Free beta access. 50 free AI tokens each week. Optional top-ups available. See what is included
← SearchMichigan2022Disadvantage
Opening Word document…

Five-thirty-eight polls good

Nate Silver 22, 6-30-2022, Silver is the founder and editor in chief of FiveThirtyEight."Why Republicans Are Favored To Win The House, But Not The Senate," FiveThirtyEight, https://fivethirtyeight.com/features/why-republicans-are-favored-to-win-the-house-but-not-the-senate/

Changes to the model Overall, we’re happy with our congressional and gubernatorial forecasts, which last underwent a major revision before the 2018 elections. They performed very well in 2018 and fairly well in 2020 (despite a challenging year for the polls in 2020; it helped that our model also considers a number of other factors in addition to the polling). Therefore, the overall methodology is largely the same. However, after assessing the performance of the models, we did make a few changes around the margins: In evaluating fundraising for congressional candidates, the model now places more emphasis on contributions received within the candidate’s state. Fundraising is a highly nationalized activity these days; Democrats in California and New York regularly contribute to Senate campaigns like that of Democrat Jaime Harrison of South Carolina, who raised a record amount of money in 2020 but nonetheless lost to Republican Sen. Lindsey Graham by 10 percentage points. As such, our analysis finds that a dollar received from within a candidate’s state is about five times as valuable as a dollar received outside of it in predicting the eventual election outcome.5 That’s why the model now applies that ratio, multiplying any funds raised from voters within the candidate’s state by 5, when assessing which candidate has the fundraising edge. The model now puts more emphasis on basic partisanship as opposed to other fundamentals. The fundamentals component of our model, used in the Classic and Deluxe versions, evaluates a series of variables such as fundraising, incumbency, candidate experience, congressional voting records and even scandals to make a forecast. The formula is fairly complicated, so we tested a simple alternative in an era of high partisanship. Specifically, we looked at what would happen if you made a forecast based solely on the partisan lean of a state or district, plus generic-ballot results. For instance, if the generic ballot favors Republicans by 4 percentage points, and a state leans Republican by 3 points, you might expect the Republican to win there by 7 points. In testing this on 2018 and 2020 races, we found that the more complicated version of our fundamentals formula does outperform this simple alternative. But it’s pretty close, and a blend of the complicated and simple formulas does better than either one taken alone.6 What this means is that state partisanship and the generic ballot will have more influence on our model going forward and the other fundamentals factors less, although they certainly matter some. We assume errors are more correlated from race to race. In presidential races, forecast and polling errors are highly correlated from state to state. It wasn’t a surprise that Trump won Wisconsin and Pennsylvania in 2016 given that he also won Michigan, for example, since the states are fairly similar demographically. (Accounting for this properly was a big reason that FiveThirtyEight’s model gave Trump a better chance than others in 2016.) This also holds in congressional races, though to a considerably lesser degree since there are different candidates on the ballot in each state. A scandal affecting the Republican candidate in Georgia won’t necessarily have any impact in Arizona, for instance. However, our analysis suggests that polling errors are becoming more correlated in congressional races, too. Certainly, 2020 was a prominent example given that Republicans won nearly every toss-up race for the House and that polls for Congress had a strong overall bias toward Democrats. Therefore, we’ve increased the degree to which race-by-race errors are assumed to be correlated in the model.7 We’ve reduced the emphasis on sample size when weighting polls. This is consistent with a change in our pollster ratings methodology and is explained in more detail there. We’ve removed a series of changes in the 2020 model that were specific to the COVID-19 pandemic. Yes, I know the pandemic isn’t over yet, but voting will take place under far more normal conditions than in 2020.

Sign in to recut, update, or save this card.