Oliver Hines @ohines.bsky.social · 11/05/2026Of course the other solution is to regress the outcome on the predictors, but that controls the bias-variance trade off in the observed not intervention distribution, so we need the same weights if we want to debias using e.g. one-step or TMLE 010
Oliver Hines @ohines.bsky.social · 11/05/2026Preview slide for my ACIC talk tomorrow: The fundamental problem of observational causal inference is that the observed data distribution is different to the intervention distribution. The main solution is to learn some density ratios (importance weights) and reweight! 140
Oliver Hines @ohines.bsky.social · 13/04/2026I must congratulate you on a successful longitudinal intervention! 020
Oliver Hines @ohines.bsky.social · 03/04/2026The bootstrap has got to be a prime example. Multinomial weights - frequentist, Dirichlet weights - Bayesian?? 230
Reposted by Oliver HinesKara Rudolph @kararudolph.bsky.social · 24/11/2025Mind meld complete—these two showing up the same day with twin buzz cuts @calebhmiles.bsky.social @ohines.bsky.social 061
Oliver Hines @ohines.bsky.social · 31/10/2025classical is anything that predates the start of my PhD and the older I get then the more classical it was 030
Oliver Hines @ohines.bsky.social · 21/10/2025New paper with @calebhmiles.bsky.social on density ratio learning! 071
Oliver Hines @ohines.bsky.social · 20/10/2025when the information matrix is singular then should it be called the 'bagel variance' (ie a sandwich with a hole)? 020
Oliver Hines @ohines.bsky.social · 11/10/2025Tbh I never understood midterm exams in the US. Why not just two weeks of exams at the end of the academic year? I always felt that I only really understood the content when revising on my own during the spring break 🤷🏼♂️ 000
Oliver Hines @ohines.bsky.social · 03/04/2025This week I started a new job as a postdoc at Columbia University! Excited to be back doing research and exploring NYC even with all the political madness that is going on 0100
