Causal Effect Identification in lvLiNGAM from Higher-Order Cumulants
Authors: Daniele Tramontano, Yaroslav Kivva, Saber Salehkaleybar, Negar Kiyavash, Mathias Drton
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate the accuracy and robustness of our approaches compared to existing methods, advancing the theoretical and practical understanding of causal inference in linear systems with latent confounders. This section presents experimental results on synthetic and experimental data for the graphs studied in Section 3. |
| Researcher Affiliation | Academia | 1Technical University of Munich, Munich, Germany 2Ecole Polytechnique F ed erale de Lausanne, Lausanne, Switzerland 3Leiden Institute of Advanced Computer Science, Leiden University, Netherlands 4Munich Center for Machine Learning, Munich, Germany. Correspondence to: Daniele Tramontano <EMAIL>. |
| Pseudocode | Yes | Algorithm 1 Proxy Variable (Fig. 2)..., Algorithm 2 Proxy Variable with an Edge from Proxy to Treatment (Fig. 3)..., Algorithm 3 Proxy Variable with an Edge from Proxy to Treatment with one Latent (Fig. 3)..., Algorithm 4 Proxy Variable with an Edge from Proxy to Treatment with one Latent with Optimization (Fig. 3)..., Algorithm 5 Underspecified Instrumental Variables (Fig. 8)..., Algorithm 6 Underspecified Instrumental Variables with Multiple Instruments (Fig. 8)... |
| Open Source Code | Yes | The code to replicate the computation can be found at https://github.com/danieletramontano/CEId-from-Moments/blob/main/Macaulay2/Non-Gaussian-Identifiability.m2. The code to replicate the experiments can be found at https://github.com/danieletramontano/CEId-from-Moments. |
| Open Datasets | Yes | To assess the practical efficacy of our method, we conduct experiments on the dataset analyzed in Card & Krueger (1993), which contains information on fast-food restaurants in New Jersey and Pennsylvania in 1992. Card, D. and Krueger, A. B. Minimum wages and employment: A case study of the fast food industry in new jersey and pennsylvania, 1993. |
| Dataset Splits | No | The original study aimed to estimate the effect of an increase in New Jersey s minimum wage from $4.25 to $5.05 per hour on employment rates. Importantly, the data were collected both before and after the wage increase in New Jersey, while the minimum wage in Pennsylvania remained constant throughout this period. All the experiments in this subsection are done on the synthetic data generated according to the specific causal structure established for it. The paper describes data generation and experimental design for real-world data but does not specify explicit training/test/validation splits of a single dataset. |
| Hardware Specification | No | No specific hardware details (GPU/CPU models, memory, or cloud instances) are mentioned in the paper's experimental sections or appendix. |
| Software Dependencies | No | The computations were done using the computer algebra software Macaulay 2 (Grayson & Stillman, 2023). In practice, we solve the optimization problem using the Python implementation of the BFGS algorithm (Nocedal & Wright, 2006, 6.1) provided in Jones et al. (2001 ). Specific version numbers for software components like Macaulay 2, Python, or SciPy are not provided. |
| Experiment Setup | Yes | To generate synthetic data, we specify all exogenous noises from the same family of distributions (with parameters sampled according to Table 1) and select all non-zero entries within the matrix A through uniform sampling from [ 0.9, 0.5] [0.5, 0.9]. Table 1. Summary of the experimental setups. Figure Causal Graph Distribution Parameters of Interest Family shape scale ... Family alpha beta |