When people read, hear, or prepare research summaries,
they sometimes have misconceptions about what is or isn't "sound
practice" regarding the collection, analysis, and interpretation
of data. Here are some of these common (and dangerous) misconceptions
associated with the content of Chapter 12.
The Fisher, Duncan, Newman-Keuls, Tukey, and Scheffé
test procedures produce valid result only if they are used in post hoc
If the omnibus F from a one-way ANOVA turns out to
be significant, at least one of the pairwise comparisons in the post
hoc investigation will definitely turn out significant.
If pairwise comparisons (either planned or post hoc)
reveal that Group A is not significantly different from Group B and
that Group B is not significantly different from Group C, then it will
also be the case that Group A is not significantly different from Group
Dunnett's test should be used only if one of the
groups in the study is either a true control group or a placebo group.
Pairwise comparisons can legitimately be conducted
only if the ANOVA's omnibus F has first turned out to be significant.
In deciding which specific test procedure to use
for making pairwise comparisons, the typical researcher weighs the risks
associated with procedures that are overly liberal vs. the risks associated
with procedures that are overly conservative.
Scheffé's procedure is good for making pairwise
comparisons in a post hoc investigation.
Planned and post hoc comparisons cannot logically
be used together in the same study.