Methodology - 2022 U.S. Senate Election Model

The 2022 election models were a big shift away from the very primitive techniques I used with the 2020 models. The only exception to this was with the U.S. House model, as I did not have time to recreate that with the updated strategy. To sum up what changed, in the 2020 models the category winners were just whoever was ahead, and the overall race rating was about who won more categories, so there was very little numerical influence. With the 2022 updates, now the final rating is based on a percentage average of the categorical data, as explained below.

View Full Model (Google Sheets)

Specific Categories

With the Fundraising category, the percentage for each party is based on their total campaign receipts in relation to the total receipts for that seat raised by all parties. If John Doe (R) has $500 and John Doe II (D) has $500, then both parties get a 50% rating for this category. This data comes straight from the FEC. For the Recent History category, I used historical data from 270ToWin to find the percentage of the past 12 or so elections that each party had won. The Last Election category was simply the percent of the vote in the last election for this particular seat, which I got from Wikipedia. The Expert Forecast category looks at 270ToWin's Consensus Forecast Map and assigns each color a percent value, ranging from 50% to 70% depending on how safe a race is. The Polls category is pretty straightforward, with it of course just being dependent on how a party's candidate is polling in that race. This data comes from FiveThirtyEight's polling averages, or if no average is available whatever the latest poll was. The 2022 House model did not account for undecided voters/remaining votes with polls, so if you had a poll result of 49% and 49%, it would not split the remaining 2% up between the two candidates like with future models.

Race Ratings

The race ratings average both party's percentages across all categories, and then whoever has a higher probability based on the average is the predicted winner of that race. This gives much more accurate and numerical results than the old model which was not specific at all. You can view the ratings and see the correspondence to the categories won above by clicking the spreadsheet link. (green button)