Examples from W. G Cochran’s paper, http://www.jstor.org/stable/3001535.

Table 1

There is an outlier plot in these data, most likely an entry error. The general test of normality (Shapiro-Wilks) and the test for skewness are larger than we might like (\(p<0.10)\), while the test for kurtosis indicates non-normality.

Note that since this column has the RATIO rating type, we will be warned about the outlier on entry. I’ve included duplicate columns, marked AL and AR, to compare the different options for dealing with non-normality.

Table 2

Cochran argues that the data have heterogeneous variances, but we miss this with a Levene’s of 0.054. Note from the wedge (pointing to the right) in the spread-location plot argues against any of our available transformations. I’ve only included an additional AR column.

Table 5

This was also used for the 2017 ASA meetings, in Beyond RCB. This illustrates spatial models.

Table 7

Not a lot to show in the ARM reports, but this illustates Tukey’s 1 d.f. test for non-additivity. We include this test in ST.