Reply #11: reproducing shapes

[OnTheOtherHand]

Frankly, the shape of the "Vote Shift Simulation with Bias" (second graph on p. 8) seems even further off than the shape of the "Simulated Response Bias" on p. 7, especially if you consider numbers of precincts.

The one on p. 7 (without vote shift) is off by 0.06 or 6% for mean WPE in the high-Bush precincts, at left. It's off by about 0.01 or 1% for mean WPE in the high-Kerry precincts at right.

The (latter, because I assume this one is supposed to be better) one on p. 8 is still off by almost 0.04 or 4% for mean WPE in the high-Bush precincts, and it is off by close to 0.06 or 6% for mean WPE in the high-Kerry precincts. And bear in mind that there are more than twice as many high-Kerry precincts as high-Bush precincts.

Honestly, I'm still staring at the spreadsheet and trying to figure out where, exactly, these numbers come from in the first place. But while it may be obvious to Baiman that the p. 8 model fits better than the p. 7 model, I don't see it.