Mousa, I don't know the details of your study so I don't know why you measured twice in post -test. However I would still say yes.  Suppose your data looked like this:                pre         post    covA

                         1            2,3         covA1

                           2           3,4          covA2     etc

Then I would arrange the data this way:

                           1                2           covA1

                             1               3           covA1

                              2                3           covA2

                             2                  4            covA2      etc

Hope this helps.  Best, David

No, ANOVA is not appropriate, because, firstly, it does not account for correlated data (within-subject correlation due to repeated measurements) and, secondly, it does not consider missing data (missings are deleted listwise), which reduces statistcial power and which may produce biased results. I recommend generalized estimating equations (GEE), which deals adequately with correlated data and missing values (see for instance Ballinger, 2004; Organizational Research Methods, 7:127-150; Gibbons et al, 2010; Annual Review of Clinical Psychology, 6:79-107).

Michael, First he said ANCOVA so the linear Model would be Post=a +bPre  + c(covA) +.... +eps. hence the correlation is accounted for. Second this is an experiment and there is no indication of missing data. However, even if there was, standard imputation methods would work here. If there were post1  Follow1 Follow2 etc I would recommend GEE or something similar.  See Dr Marie Davidian's course notes at the link below. My recommendation stands. David

http://www.stat.ncsu.edu/people/davidian/courses/st732/

David, if the outcome of interest is not normally distributed, then you cannot use the general linear model. In contrast to that, you can fit GEE to binary, ordinal or right-skewed continuous data, poisson distributions or wald distributions, etc. Moreover, according to most clinical trial guidelines the general linear model is not adequate for the analysis of longitudinal data and most journals will refuse to publish work based on ANOVA/ANCOVA. See Gibbons et al (2010; Annual Review of Clinical Psychology, 6:79-107) for a state-of-the-art review on the analysis of longitudinal data. Best, Michael

Micheal, Marie Davidian , who I cited, is a biostatistician, holds an endowed chair, and is a past president of the American statistical Association., and has done work on AIDS trials which are extremely longitudinal. Further she has written one of the top three books on longitudinal studies.(see the link). If you are interested in longitudinal studies I would check out her work.  My recommendation is one that she gave several years ago at a symposium on longitudinal studies at the Joint Statistical meetings. BTW her work has gone beyond GEEs. Check it out. Best, David

https://www.amazon.com/Longitudinal-Analysis-Chapman-Handbooks-Statistical/dp/1584886587/ref=sr_1_1?ie=UTF8&qid=1485770492&sr=8-1&keywords=Davidian+longitudinal

ANOVA TEST IS PREFERABLE 

Dear All

Thank you very much for your kind reply. Yes, my question was about a clinical trial where there were for example posttest1 Follow1 Follow2 ... (to see and compare longitudinal effect of for example a cognitive behavioral intervention between control and intervention groups). 

Best, Mousa