Unit Selection with Nonbinary Treatment and Effect

Authors: Ang Li, Judea Pearl

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we show the quality of the bounds of the benefit function obtained by Theorem 2 using four common benefit vectors. First, we set m = 2 (i.e., X has two values) and n = 3 (i.e., Y has three values). ... We randomly generated 1000 populations where each population consists of different fractions of nine response types of individuals. For each population, we then generated sample distributions (observational data and experimental data) compatible with the fractions of response types (see the appendix for the generating algorithm). ... For a sample population i, let [ai, bi] be the bounds of the benefit function from the proposed theorem. We summarized the following criteria for each population as illustrated in Figure 1: lower bound : ai; upper bound : bi; midpoint : (ai + bi)/2; real benefit : dot product of the benefit vector and the fractions of response types; Table 1: Experimental data of the clinical study. Here, 600 people were forced to take the vaccine and 600 people were forced to take no vaccine. Table 2: Observational data of the clinical study. Here, 1200 people were free to access the vaccine. 953 people chose to take the vaccine and 247 people chose to take no vaccine.
Researcher Affiliation Academia 1Department of Computer Science, Florida State University 2Cognitive Systems Laboratory, Department of Computer Science, University of California, Los Angeles angli@cs.fsu.edu, judea@cs.ucla.edu
Pseudocode Yes Algorithm 1: Check identifiability of the benefit function; Algorithm 2: Compute the bounds of the benefit function
Open Source Code No The paper does not contain any statement about releasing source code or a link to a code repository for the methodology described.
Open Datasets No The paper describes experimental and observational data from a clinical study summarized in Tables 1 and 2, and details a data generation process for simulations. However, it does not provide concrete access information (link, DOI, repository, or formal citation with access details) for these datasets.
Dataset Splits No The paper uses experimental and observational data, and also describes a data generation process for simulations, but it does not specify any train/validation/test splits, percentages, or sample counts for reproducibility.
Hardware Specification No The paper does not provide any specific hardware details such as GPU/CPU models, memory, or cloud instance types used for running its experiments.
Software Dependencies No The paper does not specify any software names with version numbers or particular solver versions required to reproduce the work.
Experiment Setup No The paper defines the problem and benefit function, and describes data generation for simulations, but it does not provide specific experimental setup details such as hyperparameters, optimizer settings, or other concrete training configurations.