Key points: A popular "hazard difference" estimator for averted outcomes is biased, but an unbiased one can be calculated from status-based aggregated data (knowing y/n vaccinated for those with and without the outcome). For avertible outcomes, this doesn't work; need to know dates of vaccination.
cisac.fsi.stanford.edu/content/cisa... In my new role at @stanfordcisac.bsky.social, I'm recruiting for a postdoc to define how benefits of risky research should be evaluated. w Tony Mills of @notredame.bsky.social . #philsci #biosecurity #scipolicy
New preprint reporting a large piece of work by Katherine Jia to define estimands and estimators for direct impact of vaccines with arbitrary rollout schedules arxiv.org/abs/2509.05508
with Alyssa Bilinski, @chrisboyer.bsky.social and me