Tight Partial Identification of Causal Effects with Marginal Distribution of Unmeasured Confounders

Authors: Zhiheng Zhang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our theoretical findings are supported by experiments. We conduct synthetic and real-world experiments to quantify the information loss of the traditional entropy- based optimization for PI in various settings, compared with our proposed PI region.
Researcher Affiliation Academia Institute for Interdisciplinary Information Sciences, Tsinghua University, Beijing, China.
Pseudocode Yes Algorithm 1 Approximation TPI algorithm. Algorithm 2 SSP(I, I , η) algorithm.
Open Source Code No The paper does not contain any explicit statements about releasing source code or provide links to code repositories.
Open Datasets Yes We conduct experiments on INSURANCE dataset (Binder et al., 1997) and the ADULT dataset (Dua & Graff, 2017).
Dataset Splits No The paper describes the datasets used and how variables are treated, but does not specify any train/validation/test splits, proportions, or cross-validation methods.
Hardware Specification No The paper does not provide any specific details about the hardware used for running experiments, such as CPU/GPU models or memory specifications.
Software Dependencies No The paper does not list specific software components with version numbers required for reproducibility.
Experiment Setup Yes We assume that each generated data sample has only two parts of data information: P(X, Y ) and confounder information P(U). Specifically, we generate U with the same analogue as the previous: U Dir([0.1, 0.1, 0.1, 0.1, 0.1]), du = 5. Moreover, following the famous sampling procedure (Chickering & Meek, 2012), X, Y is generated by P(X | ui) Dir(v ), i = 0, ...du 1, P(Y | uj, xk) Dir(s ), j [0, 1, ...du 1], k [0, 1, ...|X| 1]. In the INSURANCE dataset, we treat car cost, property cost, and accident cost (other cars) as X, Y, U. Furthermore, in the ADULT dataset, we treat this triple as relationship (unmarried, in-family, etc), income and age. We follow the division method as before in Table 6. For instance, for AGE in the ADULT dataset, we choose 65 as the cutting point to separate it into two categories: young and old.