Election 2020: Naive Forecasts of Swing States

Update: 2020-11-06 10:50 EST:

Biden is now ahead in Pennsylvania by 6826 votes, and the remaining votes to be tallied are expected to increase his margin there. The naive forecast looks like it was good for this race.

Biden now leads by 1579 votes in Georgia, where fewer than 20K ballots of any sort remain to be counted. This could go any direction. There are over 5000 rejected ballots in Georgia that could be cured, if those voters do so by the close of business today. What the naive forecast got right here was that the vote was going to be close, but there’s no telling yet who will end up with the lead.

Nevada’s remaining ballots are expected to favor Biden and increase his margin there.

Biden’s lead in Arizona has reduced to 43779 votes. The 142,000 remaining ballots do not appear to follow the same distribution as the AB%, so it is unclear where the final tally will leave us yet, and the naive forecast from this point would be unlikely to be a useful guide. A different naive forecast built on the vote distribution from deltas in recent updates would be more likely to be informative, though one then runs into issues with small sample sizes.

============================================

2020-11-05 afternoon data pull:

TL;DR: MI, AZ*, PA, and NV are likely Biden wins, though NV is a closer race with more uncertainty. GA and NC are likely Biden losses, though GA has a relatively thin margin, but the recent updates have slightly increased Trump’s lead there. *AZ is now at risk, see the update below.

Naive forecasts for swing states, 2020-11-05 ~15:25 EST data from the NYT. AB% is ‘absentee ballot percentage’. ‘WV%’ is the ‘weighted vote %’, an estimate of percentage of vote that aggregates county tallied vote data. The naive forecast assumes that the AB% or WV% per candidate seen in vote tallies to this point fairly reflect the AB% or WV% that the untallied votes will have. To the degree that applies, these forecasts will work. I have included a statement on how many more votes the projected loser would have to take from the untallied votes over the naive forecast expectation. This would indicate how easily a difference in distribution of AB% or WV% in the final tallied votes might lead to a difference in the election outcome. These are listed in descending order of Biden estimated % performance in these states.

In MI, currently Biden is at 50.6% (2788425) and Trump is at 47.8% (2639035). The unreported vote number is 110316. Using AB% per candidate, the estimated final vote tally for Biden is 50.9% (2863771) with AB% of 68.3% and for Trump it is 47.5% (2672571) with AB% of 30.4%. This indicates a Biden win is more likely than not. The estimated final vote difference is 3.4% (191200). The projected loser would have to garner 95601 more votes than expected to win.

In AZ, currently Biden is at 50.5% (1470560) and Trump is at 48.1% (1402431). The unreported vote number is 407880. Using AB% per candidate, the estimated final vote tally for Biden is 50.8% (1687144) with AB% of 53.1% and for Trump it is 47.8% (1588424) with AB% of 45.6%. This indicates a Biden win is more likely than not. The estimated final vote difference is 3.0% (98720). The projected loser would have to garner 49361 more votes than expected to win.

In PA, currently Biden is at 48.5% (3127530) and Trump is at 50.2% (3236132). The unreported vote number is 515695. Using AB% per candidate, the estimated final vote tally for Biden is 50.8% (3539055) with AB% of 79.8% and for Trump it is 47.9% (3335661) with AB% of 19.3%. This indicates a Biden win is more likely than not. The estimated final vote difference is 2.9% (203394). The projected loser would have to garner 101698 more votes than expected to win.

In NV, currently Biden is at 49.4% (604251) and Trump is at 48.5% (592813). The unreported vote number is 134456. Using WV% per candidate, the estimated final vote tally for Biden is 49.4% (670786) with WV% of 49.5% and for Trump it is 48.5% (657955) with WV% of 48.4%. This indicates a Biden win is more likely than not. The estimated final vote difference is 0.9% (12831). The projected loser would have to garner 6417 more votes than expected to win.

In GA, currently Biden is at 49.2% (2423608) and Trump is at 49.5% (2436443). The unreported vote number is 98430. Using AB% per candidate, the estimated final vote tally for Biden is 49.3% (2474890) with AB% of 52.1% and for Trump it is 49.5% (2482607) with AB% of 46.9%. This indicates a Biden loss is more likely than not. The estimated final vote difference is 0.2% (7717). The projected loser would have to garner 3859 more votes than expected to win.

In NC, currently Biden is at 48.6% (2655383) and Trump is at 50.0% (2732120). The unreported vote number is 273090. Using AB% per candidate, the estimated final vote tally for Biden is 48.8% (2796571) with AB% of 51.7% and for Trump it is 49.9% (2861019) with AB% of 47.2%. This indicates a Biden loss is more likely than not. The estimated final vote difference is 1.1% (64448). The projected loser would have to garner 32225 more votes than expected to win.

Michigan is already called for Biden.

Arizona seems likely to be called for Biden; with just over 400K votes to go, the untallied votes AB% would need to shift another 8% toward Trump, which seems possible but not likely. Update 2020-11-05 21:24: Looking at a data delta of recently tallied votes, the vote % numbers are different and more favorable to Trump than the reported AB% numbers. At that rate, the lead Biden has is at risk. A quite recent notification from the NYT indicates the separation is only 1.5%, and at the rate of change of that recent tally, the lead could flip to Trump with another 234K votes or so. This all hinges on whether the remainder of votes AZ continues to show the same distribution as the recent tally.

It seems extremely likely that Pennsylvania will be called for Biden if the vote count is completed. Update 2020-11-05 21:36: Trump’s lead has narrowed to 42,142 votes.

