Overview

The field studies the consequences of using models that omit variables, impose incorrect structure, or fail to capture changes across environments. Approaches include robust estimation such as median-of-means methods, posterior predictive calibration, residual and identifiability diagnostics, and discrepancy correction; applications span forecasting, physics-informed learning, simulation-based inference, and reinforcement learning. A central challenge is distinguishing genuine misspecification from noise, weak data, or poor optimization, then correcting it without erasing useful uncertainty or real effects. Because no diagnostic can establish that a model is fully correct, reliable practice often combines complementary checks with iterative model revision.

Latest papers

CardsList

35 papers4 papers in the last four weeks, down 56% on the four weeks before. 0.0% of all new papers.

  1. Omitted Variable Bias in Language Models Under Distribution Shift

    Feb 18, 2026Victoria Lin, Louis-Philippe Morency, Eli Ben-MichaelOOD GeneralizationGeneralization Bounds

  2. Robust Amortized Simulation-Based Inference via Learned Error Models

    Sep 6, 2025Matthew O'Callaghan, Kaisey S. Mandel, Gerry GilmoreSimulation-Based InferenceNeural Posterior Estimation

  3. Dissecting the Impact of Model Misspecification in Data-driven Optimization

    Mar 1, 2025Adam N. Elmachtoub, Henry Lam, Haixiang Lan +1Model MisspecificationDecision-Focused Learning

  4. Predictive variational inference: Learn the predictively optimal posterior distribution

    Oct 18, 2024Jinlin Lai, Antonio Linero, Yuling YaoModel MisspecificationVariational Inference

  5. Robust Linear Predictions: Analyses of Uniform Concentration, Fast Rates and Model Misspecification

    Jan 6, 2022Saptarshi Chakraborty, Debolina Paul, Swagatam DasRobust RegressionModel Misspecification