[2309.12833] Mannequin-based causal function choice for normal response varieties


Obtain a PDF of the paper titled Mannequin-based causal function choice for normal response varieties, by Lucas Kook and 4 different authors

Obtain PDF
HTML (experimental)

Summary:Discovering causal relationships from observational knowledge is a elementary but difficult process. Invariant causal prediction (ICP, Peters et al., 2016) is a technique for causal function choice which requires knowledge from heterogeneous settings and exploits that causal fashions are invariant. ICP has been prolonged to normal additive noise fashions and to nonparametric settings utilizing conditional independence checks. Nevertheless, the latter typically undergo from low energy (or poor sort I error management) and additive noise fashions are usually not appropriate for purposes through which the response will not be measured on a steady scale, however displays classes or counts. Right here, we develop transformation-model (TRAM) based mostly ICP, permitting for steady, categorical, count-type, and uninformatively censored responses (these mannequin lessons, usually, don’t enable for identifiability when there is no such thing as a exogenous heterogeneity). As an invariance check, we suggest TRAM-GCM based mostly on the anticipated conditional covariance between environments and rating residuals with uniform asymptotic degree ensures. For the particular case of linear shift TRAMs, we additionally take into account TRAM-Wald, which checks invariance based mostly on the Wald statistic. We offer an open-source R bundle ‘tramicp’ and consider our strategy on simulated knowledge and in a case research investigating causal options of survival in critically unwell sufferers.

Submission historical past

From: Lucas Kook [view email]
Fri, 22 Sep 2023 12:42:48 UTC (266 KB)
Fri, 6 Oct 2023 14:27:37 UTC (280 KB)
Thu, 14 Mar 2024 09:28:12 UTC (286 KB)

Supply hyperlink


Please enter your comment!
Please enter your name here