Originally Posted by The Mayor
Can you explain how the author is wrong in his criticism of the PPP weights? That's the part that I find intriguing.
Silver has an article where he explains why he doesn't correct for self-ID of party affiliation, even though many polls collect that data.
So the National Review article was bringing up statistics (self-ID of party affiliation) that Silver has already shown to be not useful. Again shows a poor understanding of the the 538 methodology by National Review article.
As far as whether a particular poll has a lean or not, I believe he compares changes in poll x over a given time period compared to changes in polls y, z, and w over a similar time period.
Remember, every poll has a sample error of unknown magnitude. By comparing polls against one another and considering their historical biases, you can smooth out some of that sampling error. You don't want to smooth it all out because you'll smooth signal instead of noise. But to the extent you can smooth noise your predictions will improve. In my understanding, that is what the 538 method was doing with the PPP weighting.