the data that needs analysis comprises k-independent groups over time
subjected to a standard treatment(lifestyle intervention) with each
group being measured before and after. Measurements are ratio scales -
blood pressures, weights and various blood tests. What statistical
tests would you suggest? Subjects cannot be regarded as having been
assigned randomly from a normal population.

---

Request for Question Clarification by mathtalk-ga on 16 Dec 2002 18:45 PST

Hi, vng-ga:

What is the hypothesis that the statistical tests are trying to
validate or invalidate?

thanks, mathtalk-ga

You have many options for analysis, each with strengths and
limitations.

1. You could do X separate repeated measures ANOVAs with the k groups
as the IV and the pretest and posttest measure (e.g., blood
pressure--pre and post) as the DV.  Do one ANOVA for each outcome
variable. Evaluate the interaction between groups and
pretest/posttest.

2. You could do X separate ANCOVAs with the k groups as the IV, the
pretest as the covariate, and the  posttest measure as the DV.  Do one
ANCOVA for each outcome variable (e.g., blood pressure pre and post).
Evaluate the between-group differences.

3. You could do one MANOVA (groups as a between factor ((with k
levels)) and the pretest and posttest as a within factor). Again, the
DVs would be the set of all outcome measures taken together. Evaluate
the groups by tests (pre/post) interaction. Or do a MANCOVA as above
but with the set of pretests as the covariate and the set of posttests
as the DV.

If your outcome measures are relatively uncorrelated, I would suggest
number 1. If they are moderately or highly correlated (r>.31), I would
suggest a principal components analysis, computing component  scores,
and using the component scores as DVs in number 1, instead of the
original outcome measures. If that's too unfamiliar, do the MANCOVA.

Good luck.