Testing zero variance components is a common practice under random-intercept models. Various test exist to check the need for random effects in such models. Although many of those tests have correct Type-I error rate even when the error components are not normally distributed, an empirical assessment of the performance of these tests when the is contaminated in the form of possessing heavy tails, heavy skewness, or contains outliers does not exist. This article investigates the performance Of four recently proposed variance components tests under such violations using extensive simulation studies. Results indicate that the simulation-based test based on the likelihood ratio test statistic is much preferred to the other tests unless the response space suffers from the presence of outlier.  Under the latter case, none of the competing tests revealed satisfactory performance.