Since Nevada’s AB% was a likely unrepresentative number for the untallied votes, I calculated the weighted vote percentage (WV%) from the county tallies and used that as the naive forecast basis here. With that, NV appears to be a likely Biden win, where Trump would have to take 6417 more votes than expected to take a win there.

Georgia is headed to a close finish, and Biden would need to have an additional 3.9% over the current AB% to garner a win there. The recent vote tally updates have increased Trump’s lead here. Update 2020-11-05 21:39: Trump’s lead has narrowed to 1,902 votes. To single-digit precision, Trump and Biden share the same vote percentage, 49.4%. Another 6,000 vote update with the same rate will see Biden take the lead in GA. NYT said there were 16,105 more votes to tally in Georgia when the Trump lead was at 2,497. If the recent rate holds through that, Biden will take GA. It could be exceptionally close when this is done, within recount range, and possibly a difference that might be small enough to be changed in recount adjustments to flip the other way, whichever way it lands.

North Carolina is not that close, and Biden would have to outperform the current AB% by 12% to garner a win, which would appear to put North Carolina out of reach.

==================================

2020-11-05 morning data pull:

TL;DR: PA, MI, and AZ are likely Biden wins. (Michigan has already been called for Biden.) GA, NC, and NV are likely Biden losses, though there is reason to suspect the NV AB% numbers are not useful for prediction of the untallied vote distribution, and Georgia is going to be very close even if the AB% is OK.

Naive forecasts for swing states, 2020-11-05 ~9:30 EST data from the New York Times. AB% is ‘absentee ballot percentage’. The naive forecast assumes that the AB% per candidate seen in vote tallies to this point fairly reflect the AB% that the untallied votes will have. To the degree that applies, these forecasts will work. ‘Unreported vote number’ is an estimate based on the NYT percentage reported and total votes reported numbers, and actually only have two digits of precision, since the NYT percentage is only two digits. I have included a statement on how many more votes the projected loser would have to take from the untallied votes over the naive forecast expectation. This would indicate how easily a difference in distribution of AB% in the final tallied votes might lead to a difference in the election outcome. These forecasts are listed in descending order of Biden estimated % performance in these states.

In PA, currently Biden is at 48.2% (3078666) and Trump is at 50.5% (3221132). The unreported vote number is 701947. Using AB% per candidate, the estimated final vote tally for Biden is 51.4% (3638820) with AB% of 79.8% and for Trump it is 47.4% (3356608) with AB% of 19.3%. This indicates a Biden win is more likely than not. The estimated final vote difference is 4.0% (282212). The projected loser would have to garner 141107 more votes than expected to win.

In MI, currently Biden is at 50.6% (2788425) and Trump is at 47.8% (2639035). The unreported vote number is 110316. Using AB% per candidate, the estimated final vote tally for Biden is 50.9% (2863771) with AB% of 68.3% and for Trump it is 47.5% (2672571) with AB% of 30.4%. This indicates a Biden win is more likely than not. The estimated final vote difference is 3.4% (191200). The projected loser would have to garner 95601 more votes than expected to win.

In AZ, currently Biden is at 50.5% (1469341) and Trump is at 48.1% (1400951). The unreported vote number is 407494. Using AB% per candidate, the estimated final vote tally for Biden is 50.8% (1685721) with AB% of 53.1% and for Trump it is 47.8% (1586769) with AB% of 45.6%. This indicates a Biden win is more likely than not. The estimated final vote difference is 3.0% (98952). The projected loser would have to garner 49477 more votes than expected to win.

In GA, currently Biden is at 49.2% (2413836) and Trump is at 49.6% (2432426). The unreported vote number is 245375. Using AB% per candidate, the estimated final vote tally for Biden is 49.3% (2541431) with AB% of 52.0% and for Trump it is 49.4% (2547752) with AB% of 47.0%. This indicates a Biden loss is more likely than not. The estimated final vote difference is 0.1% (6321). The projected loser would have to garner 3161 more votes than expected to win.

In NC, currently Biden is at 48.7% (2655353) and Trump is at 50.1% (2732120). The unreported vote number is 272879. Using AB% per candidate, the estimated final vote tally for Biden is 48.8% (2796432) with AB% of 51.7% and for Trump it is 49.9% (2860919) with AB% of 47.2%. This indicates a Biden loss is more likely than not. The estimated final vote difference is 1.1% (64487). The projected loser would have to garner 32245 more votes than expected to win.

In NV, currently Biden is at 49.3% (588252) and Trump is at 48.7% (580605). The unreported vote number is 167008. Using AB% per candidate, the estimated final vote tally for Biden is 44.5% (604619) with AB% of 9.8% and for Trump it is 53.6% (728240) with AB% of 88.4%. This indicates a Biden loss is more likely than not. The estimated final vote difference is 9.1% (123621). The projected loser would have to garner 61811 more votes than expected to win.

Michigan is already called for Biden.

It seems extremely likely that Pennsylvania will be called for Biden if the vote count is completed.

Arizona seems likely to be called for Biden; with just over 400K votes to go, the untallied votes AB% would need to shift another 8% toward Trump, which seems possible but not likely.

Georgia is headed to an exceedingly close finish, and Biden would need to have an additional 1.3% over the current AB% to garner a win there, which is not a large shift compared to the variance in county-by-county AB%.

North Carolina is not that close, and Biden would have to outperform the current AB% by 12% to garner a win, which would appear to put North Carolina out of reach.

Nevada would require Biden to exceed the current AB% by 37% (to get to 46.8%), which usually I’d think was insurmountable, but given Nevada’s reported AB% numbers come from one small, overwhelmingly Trump-supporting county, this might well happen.