ESRA 2027 sessions by theme

Back to Overview of Sessions

Session Details

Survey weights often evolve throughout the survey processing lifecycle as adjustments are made to improve the quality and efficiency of final estimates. Base weights reflect the sample design; nonresponse-adjusted weights improve respondent representativeness; benchmarked weights calibrate for population-level consistency; and weight smoothing can reduce weight variability and increase statistical efficiency.

These weight adjustments may rely on model-based procedures, often using generalized linear models to estimate quantities needed for weighting. Examples include linear regression for weight smoothing, logistic regression to estimate response propensities for the double expansion estimator, and linear regression to obtain the generalized regression estimator for calibration. Machine learning techniques can replace or augment such model-based components, potentially capturing more complex relationships among design, response, auxiliary and outcome variables.

This session invites applications of machine learning to conventional survey weight adjustments, with attention to their potential benefits, practical implementation, and implications for the resulting estimates.