Using %MMAOV for Mean Separation
%MMAOV automatically runs a mean separation test when treatment levels are in the model. By default, Fisher's Least Significant Difference test is run at the 5% significance level. The purpose of this page is thus to present various options for controlling the mean separation test.
But first, scientists sometimes need a mean separation value, and means that differ by at least this value are statistically different (thus "least significant difference" value). This can be useful when comparing many means, as the letter groupings may use so many letters that interpretation is impossible. Such values are printed in the SAS Log Window, and an example is show here.
PDMIX800 08.08.2003 processing
LSD values for Irrigation are 16.9962 (avg) 16.9962 (min) 16.9962 (max).
LSD values for Spacing are 9.96082 (avg) 9.40174 (min) 10.5216 (max).
LSD values for Irrigation*Spacing are 16.9896 (avg) 13.2506 (min) 20.4301 (max).
For each term in the model, an average, minimum and maximum value is given. When the data or model are unbalanced, comparison values will differ among the pairs of means. As long as the minimum and maximum are reasonably close, the one average value can be used. If not, then the convenience provided by a single value may not be possible.
For complex problems, the 26 letters of the alphabet may not be enough to represent all groups. In this case, a second, third, etc. set of letters is used, indicated by (2) and so on. If you have two means, with letters of "A" and "(2)A", then these means differ, because the A's are in different sets of 26, and are not the same letter.