Post Hoc Tests - Agricultural Statistical Data Analysis Using Stata - page 151

test 2*5.trt - (6.trt+7.trt) = 0
( 1) 2*5.trt - 6.trt - 7.trt = 0
F( 1, 35) = 2.12
Prob > F = 0.1542
test 6.trt - 7.trt = 0
( 1) 6.trt - 7.trt = 0
F( 1, 35) = 0.01
Prob > F = 0.9051
Built-in Multiple Range Tests
In many cases, and some statisticians think in most cases, the specific
treatment comparisons should be planned in advance. Frequently, the
experiment and treatments will indicate the planned comparisons you
should look at. For example, an entomologist may be interested in how
the current standard insecticide compares with new materials. These
new insecticides, in addition, may have different modes of action and
the researcher may wish to compare these different modes of action.
This kind of information before the experiment is conducted deter-
mines what the planned comparisons will be.
There are, however, legitimate cases, I believe, where planned
comparisons are not possible. Variety trials are a good example. The
comparisons of interest in such a case can encompass all possible com-
parisons. More than likely, however, the comparisons of interest will
depend on the individual viewing the information. I do variety trials
that are distributed widely to growers, seed companies, and other
researchers. Each has its own comparisons of interest. Growers may
be interested in comparing their current variety to improved or better-
performing varieties. Seed companies may be interested in compar-
ing their varieties to their competitors and researchers could have a
wide range of interests in the trial as it relates to their work. As the
number of comparisons increases, the chance of committing a Type
I error increases. For example, with 10 varieties, there are 45 pos-
sible pairwise comparisons. The comparisonwise Type I error can be
calculated, for example, at the 5% level as 45 × 0.05 = 2.25, which is