Detecting Outliers and Using Robust Methods in Linear Panel Data Model

The increasing use of panel data across various fields necessitates robust estimation methods that can resist the influence of outliers, which often lead to biased and ineffective estimates when using traditional methods like least squares. This research investigates two robust estimation techniques within fixed and random effects models for panel data, comparing their performance using the mean square error. Through a simulation experiment with varying sample sizes and contamination levels, the results for the fixed effects model indicate that the Weighted Likelihood Estimator consistently outperformed other methods across all sample sizes at a 10% contamination rate, while the S method excelled at a 20% contamination rate. For the random effects model, the former was most effective with a sample size of 200, while the latter proved superior at a sample size of 800, regardless of contamination levels.