Evaluating Panel Data Estimators Under Unbalanced Conditions Across Sample Sizes

Abstract

This paper investigates the performance of panel data under the unbalanced panel dataset. The degree of missingness is varied in increments of 5% (5%, 10%, 15%, and 20%) and across different sample sizes (N = 25, 75, 100, 150, 200, 250, and 300). Monte-Carlo simulations of panel data for different N sample sizes and different degrees of missingness investigate the behaviors of the estimators through Mean Squared Error and Mean Absolute Error criteria. The rule of thumb is choosing the estimator with the lowest value of MSE and MAE as the best estimator that performs better than others at different levels of degree and different sample sizes. For the balanced panel data, the Between estimator, ranked 1st, the Within estimator, Random estimator, and Pooling estimators have no specific rank as they are not consistent in ranking, while the First Difference estimator, which has the highest MSE, MAE, and RMSE, ranked last for the panel data set. Similar results patterns were observed for the Unbalanced panel data set, which shares the same ranking as the balanced panel dataset. The result underscores that the estimator consistently outperforms its counterparts in managing unbalanced panel data under varying degrees of missingness. This estimator was therefore considered fit for the data used.