Dear All,
I'm analyzing a longitudinal study on 400 patients where measures on a
continuous data were collected at 3 time points (T0: pre-treatment; T2:
last visit and T1 an intermediate time point);
comparison among two treatments is the aim of the study.
The best analytical approach, in this case, is the Repeated Measure ANOVA.
The analysis of the original data shown a non normal distribution, for the
variable of interest, at each time point.
After having transformed the data by a log_tranformation the shapes, at a
quick look, seems normal.
The test I used in SAS program (Proc Univariate - Test W Shapiro-Wilk
statistic for n<2000 ), instead, gives results nearly 0.01.
My question are:
- Is it a big mistake the implementation of a parametric model on such
data?
- Is the p-value calculated by the W-test the only criteria in determining
the application of a parametric or non parametric model (other possible
criteria: total N - Skewness)?
- Which non parametric model should be the candidate for the analysis of
treatment effect in repeated measures data ?
The replies will be promptly summarised.
Thanks a lot.
Luigi
Dr. Luigi Santoro
Head of Statistical Service Unit
Mediolanum farmaceutici S.p.A.
Via S. Giuseppe Cottolengo, 15 - 20143 Milano - Italy
Tel.: +39 0289132343
Fax: +39 0289132375
e-mail: [log in to unmask]
Web site Mediolanum farmaceutici S.p.A.: http://www.mediolanum-farma.com
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|