Learning Treatment Effects in Panels with General Intervention Patterns

Authors: Vivek Farias, Andrew Li, Tianyi Peng

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Computational experiments on synthetic and real-world data show a substantial advantage over competing estimators. We show both for synthetic and real data that our estimator provides a material improvement in empirical performance relative to available alternatives
Researcher Affiliation Academia Vivek F. Farias Sloan School of Management MIT Cambridge, MA 02139 vivekf@mit.edu Andrew A. Li Tepper School of Business Carnegie Mellon University Pittsburgh, PA 15213 aali1@cmu.edu Tianyi Peng Department of Aeronautics and Astronautics MIT Cambridge, MA 02139 tianyi@mit.edu
Pseudocode No The paper describes the algorithm using mathematical equations (1a) and (1b) and accompanying text, but does not provide a formal pseudocode block or a clearly labeled algorithm section.
Open Source Code No The paper does not contain any explicit statement about releasing source code or provide a link to a code repository.
Open Datasets Yes The first dataset consists of the annual tobacco consumption per capita for 38 states during 1970-2001, collected from the prominent synthetic control study [2] (the treated unit California is removed). The second dataset consists of weekly sales of 167 products over 147 weeks, collected from a Kaggle competition [41]. This dataset consists of daily sales and promotion information of 571 drug stores over 942 days, collected from Rossmann Store Sales dataset [42].
Dataset Splits No The paper states hyperparameters were 'tuned using rank r 5 (estimated via the spectrum of M )' and 'estimated via cross validation' without specifying the exact split percentages, sample counts, or the explicit cross-validation methodology (e.g., k-fold, leave-one-out).
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., GPU/CPU models, memory, or specific computing clusters) used for running the experiments.
Software Dependencies No The paper does not provide specific software dependencies or version numbers for any libraries, frameworks, or specialized packages used in the experiments.
Experiment Setup Yes The hyperparameters for all algorithms were tuned using rank r 5 (estimated via the spectrum of M ). We fix τ = σδ = M /5 through all experiments. We considered an ensemble of 1,000 instances with m1 Uni[1, n1), m2 = Uni[1, n2) for stagger patterns and m1 Uni[1, 5), m2 = 18 for block patterns. A test set Ωconsisting of 20% of the treated entires is randomly sampled and hidden.