Introducing Dark Horse Politics and Josh Taft's 2024 Election Model

Announcing the launch of a new newsletter and model for the 2024 election by my friend Josh Taft.

Editor’s note: Every election cycle, there’s always someone who comes from nowhere to completely change the game of analysis. In 2022, Josh Taft was that person. When nearly everyone in politics was predicting a red wave, he came out with his own model that boldly, and correctly, called the election as a dead heat. Now, he’s back with a new model and substack newsletter to cover the 2024 election. I’m helping launch his newsletter here, but I recommend you all subscribe to his page as soon as you can. If his track record this year is anything like his record in 2022, there won’t be any better commentators to follow.

— Ettingermentum

Hello! I’m Josh Taft.

Before I go any further, I want to thank Ettingermentum for giving me the opportunity to write this guest article on his substack. If you are reading this and haven’t already, you should subscribe to his newsletter.

A year and a half ago, I created a model to try to predict the 2022 election. Starting from a simple Google Sheet, I was eventually able to create something I felt had some value to it. At that point, I decided to share it with the Election Twitter community. It got a really good reception from a lot of people in the community (and for me to say thanks, you all should check out people like Ettingermentum and Lakshya Jain on Twitter/X, to name a few). The model ended up doing amazing, with it predicting an environment of R+0.6 while the adjusted House popular vote was about R+1.6. Now, however, I can gladly say that I am launching a model for the 2024 election! It is very similar to my previous 2022 model—using special election results, generic ballot polling, and partisan primary composition to calculate a measurement for the national environment. 

This is done through each category giving out a partisan lean (such as a generic ballot polling of D+2), with weight given through previous accuracy of the variable. For example, if approval polling were used, the current president’s approval rating would be converted to a partisan lean, with the -7 approval rating of a Republican president translating to a D+7 output. It would then be weighted by the accuracy of approval ratings to previous election results—such as Trump’s approval rating relative to the 2018 midterms results. Once all variables are calculated and correspondingly weighted, they are combined to give a model output. Right now, it predicts a D+0.8 environment if the election was held today, which is essentially a tie, although it gives Democrats the slightest edge.

Currently, there is no data to input for special elections (specifically congressional, though this will change next week) or primaries. To compensate for this, the model uses the historical average as a replacement until live data is available. However, this is only temporary, as data should be coming in soon for both categories. 

Like its predecessor, the 2024 model also has prior elections built into it, giving predictions for previous elections. It does this by using the corresponding data that would have been available to it before each election happened and giving an output off of that—essentially the current model had it been built at those points. For comparison, the actual results are also shown, giving a sense of how far off each prediction was. 

With this, we can see the model has an R² of ~.95, with it being under just about half a point off in four of the past seven elections, while the other three are off by 2.2, 2.7, and 4.1 points (the 2020, 2018, and 2010 elections, respectively). The partisan error is also split, with some elections underestimating Republicans (mainly 2018 and 2020), while others had underestimated Democrats (namely 2010). This is a massive improvement over my previous model, which had an R² of .93 and was usually off by about 2 points on average. 

I’m looking forward to this election cycle, as there is much ahead to see. Due to this, in addition to my model, I have decided to launch a substack named Dark Horse Politics. There, I will write analyses of various political events and topics, occasionally giving my opinions on matters of more subjectivity. Additionally, I will also be giving out routine updates about the model, seeing how the environment shifts over time. 

To give a taste of what’s to come, here’s a short preview of my first non-model article:

---

Democrats’ Senatorial Checklist

It is a well-known fact that Democrats’ chances of maintaining their Senate majority come this November are slim. They are on the defense in almost every major battleground, with incumbents in red states such as Montana and West Virginia (the latter of which has already been given up through Senator Joe Manchin’s retirement). Their only real offensive target is Texas, a quickly left-trending GOP stronghold. However, the Democratic Party might just catch a lucky break—which, given the history of this class of Senate Democrats with the 2018, 2012, and 2006 elections, is quite frequent—through Republicans dropping the ball left and right. Here’s a checklist of things Democrats need to go right in order to hold their fragile majority in the US Senate…

---

Every election people say it will be like nothing before, and there’s a bit of truth to that sometimes. However, if one were to look in the right places, and look at them with the correct eyes, they can find the answers they seek. That is why I am doing both my model and substack, as to try to help people see through all the data and information. In other words, “Finding clarity in chaos”, or the dark horses of politics. I am happy to start this journey, and I hope you all come along with me. Cheers!