voters misrecalled their vote, 1.9% of Bush voters.

i.e. not zero for either group. So the next question is: is the proportion of Gore voters who misrecalled their vote significantly higher than the proportion of Bush2000 voters? The two relevant statistical tests for this question are chi square, and Fisher's Exact test. Chi square is easy, and is almost certainly accurate for a sample this large, but I happen to have a program that will give me Fisher's Exact Test, so I did that.

Fisher's Exact Test gives a probability of 1 in 27,740 that the difference could have occurred by chance.

So we can probably conclude two things: yes people can forget who they voted for four years after the event (because out 6% of the survey participants did so - no inferential statistics required) and that Gore voters were significantly more likely than Bush voters to forget who they voted for in 2000 when asked in 2004.

This looks like evidence to me. There may be reasons why it is not valid evidence (but I have yet to see such a thing). It is certainly not invalidated by the sample size which provides plenty of statistical power, as is evident from the prrobability figure.