Reply #45: Sorry

[Febble]

here you go:

http://inside.bard.edu/~lindeman/slides.html

and:

http://journals.democraticunderground.com/Febble/3

And here's another plot:



X axis exit poll discrepancy (positive = count "redder" than poll; negative = count "bluer" than poll) in standard errors. ("redshift")

Y axis is percentage point swing to Bush from 2000 (positive = swing to Bush; negative = swing to Kerry.

So "redshift" means count redder than exit polls and "swing" means voteshare redder than in 2000. They aren't the same variable!

You can see that there are many precincts in which the discrepancy is several standard errors from zero, in both directions, but you can also see that the net discrepancy is in the "redshift" direction. You can also see that while Bush did worse than 2000 in some precincts and better in others, overall he gained voteshare.

You can also see that there is absolutely no correlation between the two. There are 1250 precincts in the analysis, and the R squared is given, and the regression line is visibly (but insignificantly) negative, so you can even work out the confidence interval of the regression line if you like.

As for precinct level data within state - there are only tens (49 in Ohio) of precincts polled in each state, so there is very little statistical power. ESI published a precinct level study of the exit polls in Ohio, and essentially found the same as I have posted here - no correlation between swing and shift, ergo, no evidence FOR fraud. However, because of the much lower statistical power, it is not evidence against. Because of the very large statistical power in the plot above, it is actually contra-indicative of widespread massive fraud.

Feel free to ask if you have any more questions.

Cheers

Lizzie

edited to correct unfortunate freudian slip