Disturbance generation for targeted Monte Carlo with cluster fixed effects

Determine a principled method for generating disturbances in targeted Monte Carlo experiments for linear probability or linear regression models with cluster fixed effects, specifying realistic within-cluster correlation structures that are not absorbed by fixed effects and ensuring meaningful evaluation of cluster-robust inference methods.

Background

Targeted Monte Carlo experiments can quantify the reliability of inference for a specific dataset and model, but require a realistic disturbance-generating process. In models with cluster fixed effects, standard random-effects specifications can eliminate within-cluster correlation, undermining the need to assess clustered inference. A clear approach to generating disturbances in this setting is currently lacking.