Based on Actual Math: New Hampshire Primary

This has become my go to "here's my methodology" post so I'm moving it to the top and the NH specific stuff down below.

Bayesian Methology

In case you didn't know, the New Hampshire Primary is tonight. Lots has been written about it already, and I'm sure lots more will be written.

I have my own small contribution to share: a model to better extrapolate results that have come in so far and project who will win, who will get 2nd, 3rd, fourth, and so on.

It's a Bayesian model that starts with each candidate's projected vote totals* and applies them to each county in New Hampshire.

First (like right now), the model is based only on the projected vote totals. Each candidate's odds of winning look like this:

Once votes start coming in the model learns how the county is actually voting and updates itself accordingly, balancing its prior expectation with that new information**.

As more and more votes come in the models leans on those votes more and more heavily and begins discarding its prior assumptions.

Then, when enough votes have come in the model extrapolates full county results using only the votes that have been cast thus far and no prior assumptions.

*I'm using 538s projected vote totals as my priors
**I'm using a beta distribution as my Bayesian prior updated by its conjugate distribution (binomial)

NH Results

I'm out! Gnight, thanks for following along. Time to dust off my South Carolina Model.