New from me in @WEPsocial (open access): . Fragmentation revisited: the UK General Election of 2024. 4,000 words on the who, what, where, why of the 2024 UK General Election - the most fragmented election in British democratic history.

If you're interested in doing a PhD in something to do with elections, political psychology, public opinion, or anything to do with British politics (or know someone else who is), then get in touch - we have an open funding competition at the moment!.

All this and more - open access, so free to read - here:

Unsurprisingly then, the relationship between Reform performance and Conservative performance in 2024 is pretty obvious - in the places where Reform did the best, the Conservative vote went down a lot!

Reform 2024 and UKIP 2015 supporters were pretty similar, but not identical. Reformers were a bit more educated, less retired, and better off. Most importantly, Reformers were much more likely have switched directly from the Conservatives - more than 3/4 came from 2019 Cons.

Using a unique dataset of redistricted election results, we look at the factors associated with Reform UK performance at the 2024 election and compare it to UKIP in 2015 - the similarity between the two are pretty obvious:

New from @olhe, me, Humphrey Southall and Paula Aucott (@gbhgis ) in @po_qu: . The 2024 General Election and the Rise of Reform UK

RT @jon_mellon: New paper w/ @caprosser at @apsrjournal. A recent paper claim rising mass polarization in US is actually declining survey c….

RT @apsrjournal: Just published on APSR First View: "Regularized Regression Can Reintroduce Backdoor Confounding: The Case of Mass Polariza….

Along the way we touch things of wider relevance to social scientists like the equivalence of informative priors and different types of regularization, and the utility of simulation as part of our workflows. Check it out ⬇️⬇️⬇️

We argue that a recent claims that survey non-response has inflated estimates of mass polarization ( are driven by the use of ridge regression. See this thread from @jon_mellon

Because there is residual variance correlated with z, it can be accounted for by increasing the coefficient for z, so rather than regularization shrinking the z coefficient, it can *inflate* it. Having closed the backdoor path by covariate adjustment, regularization reopens it.

If you add regularization, e.g. by using ridge regression, this will shrink the coefficient for x. The consequence of this is that were will be residual variance that is *still correlated* with x, and because it is correlated with x it will be correlated with z.

If you were to estimate the model:.y ~ z.You would spuriously find an association between y and z, but you can block this backdoor path by estimating:.y ~ x + z. So far so straightforward.

Imagine you have a DAG that looks like this, which has a backdoor path between Z and Y that runs through X:

New from @jon_mellon and me in @apsrjournal looking at the impact that regularization (e.g. Ridge regression, LASSO etc) can have on adjusting for confounders, with particular application to claims about the effect of survey non-response on estimates of polarization. Thread⬇️.

RT @JeanMonnetRHUL: Looking forward to chairing this evening's round table on Populism and the British and European Elections of 2024 at @R….

Grant me the confidence of the peer reviewer who doesn’t know what a marginal effect is but thinks they’re qualified to offer statistical advice.

(Shame about the weather).