Learning Smooth Populations of Parameters with Trial Heterogeneity

TL;DR

This paper develops fast error rate estimators for smooth population parameters in heterogeneous binomial trials, improving existing methods and applying them to criminal justice data to compare conviction rates.

Contribution

It introduces new error bounds for kernel density estimators under trial heterogeneity and smoothness, advancing the estimation of mixing distributions in complex scenarios.

Findings

Improved error rates depend on the harmonic mean of trials.

Enhanced estimation accuracy over previous methods.

Application reveals higher conviction rates for appointed counsel.

Abstract

We consider the classical problem of estimating the mixing distribution of binomial mixtures, but under trial heterogeneity and smoothness. This problem has been studied extensively when the trial parameter is homogeneous, but not under the more general scenario of heterogeneous trials, and only within a low smoothness regime, where the resulting rates are slow. Under the assumption that the density is s-smooth, we derive fast error rates for the kernel density estimator under trial heterogeneity that depend on the harmonic mean of the trials. Importantly, even when reduced to the homogeneous case, our result improves on the state-of-the-art rate of Ye and Bickel (2021). We also study nonparametric estimation of the difference between two densities, which can be smoother than the individual densities, in both i.i.d. and binomial-mixture settings. Our work is motivated by an application in criminal justice: comparing conviction rates of indigent representation in Pennsylvania. We find that the estimated conviction rates for appointed counsel (court-appointed private attorneys) are generally higher than those for public defenders, potentially due to a confounding factor: appointed counsel are more likely to take on severe cases.