Optimal trade-off between sample size, precision of supervision, and selection probabilities for the unbalanced fixed effects panel data model