Long-term Causal Effects via Behavioral Game Theory

Authors: Panagiotis Toulis, David C. Parkes

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we apply our methodology to experimental data from Rapoport and Boebel [18], as reported by Mc Kelvey and Palfrey [15].
Researcher Affiliation Academia Panagiotis (Panos) Toulis Econometrics & Statistics, Booth School University of Chicago Chicago, IL, 60637 panos.toulis@chicagobooth.edu David C. Parkes Department of Computer Science Harvard University Cambridge, MA, 02138 parkes@eecs.harvard.edu
Pseudocode Yes Algorithm 1 Estimation of long-term causal effects Input: Z, T, A, B, G1, G0, D1 = {a1(t; Z) : t = 0, . . . , t0}, D0 = {a0(t; Z) : t = 0, . . . , t0}. Output: Estimate of long-term causal effect CE(T) in Eq. (1).
Open Source Code No The paper does not provide any statement about releasing source code or a link to a code repository.
Open Datasets No The paper uses "experimental data from Rapoport and Boebel [18], as reported by Mc Kelvey and Palfrey [15]". While these are cited papers, no direct access information (link, DOI, specific repository) for the dataset itself is provided, nor is it explicitly stated to be a well-known public dataset with clear access.
Dataset Splits Yes To evaluate our method, we consider the last period as long-term, and hold out data from this period.
Hardware Specification No The paper does not provide any specific details about the hardware used for running its experiments.
Software Dependencies No The paper does not specify any software dependencies with version numbers.
Experiment Setup Yes We choose diffuse priors for our parameters, specifically, φ U(0, 10), ψ U( 5, 5), and λ U( 10, 10). Given φ we sample the initial behaviors as Dirichlet, i.e., β1(0; Z) Dir(φ) and β0(0; Z) Dir(φ), independently. As the temporal model, we adopt the lag-one vector autoregressive model, also known as VAR(1). For the behavioral model, we adopt the quantal p-response (QLp) model [20], which has been successful in predicting human actions in real-world experiments [22]. We choose p = 3 behaviors, namely B = {b0, b1, b2} of increased sophistication parametrized by λ = (λ[1], λ[2], λ[3]) R3.