ABSTRACT
Outliers may be detected in simple scatterplots of the data. However, more sophisticated approaches can be useful. Boxplots and the so-called Q-Q or normal probability plots for the residuals may be useful. Signs of nonnormality include skewed (lack of symmetry), light-tailed, or heavy-tailed distributions. Although specific tests for normality exist, these tests may lack power in small data sets, and in large data sets statistically significant departures from normality may have a trivial impact. Therefore, the important issue to address is whether outliers had an important effect on results, not whether outliers or nonnormality existed.
