Allocation Requires Prediction Only if Inequality Is Low

Authors: Ali Shirali, Rediet Abebe, Moritz Hardt

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We present a simple mathematical model to compare prediction-based versus unit-based allocations. We show that prediction leads to superior allocations only when between-unit inequality is low, and the allocation budget is high. (See Fig. 1 for a high-level view of inequality and budget regimes covered in our results.) Our analyses cover a broad range of settings for the price of prediction, treatment effect heterogeneity, and unit-level statistics learnability. Figure 3. ULA outperforms ILA in a simulated setting with high inequality. Figure 4. ULA outperforms ILA in a real-world high inequality setting. Eight school districts from the greater Los Angeles (LA) area are considered.
Researcher Affiliation Academia 1University of California, Berkeley 2Harvard Society of Fellows 3Max Planck Institute for Intelligent Systems, T ubingen and T ubingen AI Center.
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not include a statement or link indicating the release of open-source code for the methodology described.
Open Datasets Yes Here, we utilize the American Community Survey Children s Education Tabulation,2 an annually updated custom data collection of demographic, economic, social, and housing characteristics about school-age children and their families, developed from the U.S. Census Bureau s 2017-2021 data. 2https://nces.ed.gov/programs/edge/Demographic/ACS
Dataset Splits No The paper does not provide specific details regarding training, validation, and test dataset splits.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments or simulations.
Software Dependencies No The paper does not provide specific software dependency details, such as library names with version numbers, needed to replicate the experiments.
Experiment Setup Yes We consider a similar parameter setting as in the synthetic data example with δ = 0.3, q = 0.3, q 0, p(ϵ) = 0.2 B, and a budget B sufficient to treat half of the units.