Control Function Instrumental Variable Estimation of Nonlinear Causal Effect Models
Authors: Zijian Guo, Dylan S. Small
JMLR 2016 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The pretest estimator is shown to work well in a simulation study. An application to the effect of exposure to violence on time preference is considered. Our simulation study compares the usual two stage least squares (2SLS) estimator, the control function (CF) estimator and the pretest estimator. We will consider several different model settings where the assumptions of the control function estimator are satisfied, moderately violated (two settings) and drastically violated and we will also investigate the sensitivity of the control function estimator to different joint distributions of errors. |
| Researcher Affiliation | Academia | Zijian Guo EMAIL Department of Statistics University of Pennsylvania Philadelphia,19104, USA Dylan S. Small EMAIL Department of Statistics University of Pennsylvania Philadelphia,19104, USA |
| Pseudocode | Yes | Algorithm 1 Two stage least squares estimator (2SLS) ... Algorithm 2 Control function estimator (CF) ... Algorithm 3 Pretest estimator |
| Open Source Code | No | The paper does not provide concrete access to source code. There are no explicit statements about code release, links to repositories, or mentions of code in supplementary materials. |
| Open Datasets | Yes | The data set consists of 302 observations from a field experiment in Burundi (Voors et al (2012)). The data set is from Bouis & Haddad (1990) and comes from a survey of farm households in the Bukidnon Province of Philippines. The data set consists of 1388 observations from the Child Health Supplement to the 1998 National Health Interview Survey. |
| Dataset Splits | No | The paper mentions a sample size of "10,000" for the simulation study and specific observation counts for the real-world applications (302 for Burundi, 1388 for smoking data), but it does not specify how these datasets were split into training, validation, or test sets for experimental purposes. |
| Hardware Specification | No | The paper does not provide specific hardware details used for running its experiments. No mentions of CPU, GPU, or other accelerator models are found. |
| Software Dependencies | No | The paper does not provide specific ancillary software details with version numbers. It describes statistical methods and models but does not mention any software packages, libraries, or solvers with their versions. |
| Experiment Setup | Yes | The sample size is 10,000 and we implement 10,000 simulations for each setting. The outcome model that we are considering is y1 = β0 + β1z1 + β2y2 + β3y2 2 + u1; In the following subsections, we will generate y2 according to different models. For the setting where the assumptions of control function estimator are satisfied, the model we consider is: y1 = 1 + z1 + 10y2 + 10y2 2 + u1; y2 = 1 + z1 + 1/8z2 + 1/8z2 2 + v2; where z1 ∼ N(0, 1) and z2 ∼ N(0, 1) and u1, v2 ~ N(0, [1 0.5; 0.5 1]). |