While we await the probable victory of Biden with 306 electoral votes and 5-6% popular vote …
The above chart only shows the status that have been called by the media and when the number completed is> 95% (results from THE CURRENT). For example, California, Arizona, Nevada, Pennsylvania were disqualified. Everything is subject to change.
In just two states, North Carolina and Florida, the polls went in the wrong direction, where “wrong” means an indication of the voting amplitude and the result is the opposite, and is indicated in red ( North Carolina is not fully implemented yet). From a public consumption standpoint, however, it was the outcome: before the election, expectations were raised for a viable Election Night solution … then four suspenseful days. (I’m updating this on Saturday 11/7, after the network finally became known as Pennsylvania).
But if we go into detail, there is a notable flaw in the states polls. It has two components: (1) In states where polls in favor of Biden, the actual percentage of votes in favor of Trump increased by 2.6 points. (2) In states where polls favor Trump, Trump has done a lot better – averaging 6.4 points and steadily increasing his share of the vote.
Before you all get excited about that larger number… I should point out that it’s a known phenomenon in the landslides. Pandemic effect? Quite a while ago, I pointed out that states with margins tend to have polls that underestimate a winner’s rate of return. (See “Voters’ excitement was higher, ”November 12, 2008.) In this case, greater payoffs in strong Trump states could arise from the enthusiasm of the Democratic voters, who perceive that the ballot their have no influence. If true, it is interesting that in strong Biden states there is not yet the same effect (although we do not have enough data on IL / NY / NJ / CA yet, so it may not show up yet).
This growing error can be a pandemic effect. Democratic voters have paid more attention to public health. It would be worth going back and surveying red states voters to see if capable but non-Biden voters are at risk of discouraging disease. One can test this by seeing if the system errors are smaller in dark red states with lots of mail voting.
This can be of great importance in Senate races, where poll errors are quite large: on average 4.5 points, the largest error in the past 20 years. I note that Senate candidates’ vote rate is more like Trump’s vote rate than their own:
I think there’s a lot more to this Senate story, but I’ll just leave it alone for now.
Of the states with a voting margin of less than 10 points, the largest voting error occurred in Iowa (7 points), Florida (6 points) and Wisconsin (7.5 points). The errors were smaller in Texas (3.3 points), Michigan (4.5 points), Virginia (3 points) and NH / MN / CO / ME / NM (less than 3 points). The following main states do not have enough counts to get an accurate estimate: OH, GA, NC, AZ, PA, NV. They cannot seem to change the overall model.
Difficulty sampling non-university and Hispanic population? For the purpose of tracking the elections using polls, the other thing to explain is the 2.6 point error. Poll shows Trump has won support this year among non-college voters and Hispanics [GA] [TX] [WI]. These groups are known to be difficult to reach. It is possible that the survey respondents were not a representative sample of those groups.
In other words… these are not monolithic groups. It is a good idea to find a way to stratify within each group and identify different groups. That can be done by asking additional questions about engagement with news, social connections, etc.It may not take long and one can even identify what questions do not do. affect the later responses, a known issue in the poll.
Shame… or just undecided? Another possibility is called “undecided” voters. They were thought to be a factor in 2016, but less so this year. This year, they tend to change inversely with Trump’s support, IE If the undecided things go down by one point, then Trump’s support will increase. It seems that some of Trump’s supporters are unaware of their own biases. In 2008, I wrote about the cognitive science of how one can be do not know about their own hobbies.