Klaus Pforr: "very interesting paper on caus…" - sciences.social

Recent searches

Search options

Only available when logged in.

Klaus Pforr @klauspforr@sciences.social

very interesting paper on causal inference methods https://muse.jhu.edu/article/895650 #causality #causalinference #causalfriday

abstract abbreviated by openai: Introducing new causal estimators often involves simulation studies by statisticians, but this approach has drawbacks. Simulation design biases favor the new method, unfavorable results may go unpublished, and comparing estimators is challenging. The American Causal Inference Conference (ACIC) data challenges provide an alternative. As organizers of the 2022 challenge, we created realistic data sets for policy evaluations with hidden causal impacts. Participating teams used advanced methods to estimate these effects. 20 teams submitted results from 58 estimators using various approaches. We discovered important factors driving performance not typically used in policy evaluations, suggesting ways to achieve more precise estimates. Top-performing methods employed flexible modeling of outcome-covariate and outcome-participation relationships, along with subgroup estimate regularization. Model-based uncertainty intervals outperformed bootstrap-based ones. Surprisingly, analyzing large-n patient-level data didn't improve performance compared to analyzing smaller-n data aggregated to the primary care practice level. In our simulated data, practices, not individuals, decided to join the intervention. Ultimately, this competition aimed to identify methods best suited for evaluating the impact of social policies on individuals and communities.

Jun 02, 2023, 08:03 AM··Web

11boosts·10favorites

Drag & drop to upload