Browse Teach Curate

Surrogate-Augmented Optimal Subsampling for Generalized Linear Models

A subsampling strategy for fitting a generalized linear model when the gold-standard response is costly to observe but a cheap, error-prone surrogate is available for every record. Sampling probabilities are derived from the A-optimality criterion and combined with an augmented estimator that projects the response-based estimator onto the surrogate-based one, yielding a consistent, asymptotically normal estimator with lower variance than surrogate-free optimal subsampling.

Data Text

Optimal Sampling for Generalized Linear Model under Measurement Constraint with Surrogate Variables

This paper introduces OSUMCS, an optimal subsampling strategy for fitting generalized linear models when the gold-standard response is expensive to observe but a cheap, error-prone surrogate variable…