I think there is merit in coming at this from the other direction, looking at actual results from adjacent counties with similar populations, media markets .... and seeing what how much difference is left after using the same variables used in poll adjustments. Look at adjacent counties in northwestern Wisconsin. Washburn County (Crowley: 39.7% to 31.6% with 1,669 total votes) and Sawyer County (Hong: 37.3% to 35.7% with 1,845 votes). These vote totals are similar in size to good poll sizes.
If the factors used in adjusting polls can't close this 9.7% swing, then there is a residual uncertainty that no additional polling/analysis will close and we will have a better estimate of the underlying uncertainty in the polls. This is the methodology used as an aspect of election night forecasting (once I know County x then I can refine my estimate for County y).
I do think that Elliott's various analyses do show that Hong's numbers were about right and that the consolidation around Crowley would have been very hard to capture given the slew rate of the change so near the election. What has been labelled as a bad miss really is not.
Realized that it might be easier to use the state polls to predict results for Washburn and Sawyer and see how well (or not) that delta matches the 9.7% observed. That methodology could be used across all counties to calculate a residual error that represents floor for polling error. Of course there is no limit on the error ceiling!
I think there is merit in coming at this from the other direction, looking at actual results from adjacent counties with similar populations, media markets .... and seeing what how much difference is left after using the same variables used in poll adjustments. Look at adjacent counties in northwestern Wisconsin. Washburn County (Crowley: 39.7% to 31.6% with 1,669 total votes) and Sawyer County (Hong: 37.3% to 35.7% with 1,845 votes). These vote totals are similar in size to good poll sizes.
If the factors used in adjusting polls can't close this 9.7% swing, then there is a residual uncertainty that no additional polling/analysis will close and we will have a better estimate of the underlying uncertainty in the polls. This is the methodology used as an aspect of election night forecasting (once I know County x then I can refine my estimate for County y).
I do think that Elliott's various analyses do show that Hong's numbers were about right and that the consolidation around Crowley would have been very hard to capture given the slew rate of the change so near the election. What has been labelled as a bad miss really is not.
Realized that it might be easier to use the state polls to predict results for Washburn and Sawyer and see how well (or not) that delta matches the 9.7% observed. That methodology could be used across all counties to calculate a residual error that represents floor for polling error. Of course there is no limit on the error ceiling!
🧄🧄🧄🧄