🚨MODEL DROP🚨: @notkavi and I have built a model (@bot_2024) that will provide a host of possible 2024 maps based on demographics, coalitions, the continuation/reversal/acceleration of differing trends, and varying national environments. It will tweet one map a day. 
We estimate demographics by taking ACS estimates and projecting 2024 totals based on demographic trends. Electoral vote totals are per @270toWin’s estimate of reapportionment after the census. Close states (<0.5% margin) are made a lighter shade of the winning party's color.
By perturbing specific feature weights, we (largely) preserve county correlations, which should avoid some ridiculous maps. The purpose of the exercise is not to predict, but to showcase a range of scenarios on what 2024 could look like.
The model emits maps at a rate of 54% Democratic wins, 46% Republican wins, which is mainly due to it being centered around the 2020 election. That's not to say it's our 2024 forecast; just that it is the rate at which the model will emit winning maps for each party by its design
Do not take this as a predictive model. We do not know the national environment and we do not know the candidates. But what we are showing is the variety of outcomes and coalitions in 2024 based on underlying factors and trends, given how predictive demographics are.
The (immense amount of) programming was largely done by @notkavi (go follow him!). The feature design and the demographic estimation/coalition projection approach was devised by @lxeagle17 and @notkavi. The data was provided by the ever-helpful @Mill226.
You can find the code for the model here github.com/kavigupta/2024…. Any suggestions/improvement fixes are welcome!
• • •
Missing some Tweet in this thread? You can try to
force a refresh
