To change mean separation method (test): See a comparison of methods here
The default mean separation test is Fisher's Least Significant Difference. Other mean separation methods supported by Proc MIXED can be requested, using the ADJUST= option.
An example is %MMAOV(one,yvar,CLASS=treat, FIXED=treat, ADJUST=tukey);
Using the ADJUST= option, you can choose one of these methods:
BON, DUNNETT, SCHEFFE, SIDAK, SIMULATE, SMM and TUKEY.
To change the significance level:
Use the ALPHA= option to change from the default 5% significance level, for example:
%MMAOV(one,yvar,CLASS=treat, FIXED=treat, ADJUST=tukey, ALPHA=.01);
To sort the means:
By default, means are sorted by level name, with treatment A listed before treatment B. But if there are many treatments, sorting by mean value can make interpretation of the mean separation letters easier. Do this with the SORT= option, for example:
%MMAOV(one,yvar,CLASS=treat, FIXED=treat, ADJUST=tukey, SORT=yes);
To see individual comparison probabilities:
The conventional letter grouping only shows if means differ at a chosen significance level. You may want to know the P-value for comparing two means. Request this additional output using the MIXDIFFS option,
%MMAOV(one,yvar,CLASS=treat, FIXED=treat, MIXDIFFS=yes);
Beware that extensive output can be produced, since a line is printed for every pair of means.
To test if means are equal to 0:
Proc MIXED automatically tests if each mean is equal to zero, but this is generally of no practical use, so %MMAOV does not print the tests (default TEST0=no). If you would like to see these results, use TEST0=yes. For example:
%MMAOV(one,yvar,CLASS=treat, FIXED=treat, MIXDIFFS=yes, TEST0=yes);