Difference-in-Differences Meets Tree-based Methods: Heterogeneous Treatment Effects Estimation with Unmeasured Confounding
Authors: Caizhi Tang, Huiyuan Wang, Xinyu Li, Qing Cui, Longfei Li, Jun Zhou
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we conduct experiments on both simulated and real-world datasets to verify the effectiveness of our method. We evaluate Di D Causal Tree against state-of-the-art causal inference algorithms |
| Researcher Affiliation | Collaboration | 1Ant Group, Zhejiang, China 2School of Mathematical Sciences, Peking University, China 3Department of Biostatistics and Epidemiology, University of Pennsylvania, 210 Blockley Hall, 423 Guardian Drive, Philadelphia, Pennsylvania 19104, U.S.A.. |
| Pseudocode | No | The paper describes the proposed method in detail using mathematical formulas and descriptive text within the "Methodology" section, but it does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper provides links to third-party baseline code (e.g., "The BCF is from https://github.com/socket778/XBCF, GRF is from https://github.com/grf-labs/grf and the others are from https://github.com/microsoft/Econ ML."), but it does not provide an explicit statement or link for the authors' own implementation code for Di DTree or Gradient Boosting Di D-Tree. |
| Open Datasets | Yes | The dataset comes from a randomized controlled trial (RCT) by a commercial finance company aimed at assessing users heterogeneous responses to increasing credit lines of credit card (Tang et al., 2022). |
| Dataset Splits | Yes | The assignment of treatment is by a function ps(X, φ), where φ controls the ratio of the treated instances, i.e., φ = #control #treated+#control. ... generate a total of 20000 instances which are randomly split into training and validation sets by 10 times. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., CPU, GPU models, memory, cloud instances) used for running the experiments. |
| Software Dependencies | No | The paper mentions that baselines use third-party libraries but does not provide specific version numbers for these or for any other software dependencies used in their own experimental setup. |
| Experiment Setup | Yes | For example, the max number of trees in ensemble models (including boosting and bagging) is 500, the subsample ratios of instance and feature are 0.8 and 0.8, and the learning rate is 0.05. The max depth of each tree in forestbased (GRF and BCF) and boosting-based (meta-learners and Di DTree) methods are 10 and 3 respectively |