| |
Statistical software
can be very helpful for exploring the relationship between sample
size and statistical power in the context of repeated measures
of analysis of variance. Such software typically accommodates
both univariate and multivariate repeated measures models, and
allows one to compute the powers of the tests for the main effects
and their interactions for a given sample size and level of significance.
If the univariate analysis is selected, one must enter the unadjusted
error variance and the assumed correlation between repeated measurements.
If the multivariate analysis is selected, one must input an estimate
of the covariance matrix.
Using the experimental
design described above, the investigator studies alternatives
to the null hypothesis of "no main and interaction effects"
in which infusion of either the painful or non-painful stimulus
causes an increase or decrease of the measurements relative to
baseline, and returns to baseline value with the passage of time.
The required number of cases in each group will ensure powers
of 80% for alternatives considered to be of clinical importance,
given, as an example, that the pain-specific main effect is expected
to be at least 20%.
|