Indecision Modeling

Authors: Duncan C. McElfresh, Lok Chan, Kenzie Doyle, Walter Sinnott-Armstrong, Vincent Conitzer, Jana Schaich Borg, John P. Dickerson5975-5983

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We test these models using data collected from an online survey where participants choose how to (hypothetically) allocate organs to patients waiting for a transplant. Our Contributions. First, we conduct a pilot experiment to illustrate how different interpretations of indecision lead to different outcomes ( ). To test the utility of these models, we conduct a second experiment to collect a much larger dataset of decision responses ( ). We take a machine learning (ML) perspective, and evaluate each model class based on its goodness-of-fit to this dataset.
Researcher Affiliation Academia Duncan C Mc Elfresh,1 Lok Chan,2 Kenzie Doyle,3 Walter Sinnott-Armstrong,2 Vincent Conitzer,2 Jana Schaich Borg,2 John P Dickerson1 1 University of Maryland, College Park 2 Duke University 3 University of Oregon
Pseudocode No The paper describes mathematical models and processes but does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes All code used for our analysis is available online, 9 and details of our implementation can be found in Appendix ??. 9https://github.com/duncanmcelfresh/indecision-modeling
Open Datasets No The paper mentions data collected from an online survey for Study 1 and Study 2 and notes that the datasets collected in Study 1 and Study 2 provide some insight, but it does not provide concrete access information (link, DOI, specific repository, or formal citation with author/year) for these datasets to be publicly available.
Dataset Splits No For each participant we randomly split their question-response pairs into a training and testing set of equal size (20 responses each). No explicit mention of a validation set split is found.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper mentions 'scikit-learn' used for the MLP, citing Pedregosa et al. (2011), but does not specify its version number or other software dependencies with versioning.
Experiment Setup Yes We take a maximum likelihood estimation (MLE) approach to fitting each model: i.e., we select agent parameters u and λ which maximize the loglikelihood (LL) of the training responses. Since the LL of these models is not convex, we use random search via a Sobol process (Sobol 1967). The search domain for utility vectors is u [ 1, 1]N, the domain for probability parameters is (0, 1), and the domain for λ depends on the model type (see Appendix ??).