While I will say that I do not believe in re-weighting partisan ID in a likely voter screen, my beef is not with the LV screen.
In FL they have the thing at 28 D, 31 R, 35 I. Real FL registration is 37 D, 35 R, 26 I. Now again we are not talking about the LV model, we are talking about the original random sample, which is what the LV screen is based upon.
9 POINTS light on D? 9 points heavy on I? 4 points light on R? That just appears to be an off sample.
Similar but less influential phenomenon in AZ. But let me allay your fears-Joe is outperforming Hillary with whites by a point in FL, and 3 in AZ. So a proper extrapolation, not an unskewing, has AZ at about 52-47 which sounds right, and FL at 50-49, which in FL, is about always the case.
So what can account for errors in LV model? Undersampling minorities maybe. ABC does not give a readout of black support. And in AZ, he is matching Hillary with 61 percent among Latinos.
Let’s look at AZ, shall we?
In 2016 the electorate was 75 percent white. Hillary won 40 percent of that. In THIS poll, the electorate is 64 percent white and Joe is winning 43 percent of it.
Do the math. 43 times .64 is 27.52. 61 percent of 21 Hispanic is 12.6. That is 40.12. Black vote at say 90 percent is 3.6 that is 43.72. So does Joe win over 60 percent of remaining non white vote? He certainly wins more than 50, which is what you would have to believe for these numbers to work.
The problem with polling is simple-demographics rarely shift. The story this year is Joe is moving his numbers with whites, mainly college educated, by 5 points. The rest you can pretty much historically rely upon. 90-95 with black vote, 60-65 with Hispanics, 65 with Asians, though more recently that has pushed into the 70s in some places.
But this is the problem-out of 701 likely voters in AZ, 28 in this poll are AA. 28. So if you get a flat out weird sample, and plug in, 50-50 instead of 90-10, you get a three point swing ON THE WHOLE POLL. Polls need to be re-evaluated.
The best thing you can do is take wide swaths of data of every conceivable demographic, THEN project who is turning out based on an LV screen. We know the black community is not magically more conservative to that extent in AZ. And ABC’s own numbers show a win in FL, because Hillary lost by 1.2 points there and Joe is doing better with whites with less white turnout in this poll than 2016, and all other demos about the same. How? Crucially, he is doing better with whites than previous polls showing mid-single digit leads.
At 38 percent with whites in FL, and 43 percent in AZ, Joes wins AZ handily and FL narrowly.
Remember-base your projections from demographics, not partisan ID.
Whenever you see a panic inducing poll look at the white support percentage. Anything 43 and up pretty much anywhere is a win for Joe.
But going forward we have to make progress on the economic message. Not sure how Trump has solidified position after Woodward-it is almost like people enjoy being lied to.
Our job is to make them want the truth, then ensure they believe it when they hear it.
-ROC