Covariate-Assisted Estimation of Grouped Panel Data Models

• Authors: Oriol González-Casasús
• BSE Working Paper: 1612 | November 2026
• Keywords: grouped fixed-effects, misclassification, auxiliary covariates, k-means
• JEL codes: C13, C23, C38

Grouped panel data models are usually estimated by classifying units on the residuals of the outcome equation, which ignores variables outside the model informative about the groups. This paper proposes the covariate-assisted estimator, which classifies units by a weighted two-view k-means rule that weights each view by its estimated precision. Auxiliary covariates are redundant when classification is perfect but valuable when it is not. At the precision weight, the squared separation-to-noise ratios of the two views add up and the endogenous bias offset of residual-based classification is preserved, so that misclassification and bias are lower than under either view alone. When the auxiliary covariate is an included regressor, the first-order bias vanishes despite misclassification. In a county panel of minimum wages and restaurant employment, manufacturing employment disciplines a residual-based partition that isolates outlying counties. In the income–democracy panel, the savings rate stabilizes the estimates across the number of groups.