Instrumental Variable Estimation of Average Partial Causal Effects

Authors: Yuta Kawakami, Manabu Kuroki, Jin Tian

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We illustrate them on synthetic and real-world data.
Researcher Affiliation Academia 1Department of Mathematics, Physics, Electrical Engineering and Computer Science, Yokohama National University, Yokohama, Kanagawa, JAPAN 2Department of Computer Science, Iowa State University, Ames, Iowa, USA.
Pseudocode Yes Algorithm 1 Nonparametric APCE (N-APCE) estimator. Algorithm 2 Parametric APCE (P-APCE) estimator.
Open Source Code No The paper does not contain any explicit statement or link indicating that the source code for the described methodology is open-sourced or publicly available.
Open Datasets Yes We take up an open dataset in the R package wooldridge (https://cran.r-project. org/package=wooldridge), which was analyzed by Griliches (1977) and Blackburn & Neumark (1992).
Dataset Splits Yes To determine the best degree of the model, we separate the data set into training set D and validation set D , estimate ˆθ by the training set, and evaluate the trained model using the performance measure (19).
Hardware Specification No The paper does not specify any particular hardware (e.g., CPU, GPU models, or cloud computing instances with detailed specifications) used for running the experiments.
Software Dependencies No The paper mentions 'R package wooldridge' but does not provide a specific version number. No other software components are mentioned with version details.
Experiment Setup Yes Settings of N-APCE (Algorithm 1) We let X = {0, 0.3, . . . , 2.7, 3}; and the N-APCE estimator at X = 0 is not defined since x0 is 0. We calculate the numerical integration using the left-hand rule. We let the initial function ˆθ1 be a zero function, and the stop threshold ϵ be 10. We choose the step size as the smallest one from (1, 0.5, 0.1, . . .) when Algorithm 1 stops before 100 iterations, and the chosen step size α is 0.5. Settings of P-APCE (Algorithm 2) We use the polynomial basis functions ϕp(x) = xp 1 for p = 1, 2, . . ., and calculate the solution of the equation (18) by ( ˆDT ˆD) 1 ˆDT ˆc. To determine the best degree of the model, we separate the data set into training set D and validation set D , estimate ˆθ by the training set, and evaluate the trained model using the performance measure (19). From the results (shown in Table 5 in the appendix), we decide that the highest degree of the polynomial functions in the P-APCE estimator will be 3, when the mean of the performance measure is the smallest.