Methodology - 2024 U.S. Senate Elections Nowcast

Updated April 22, 2024 at 12:14 AM

View Full Nowcast (Google Sheets)

About the Nowcast

First off, what even is a Nowcast? It's basically just a model that simulates how the Senate elections would go if it were today, rather than how it will go on November 5th. By doing this, we get a more real-time idea of where different races stand today, and where polls indicate voters are leaning. I will be updating both the Nowcast and the forecast (once it's released), and both will help determine how the election is going for both parties. It's important to note that the Nowcast only uses polls, meaning it can't balance out any polling errors, and this is why it's less reliable. It's also more aggressive and confident in it's outcomes, because it assumes that polls today are as reliable as they should be in November, giving it less uncertainty. Below, I'll briefly explain how the Nowcast works.

Step 1: Create polling averages and state nowcasts

Given that the nowcast is solely reliant on polling, it's essential that I gather and average polls at the state levels in the various Senate races. This is explained further at the polling averages page. If a race has no polling yet, I use the 2018 results for that seat instead.

Step 2: Plug in state odds and run the simulation

This is actually pretty simple: I run a Monte Carlo simulation of 100,000 Senate outcomes by feeding the model each state's Democrat/Republican chances of winning, and it generates win percentages in terms of which party won more simulations, as well as the average seats for both Democrats & Republicans, the range of outcomes, and more. *Note: Although this is a Nowcast, given that uncertainty is fairly low on the state probabilities, I do add in some uncertainty in the simulation as a whole, basically telling it that in each simulation, the winning chances can deviate by up to 25%, assuming there can be error, which adds a bit of uncertainty to the Nowcast overall.

Step 3: Maps, visuals, tables, etc.

That's pretty much it! I then create all necessary maps and fill in tables and content on the model page, and update the charts, and that's the Nowcast! The forecast will be much more complex, and will have it's own methodology.