Blog

Here we detail the actual electoral predictor itself: vote flows, simulated uncertainty, and a tactical-voting system, all to produce some hopefully-defensible seat-by-seat forecasts

As a country, we do tend to obsess over how downtrodden or genteel a neighbourhood, town or city is. Here, we try to quantify that obsession, to gauge the value of a house by its objectively measured 'affluence'. And as per the timeless national joke, London does its own thing

With Uniform National Swing (UNS) as dead as a dodo, every UK election model eventually has to decide what a constituency ‘is,’ part 1 found that doing so with a single label was about as accurate as describing a person by their star sign. Here we stop doing that, and start letting each of Britain's 632 seats be several different kinds of place at once.

I threw the kitchen sink of clustering methods at Britain's constituencies: K-means, hierarchical clustering, DBSCAN, GMM, run across two different embeddings and scored four different ways, in the hope that at least one of them would tell me something definite. None of them quite did, and this is the post where I admit that in some detail.