Hello I have been struggling for many days with this problem. The situation and data set are following. The data set comes from an experiment that asked subjects (ID) to compare several CVs on different experimental conditions. Each CV had two factors, picture(no picture, modified picture, normal picture) and gender (level: male/female). In order to rate CVs, participants were asked to complete likert scale questionnaire with minimum score of 7 to 49. Higher score indicated they "liked" the CV more. Furthermore the participants were sampled in 4 different settings, so the setting as well had a Factor (level: A,B,C,D).

Meaning that setting was between-factor, while picture and gender within-factor.

data.frame looks like this:

      ID  Setting Male_Normal Male_Modified Male_No_Picture Female_Normal....and so on..
        1      A         7          11            16               49    
        2      B        10          16            30               30
        3      C        11          30            20               20
        4      D        30          11            2                10
        5      A        20          15            13               15
        6      B        10          11            11               10
    ..and    ..so       ..on        ..            ..               ..

Main hypothesis: H1:Scores differ based on Setting H2:Different scores for Gender H3:Different scores for Pictures

Question is how to analyze it correctly in SPSS or/and R and how to do post-hoc afterwards, should univariate test be used or discriminant analyses?


I was trying to run repeated measures MANOVA in SPSS and in I tried approach using the car package R.

In R I did not get past formatting data set so it makes sense to R, or in other words I was not able to create factors for Gender and Pictures properly...

In SPSS I did large rep.measures MANOVA, found significance for everything literally and then I am not sure how to break the effects down. For example if GenderPictureGroup is significant, what would you do?

I know the question is probably not an easy one, though any leads are highly appreciated.

Thank you very much!

  • $\begingroup$ I always prefer to analyze this sort of thing as a multievel model - it's much easier to think about. $\endgroup$ Commented Nov 19, 2015 at 1:45
  • $\begingroup$ Hello. Thank you for an answer. How would you define the model in this case? Does it also matter what is nested in what? E.g. Gender in Weight or Weight in Gender? I have no practical experience with multi level so I will have to go from basics. Thank you. $\endgroup$
    – gofraidh
    Commented Nov 19, 2015 at 14:06
  • $\begingroup$ I think that measures are nested within people, and that the different conditions are variables. Surprisingly (at least to me) there's a wikipedia page: en.wikipedia.org/wiki/Multilevel_modeling_for_repeated_measures $\endgroup$ Commented Nov 19, 2015 at 17:40
  • $\begingroup$ Also this answer: stats.stackexchange.com/questions/24314/… and Google will find you lots more. $\endgroup$ Commented Nov 19, 2015 at 17:41
  • 1
    $\begingroup$ This was tough decision but at the end of day I/we decided to do several analyses. First we run MANOVA, then we continued by analyzing significant interactions with repeated ANOVAs or ANOVA. Probably standard approach. What was interesting is that all interactions in MANOVA were significant. This proved to be quite difficult to handle in follow-up analyses. When I will have more time, I will try the Multilevel approach and see where it leads. Thank you for your suggestion Jeremy. $\endgroup$
    – gofraidh
    Commented Nov 30, 2015 at 16:07


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.