Search for: All records

Abstract Bayesian Improved Surname Geocoding (BISG) is a ubiquitous tool for predicting race and ethnicity using an individual’s geolocation and surname. Here we demonstrate that statistical dependence of surname and geolocation within racial/ethnic categories in the US results in biases for minority subpopulations, and we introduce a raking-based improvement. Our method augments the data used by BISG—distributions of race by geolocation and race by surname—with the distribution of surname by geolocation obtained from state voter files. We validate our algorithm on state voter registration lists that contain self-identified race/ethnicity. 

Abstract Label bias occurs when the outcome of interest is not directly observable and instead, modelling is performed with proxy labels. When the difference between the true outcome and the proxy label is correlated with predictors, this can yield systematic disparities in predictions for different groups of interest. We propose Bayesian hierarchical measurement models to address these issues. When strong prior information about the measurement process is available, our approach improves accuracy and helps with algorithmic fairness. If prior knowledge is limited, our approach allows assessment of the sensitivity of predictions to the unknown specifications of the measurement process. This can help practitioners gauge if enough substantive information is available to guarantee the desired accuracy and avoid disparate predictions when using proxy outcomes. We demonstrate our approach through practical examples. 

A taxonomy delineates where trust can break down in a probabilistic machine learning workflow that informs critical decisions. 

We combine weighting and Bayesian prediction in a unified approach to survey inference. The general principles of Bayesian analysis imply that models for survey outcomes should be conditional on all variables that affect the probability of inclusion. We incorporate all the variables that are used in the weighting adjustment under the framework of multilevel regression and poststratification, as a byproduct generating model-based weights after smoothing. We improve small area estimation by dealing with different complex issues caused by real-life applications to obtain robust inference at finer levels for subdomains of interest. We investigate deep interactions and introduce structured prior distributions for smoothing and stability of estimates. The computation is done via Stan and is implemented in the open-source R package rstanarm and available for public use. We evaluate the design-based properties of the Bayesian procedure. Simulation studies illustrate how the model-based prediction and weighting inference can outperform classical weighting. We apply the method to the New York Longitudinal Study of Wellbeing. The new approach generates smoothed weights and increases efficiency for robust finite population inference, especially for subsets of the population.