Prediction: GA Runoff Turnout
Overview and Background
Today's date is December 23, 2020. On January 05, 2021 Georgia will have two runoff elections for senator. The last day of early voting is December 31, 2020. Many counties have no voting on December 24 & 25. Additionally, GA's largest county (Fulton County) will not have voting on December 31st in observance of New Years Eve. Since early voting started on December 14, daily runoff turnout has been well defined by the function Runoff Voting = .94 * corresponding_GE_day_of_voting. I believe that for days on which early voting is available, this trend will continue. Additionally, I believe that the best proxy for Runoff election day will be early in person voting, which has been running at about 103% of GE Election day voting. This data is derived from the website georgiavotes.com.
In creating this model, there were effectively two variables I needed to think about. 1) Runoff Early Voting as a percent of corresponding GE Early Voting Day and then 2) Runoff Election Day Voting as a percent of GE election day voting. Let's first analyze Early Voting
Early Voting
We have 7 days of early voting left. Since December 14, early voting has stood at around 92 - 94% of the equivalent election day and there is no sign of this stopping. Therefore, best guess is for this trend to continue. Expect about 3m early votes. For our sensitivity table, we well test values between 80% - 100% but we expect the range to end up somewhere between 87.5% to 95% of the equivalent GE early voting day.
Toggles
There are three toggles that I have put into my model. Firstly, there is a question of how many votes will arrive from Jan 1 - Jan 5. Even though early voting will be closed by then, mail in ballots will continue to arrive as the last day to request a mail in ballot is Jan 1. During the corresponding time period, the general election had an additional 111,534 ballots come in. I believe that we will see roughly 78% of this in the runoff as my expected mail in ballot requests sum to 78% of general election numbers.
The next toggle is the hours modifier. Some counties have a small amount of voting open on Dec 24 which is not reflected in the assumption that 0 votes will occur on Dec 24. Additionally, many counties have fewer voting hours on December 31 in observance of new years eve. Toggling on the Hours modifier changes the expected vote function to the following for december 24 and december 31.
GE votes * expected percent of GE early voting modifier * (holiday hours/typical hours weighted by county turnouot). This means we expect maybe 10 - 20k votes on December 24 and significantly fewer votes on Thursday, December 31 than we might otherwise expect.
The last toggle is 31 of December vote basis. When toggled on, the 31st of December expected votes are based on the last early voting Thursday from the General Election. When toggled off, the votes are based on the last early voting Friday from the General Election. This modification is made because voters tend to turn out more on the last day of early voting because they like to procrastinate.
Election Day Voting
Runoff election day turnout will likely be a function of GE Election Day turnout. Additionally, the ratio will likely be similar to the ratios seen in in person early voting because both modes of voting require that you actually get up and go to a polling place. Thus, they are subject to similar conditions (pandemic fears, voting fraud fears, etc.) In person early voting has stood at around 100% of general election day data, but I think that this is too high because the weather is meant to be bad on Jan 5. Therefore, let's say the likely range for election day turnout is 87.5% - 95% of GE election day turnout.
Scenarios
There are three toggle scenarios we need to worry about. The specific toggles can be seen in the box below the sensitivity table. Here I lay them out.
High Toggle
Best Guess Toggle
Conclusion
In conclusion, expect a range from 3,780,161 to 4,139,673 votes. That spans brackets B5 - B7. Best bet is B6.
Appendix - Current Pricing (12/23/20 9:30pm)
Regular Election Runoff
Special Election Runoff
December 24 Update
On December 23, we saw either 156,412 votes come in or 178,902 votes come in. The uncertainty is a result of backdating on mail in ballots. Should we count a mail in ballot vote on the day it arrives at the poll or on the day it is sent out? I will do some more research to make sure that we are making an apples to apples comparison. I suspect that 178,902 is the correct number to go by from a data comparison standpoint with the GE and will use that number for the rest of today's analysis. 178,902 votes is 87% of general election votes - below even the lowest percent range. Rather scary.
I have added an additional analysis type to my model. I call this the "Proportion of Early Turnout Model." During the general election, election day turnout made up 19.7% of total votes. Thus, with this analysis I flex early voting as a percent of the total vote and then flex election day votes as a function of early votes cast. This method shows a significantly larger range of possibilities than previously expected and underscores the level of uncertainty even 10 days out. NOTE: it is logical to expect that runoff election day turnout will make up a higher proportion of total votes than it did in the general election. This is because 1) fewer early voting days in runoff 2) holidays will depress early turnout near xmas 3) GOP vote conspiracies will encourage right leaning voters to vote on ED instead of early.
