Nested random effects estimation in unbalanced panel data

Panel data in many econometric applications exhibit a nested (hierarchical) structure. For example, data on firms may be grouped by industry, or data on air pollution may be grouped by observation station within a city, city within a country, and by country. In these cases, one can control for unobserved group and sub-group effects using a nested-error component model. A double-nested unbalanced panel is examined and a corresponding maximum likelihood estimator is derived. A generalization to even higher-order nesting is feasible. A practical example and a Monte-Carlo simulation compare the new estimator against the non-nested ML estimator. The style of presentation is intended to aid applied econometricians in implementing the new ML estimator.