0
$\begingroup$

I would like to test the following hypothesis:

High trait anxiety, depressive traits, and fear of movement will be higher, whilst optimism will be lower, for group 1 (patients) as compared group 2 (healthy controls).

My idea on how to test the hypothesis: Conduct a RM ANOVA with group type (patients/controls) as between factor, whilst questionnaire scores (5 different questionnaires) will act as within factor.

My question: would this be a feasible and correct way of statistical analysis? I'm a little uncertain at this point in time.

Thanks in advance. Kind regards, Jeroen

$\endgroup$
  • 1
    $\begingroup$ The way I interpret this, you have 5 different outcomes because those aren't the same questionnaire, in which case these aren't repeated measures. What is it you are measuring multiple times? $\endgroup$ – Frans Rodenburg Sep 14 '17 at 7:46
1
$\begingroup$

If your assumption is that there is dependence between the different questions, ie that they are associated with the same "construct" then it makes beautiful sense to use repeated measures modelling.

I would make sure in a first step to make sure to transform the variables to the same direction and size:
- you may have completely different variance and offset, so I would centre and scale the variables first.
- anxiety and contentment are obviously going in opposite direction. So after catering and scaling it would make sense to multiply opposing variables by -1.

After that you can perform a rm model either using individual as random factor or by marginal modelling (in sas this would correspond to the repeated statement. In r u can use e.g. the geepack library for marginal modelling.

There are also other ways to deal with and investigate commonality between variables, such as factor analysis and calculating cronbachs alpha for constructs.

Best of luck with your analysis! Carl

| cite | improve this answer | |
$\endgroup$
0
$\begingroup$

Using repeated measures could be useful if you wanted to compare the anxiety level to the depression level (and the other trait levels from the other questionnaires), assuming they were on the same scale. You might be interested in the degree to which patients were more depressed than anxious, as compared to controls (a trait*group interaction).

However, in your case you have specified 5 independent predictions, so you probably want to do 5 separate analyses (although you only specified 4 in the question). These would probably be t-tests or Wilcoxon-Mann-Whitney U tests.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ If I understand correctly, questions are not independent, but reflecting the same construct. They can therefore be seen as repeated measures of the construct. $\endgroup$ – CarlBrunius Sep 14 '17 at 23:00
  • $\begingroup$ @CarlBrunius: You mean like 5 questionnaires for anxiety, and 5 questionnaires for depression? Or do you mean that the questionnaire for anxiety and the questionnaire for depression are not independent? What construct would they both be reflecting? $\endgroup$ – Kayle Sawyer Sep 15 '17 at 0:07
  • $\begingroup$ Sorry. Should've said potential construct or something similar. Depends on if you look for constructs from data or from in-field expertise. From a data-driven perspective, anything can be tested as a construct whether or not it makes sense from in-field expertise. So if questions aren't independent, they can be tested as a construct either using repeated measures or e.g. alpha or omega tests. And for a lay-person (like me) questions likely aren't independent (it's also the hypothesis of the asker), which justifies trying different tests... $\endgroup$ – CarlBrunius Sep 20 '17 at 22:06

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.