All Pairwise Comparisons among Means: Recapitulation
and Recommendations (2 of 3)
There is no "correct" procedure to use; the various procedures
trade off power
for control of the
in different ways. There is a consensus that the
Duncan's and Fisher's LSD procedures result in too high an EER and should not
be used. The choice between the Newman-Keuls and Tukey HSD is a close call. You
have to decide how important it is to control the EER completely. If you want
to be sure that you have controlled the EER, then the Tukey HSD should be used.
Most statisticians now consider the Newman-Keuls unacceptably liberal because
of situations in which it allows an inflated EER.
However, it is up to the researcher to weigh the balance between power and
controlling the EER.
Be careful not to accept the null hypothesis for comparisons for
which the null hypothesis is not rejected.
Otherwise you may be left with contradictory conclusions. You may
find, for example, that M1
is significantly different from
but that M1
is not significantly different
and that M2
is not significantly
different from M3