Another change I have made in order to reflect potentially lower turnout is to add a white box displaying turnout from 80% to 87.5% of GE numbers in my "Percent of GE Turnout" model. Let's run through the scenarios. I will only be running the "Best Guess" toggle scenario because otherwise this blog post will get way too long. From now on, I will show both the "Percent of GE Turnout" model and the "Proportion of Early Turnout" model.
January 1, 2021 Update
Happy new year! Early in person voting wrapped up yesterday with 3,001,017 votes. In this thread, I want to analyze two factors which will decide final turnout: VBM from Jan 1 - 5 and election day votes. First, let's look at pricing as of Jan 1, 2020 5:30pm. The special election has roughly the same pricing (give or take a cent) so I'll just post this snapshot for brevity's sake.
GA Regular Election Turnout Pricing
If you had invested in B6 as recommended by me on December 23rd, you would be up 52/31 = 67% on your investment. If you invested in the range B5 - B7, you would be up 96/72 = 33%.
First, let's look at what the previous model is telling us:
Essentially, according to the old model this is a race between B6 & B7. The issue is, this model is unable to flex VBM votes to be received from Jan 1 - Jan 5 (it's just stuck at a static 86k. While this is a good guess, I don't like the idea of not having a scenario analysis as it can lead to overconfidence.) Therefore, we need new models. These new models need to flex two factors: VBM votes from Jan 1 - Jan 5 and election day votes. To this end, I have created two new models: the "GE Vote Basis Model" and the "EV Vote Proportion Model."
Let's walk through these models.
GE Vote Basis Model
The "GE Vote Basis Model" flexes expected VBM to be received from Jan 1 - Jan 5 by expected election day turnout. Thus far, we have had 1,373,691 mail in ballots requested. This should stay constant as today was the deadline to request a mail in ballot. 928,069 mail in ballots have been returned, for a return rate of 68%. In the general election, the return rate was 74%. In this model, we flex VBM left through the following function:
VBM Left = 1,373,691 * (expected VBM return rate) - VBM already returned
Election Day turnout is flexed as a percentage of GE election day turnout.
We have a few factors building into the election day turnout assumptions. Please note that election day turnout is highly uncertain because of multiple idiosyncratic factors (unprecedented doubt from GOP in election security, Coronavirus, competitive runoff election in GA, level of spend on GA senate race.)
Let's try to think through the facts.
1. On days unimpacted by the holiday season, we have seen turnout stand at roughly 90% - 95% of the corresponding GE voting day.
2. Trump is having a rally on Jan 4 in a conservative, low turnout area of GA. Increase turnout.
3. Weather will be good on Jan 5. No effect but avoid negative of bad weather.
4. Trump/Biden not on ballot will likely depress turnout.
5. GOP voters more likely to vote on election day due to widespread fear of voter fraud on right. Corroborated by low GOP turnout during early voting (this might also mean that GOP voters don't care without Trump on the ballot.)
6. Fewer polling places will be open because it is a runoff. Depress Turnout.
Based on this, my entirely unscientific guess puts us at election day turnout of 75% to 95% of General Election election day turnout for 730k - 930k votes.
EV Vote Proportion Model
The "EV Vote Proportion Model" flexes VBM in exactly the same way as the "GE Vote Basis Model." It flexes election day vote as a proportion of total turnout. The exact formula is thus:
Election Day Vote = (Total Early Votes)/(1 - [ED expected percent of Total Vote])
The election day vote was 19.7% of the total vote in the general election. I believe it is likely to be a higher proportion of the total vote due to right wing fears about early vote security. The counterargument is that we will see way fewer right wing ballots because Trump is not running on Jan 5. I believe that this point is mitigated by the Trump Jan 4 rally. Therefore, I think we will see election day vote take up something like 19% - 23% of the total vote.
Based on this, I am expecting 720k - 920k votes on election day.
B6 is the clear favorite to win. Fair pricing based on current data is probably:
B5 12%
B6 70%
B7 15%
B8 3%
B8 is a bit above my fair pricing, but I still think it is worthwhile to buy as a hedge against absurdly high turnout (>100% of GE).
Comments
Post a Comment