C. Check normality (continued).
Three more examples of normality diagnostics are shown here to show the range of results and the conclusions they lead to.
Example 1. This illustrates excellent normality, with a W well above 0.90 (ignore the statistically significant p-value), and data following the reference normal distributions in the graphs.
Example 2. This example shows poor normality, due to the distribution being too flat, not bell-shaped, and the probability plot shows that the data values occur in groups, raising concerns about the continuous nature of the measurements. This could potentially be analyzed as is, since the W is close to 0.90, but the data should be examined carefully for errors and scientific meaning. If the data are categorical, then alternative statistical methods should be considered.
( Step D. Check for model validity )
|H I N T S :|
|Notice the asterisks in the second boxplot. That usually indicates outliers, but in this case, it is due to the non-normality. Thus, you might want to check normality before discarding outliers.|