On Sat, 20 Nov 2004 hari.kumar at sympatico.ca wrote:
> Is not one of the problems with Multiple Regression that the factors are
> independent of each other? Most factors loaded into this analysis are
> unlikley to be so. H
If all of the independent variables in a regression model are highly intercorrelated, then the estimated slopes get wacky (that's formally known as the multicollinearity problem). However, the indepedent variables do not need to be completely independent of one another; moderate correlations are okay. From the "working paper" the researchers released, multicollinearity doesn't look like a major issue here.
Determining whether or not the use of electronic voting in FL caused vote miscounts is tricky. The researchers clearly demonstrated that electronic voting areas produced a higher vote count for Bush in 2004 compared to 2000, even after statistically controlling for a variety of sociodemographic factors that could explain the difference. However, as ravi (I think) pointed out, it is possible that some factors not included in the regression model covary with the use of electronic voting, and those confound factors could be the true cause of the bigger vote margin for W. (To use statistical jargon, there may be a specification error in the regression model.)
In any case, I'm glad this has caught the attention of some social scientists, and I look forward to further analyses (we do have a huge, relevant public archive of data to work from here!) Hmmm: maybe it's time for a sabbatical--
Miles