Reply #12: most of this

[OnTheOtherHand]

I think I've basically answered above. Maybe the fit is substantially better in the midrange, but not much. The fact that the simulation on p. 8 goes a little bit up from category 1 to 2, and then a little bit down from 2 to 3, for me isn't enough to prove very much. You have to imagine error bars around these results.

I'm still a little puzzled by the shape of that second p. 8 simulation line. Actually, it seems to posit a flat (or maybe downward-sloping?) regression line even for WPE -- i.e., lower errors toward the high-Bush precincts -- which is not the result I thought Baiman was looking for.

Most of my post 4 was trying to make sense out of Appendix F on pages 19-20, which I admit I've never really understood (and I swear I have tried), so I am open to instruction there. Or maybe we should leave Appendix F out of it for now. But what I take from Table 1 is that the errors should be lower in high-Kerry precincts (smaller alphas) and higher in high-Bush precincts (much larger alphas). The second simulation on p. 8 -- the one that supposedly shows the telltale zig-zag from category 1 to 2 to 3 -- indicates _virtually no difference_ between the high-Bush and high-Kerry precincts. Even accounting for the distinction between WPE and alpha, that isn't going to fly (which is why, in fact, the simulation line is so very far from the E-M results). So this simulation seems to be at war with Table 1 on page 19. If Ron is responsible for both, I hope he will tell us which one to believe.