11-10-2022, 05:55 PM
(This post was last modified: 11-10-2022, 05:56 PM by Eric the Green.)
(11-10-2022, 02:00 PM)David Horn Wrote:(11-09-2022, 04:41 PM)Eric the Green Wrote:(11-09-2022, 12:03 PM)David Horn Wrote: There is no doubt that the polls were less effective than usual, because the actual vote margins were often so close. I suspect that the typical margin of error is actually a lot larger than pollsters admit, because the underlying theorem used to determine MoE has been so totally violated that the polls are simply best-guesses.
I think the polls worked pretty well for the close senate races, as shown in the table above. Those from stronger blue and red states were less accurate, as the voters reverted to their usual habits, and so the polls didn't make much different there, and Florida sunk deeper and deeper into their swamp. And the hurricanes that they vote for time after time continue to hit them, and they ignore the lesson, preferring to indulge their gaddamm stupid prejudices. We should give up on Florida, and I hope they sink into the sea. I would swear at them more but I'm tired a'swearin'.
The poll predicted the House would be +1 Republican, and some late polls favored Democrats or were even. Quite accurate I would say.
I suspect that a lot of the accuracy was due to the pollsters putting a thumb on the scale to move the results to more likely places. A lot of the thumb work ware "adjustments to the model", which is statistics-talk for fixing the sampling problem. To be honest, there is no other option these days. Adding wieghting or alternate sources to get enough samples is the norm now.
I think I vaguely remember this discussion before. The polls turned out to be pretty accurate this time, and I know fivethirtyeight does make "adjustments" and that they analyze their polls this way and in other ways so that their average is more accurate, but polls can never taken to be as accurate in forecasting an election as the actual election returns.