[ABE-L] Data Science Seminar - Insper

[Hedibert Lopes]

May 05, 2022
12:00 p.m. de São Paulo


Title: Targeted smooth Bayesian causal forests with basis functions: an application to evaluate mindset interventions
 
Speaker: Pedro H. F. Santos

University: The University of Texas at Austin
 
 
Abstract: Among the methods used to estimate nonlinear functions, Bayesian Additive Regression Trees (BART) priors have shown promising results, especially for prediction, being used in a wide range of different problem settings, including causal inference and variable selection. BART priors represent functions as the sum of many small trees, each parameterizing a step function. However, smoothness is often desirable for parsimony, accurate interpolation, and extrapolation, while keeping the efficiency. We expanded the BART model by replacing locally constant predictions with Gaussian Processes using vectors of basis coefficients attached to the leaves. We use these models in a randomized controlled trial to evaluate the effect of mindset intervention as a way to protect adolescents from the consequences of social stressors.
 

-------------- Pr�xima Parte ----------
Um anexo em HTML foi limpo...
URL: <http://lists.ime.usp.br/pipermail/abe/attachments/20220504/ff7eefc6/attachment.htm>