Debiased Bayesian inference for average treatment effects

Authors: Kolyan Ray, Botond Szabo

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We numerically illustrate our method on simulated and semi-synthetic data using GP priors, where our prior correction corresponds to a simple data-driven alteration to the covariance kernel. Our experiments demonstrate significant improvement in performance from this debiasing.
Researcher Affiliation Academia Kolyan Ray Department of Mathematics King s College London kolyan.ray@kcl.ac.uk Botond Szabó Mathematical Institute Leiden University b.t.szabo@math.leidenuniv.nl
Pseudocode Yes Algorithm 1 Debiased GP with PS correction
Open Source Code No The paper mentions using the "GPy package" but does not provide a link to its own source code or state that it is open-source.
Open Datasets Yes We consider a semi-synthetic dataset with real features and treatment assignments from the Infant Health and Development Program (IHDP), but simulated responses. The IHDP consisted of a randomized experiment studying whether low-birth-weight and premature infants benefited from intensive high-quality child care. The data contains d = 25 pretreatment variables per subject. Following [18] (also used in [2, 21]), an observational study is created by removing a non-random portion of the treatment group, namely all children with non-white mothers.
Dataset Splits No The paper describes the datasets used (synthetic and IHDP) and their generation, but it does not specify explicit train/validation/test splits (e.g., percentages or counts) or cross-validation details for the models trained and evaluated. It mentions running simulations 200 times but not how data is partitioned within each run for model training and evaluation.
Hardware Specification No The paper does not provide any specific details about the hardware used for running the experiments.
Software Dependencies No The paper mentions using the "GPy package" and the "scaled conjugate gradient method option in the GPy package" but does not provide version numbers for GPy or any other software dependencies.
Experiment Setup Yes We optimize the hyperparameters (ℓi)d+1 i=1 , ρm and σn (noise variance) by maximizing the marginal likelihood (using the scaled conjugate gradient method option in the GPy package). We set νn = 0.2ρm/( n Mn) for Mn = n 1 Pn i=1[Ri/ˆπ(Xi) + (1 Ri)/(1 ˆπ(Xi))] the average absolute value of the last part of (5).