Silver's Polster Adjustments Are Minor
Silver's averages give Harris 58.9% victory odds, Real Clear Politics gives her 54.2%
Haters have long criticized Nate Silver’s method of averaging polls, suspecting him of some sort of partisan hackery. His process involves four main steps. First, each poll is adjusted to account for pollsters’ “house effects.” This adjusts the results of pro-Democratic pollsters towards Trump and vice-versa. Next, each poll is given a weight which depends mainly on three things: (1) a quality score for each pollster; (2) the recency of the poll, using an exponential decay function; and (3) the poll’s sample size.
Adjusting for recency and sample size should not be controversial. Voters can change their minds. Run enough regressions and you can show that polls released just before an election are more predictive than earlier polls. The relationship of sample size to accuracy is literally axiomatic because the standard error of a sample estimate is inversely proportional to the square root of the sample size. However, the house effects and pollster quality adjustments are somewhat subjective. I trust Silver’s judgment— I’ve checked his numbers enough to respect his work— but they create the potential for importing bias into a supposedly “objective” model.
However, Silver’s pollster ratings and house effects adjustments only have a big effect when polling is thin. The latest polling averages from the Silver Bulletin and RCP are as follows:
Silver RCP
AZ 49.35 49.15
FL 48.25 *
GA 49.65 49.25
MI 51 50.65
MN 52.95 *
NV 50.9 50.55
NC 49.8 49.7
PA 50.6 50
WI 50.95 50.4
TX 47.1 *
RCP’s numbers are 0.1 to 0.6 points more favorable for Trump. It does not publish battleground averages for Florida, Texas and Minnesota. However, because these states only flip in the event of a significant victory, they don’t matter much. For these states, I used Silver’s averages minus 0.5% to produce a forecast based on the RCP model.
Here’s what happens running my model using Silver’s polling averages:
HARRIS WON 588141 out of 1 Million times
She got exactly 270 EVs 14173 times
The Blue Wall (PA/MI/MN/WI and NE-2) holds 491814 times
Blue Wall Cracks 508186 times of which Harris salvages 96327 wins
Harris wins a southern state (GA/AZ/NC/FL/TX): 668111 times
State wins w/ state losses w/ state wins w/o state losses w/o state
AZ 390917 22431 197224 389428
GA 345671 107940 242470 303919
FL 279942 68 308199 411791
MI 564183 72558 23958 339301
MN 581277 255237 6864 156622
NV 535240 82561 52901 329298
NC 408436 64766 179705 347093
PA 547121 31767 41020 380092
TX 155274 11002 432867 400857
WI 544237 79694 43904 332165
NE-2 574254 303482 13887 108377
And here’s the output when I use the RCP numbers:
HARRIS WON 538943 out of 1 Million times
She got exactly 270 EVs 12088 times
The Blue Wall (PA, MN, WI, MI and NE2) holds 426582 times
Blue Wall Cracks 573418 times of which Harris salvages 112361 wins
Harris wins a southern state (GA/AZ/NC/FL/TX): 646548 times
State wins w/ state losses w/ state wins w/o state losses w/o state
AZ 378912 36376 160031 424681
GA 299429 102038 239514 359019
FL 226728 31 312215 461026
MI 526563 111072 12380 349985
MN 531043 261850 7900 199207
NV 485370 89156 53573 371901
NC 385941 74787 153002 386270
PA 479694 21584 59249 439473
TX 114009 8155 424934 452902
WI 480891 72805 58052 388252
NE-2 522452 319292 16491 141765