3 ms·

Naive question, can we just "backfit" sets of county metrics based on results of the last few elections to determine which ones have effect? Surely with a large

Naive question, can we just "backfit" sets of county metrics based on results of the last few elections to determine which ones have effect? Surely with a large enough set of metrics (i.e. all combinations of density, race, and gender), a large enough sample size (all counties in this case), and over multiple elections (albeit all reasonably recent to avoid historical shifts), we can determine the set of metrics which are the highest predictors with the fewest exceptions without appearing to rely too much on correlation.

microcolonel 8y ago
> Naive question, can we just "backfit" sets of county metrics based on results of the last few elections to determine which ones have effect? Despite what people say, the candidates in a given party differ considerably between elections (sometimes even if they are exactly the same person). "Backfitting" could only give you the trend among "lifelong democrats" and "lifelong republicans", which thankfully are not a majority of voters.