Not every credible interval is credible: Evaluating robustness in the presence of contamination in Bayesian data analysis