Strategic Instrumental Variable Regression: Recovering Causal Relationships From Strategic Responses
Authors: Keegan Harris, Dung Daniel T Ngo, Logan Stapleton, Hoda Heidari, Steven Wu
ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We empirically evaluate our model on a semi-synthetic dataset inspired by our running university admissions example. We compare our 2SLS-based method against ordinary least squares (OLS), which directly regresses observed outcomes y on observable features x. |
| Researcher Affiliation | Academia | 1School of Computer Science, Carnegie Mellon University, Pittsburgh, USA 2Computer Science Department, University of Minnesota, Minneapolis, USA. |
| Pseudocode | No | The paper describes the steps for Two-Stage Least Squares regression (2SLS) in Section 2, including derivations in Appendix A.2, but it does not provide any explicitly labeled pseudocode blocks or algorithm figures. |
| Open Source Code | No | The paper provides a link to a publicly available dataset ('https://www.openintro.org/data/ index.php?data=satgpa') but does not include any explicit statement or link for the source code of the methodology described in the paper. |
| Open Datasets | Yes | We constructed a semi-synthetic dataset based on the SATGPA dataset, a collection of real university admissions data. ... Originally collected by the Educational Testing Service, the SATGPA dataset is publicly available and can be found here: https://www.openintro.org/data/ index.php?data=satgpa. |
| Dataset Splits | No | The paper describes the construction and characteristics of its semi-synthetic dataset, including how data points are generated, but it does not provide explicit details about how the dataset is split into training, validation, or test sets (e.g., percentages, sample counts, or cross-validation setup). |
| Hardware Specification | Yes | We ran our experiments on a 2020 Mac Book Air laptop with 16GB of RAM. |
| Software Dependencies | No | The paper describes its methodology and experimental setup but does not list any specific software dependencies (e.g., programming languages, libraries, or frameworks) along with their version numbers. |
| Experiment Setup | Yes | For simplicity, we let the true effect θ = [0, 0.5] . ... For any applicant t, we randomly deploy assessment rule θt = [θSAT t , θHS GPA t ] where θSAT t N(1, 10) and θHS GPA t N(1, 2). ... For each applicant t, we perturb E[Et] with random noise drawn from N(0.5, 0.25) to the top left entry and noise drawn from N(0.1, 0.01) the bottom right entry to produce Et. ... We ran each method for 1000 time-steps with a decaying learning rate of 0.001 / T. |