Hi Sander Gr , given the information you provide, it's very difficult to give a good answer. Particularly because the choice of method does not soley depend on the data you have available but also on your theoretical background and thus the research questions and/or proper hypotheses that you want to answer with them. If you - for whatever reason - however do not happen to have any hypotheses at all, the classical throw-everything-in-and-see-that-happens method for metric data would me a regression analysis and the classical method to compare between treatment conditions would be an ANOVA.

Good luck with your research

What is your N?

Thank you Annika Hamachers , my main hypothesis is to examine whether an increase in social activity, measured by two questionnaires, leads to a decrease in symptoms at a later timepoint more often than the other way around (first decreasing symptoms and then increase in social behavior). For this I would be able to use the three timepoints. I would furthermore like to examine this effect for patients with different medical conditions, measured by another questionnaire, as well as to add some demographics, using the 'see what happens method' ;).

I came up with a rather simple idea: compare the complaints score, as well as the social behavior score means at each of the three time points for the subgroups in order to see whether a change occurs in behavior before a decline of symptoms happens. More in detail:

Comparing the percentage of participants who experienced a change in social behavior before a change in symptoms occurred, between the different groups of patients as well as taking demographical information, such as age and gender into account. Comparing the scores from each of the three questionnaires to itself over the course of the three timepoints (T1 to T2, T2 to T3 and T1 to T3), was expected to show an overall average increase in social behavior, prior to an overall average decrease in symptoms. Due to the large effect expected from social behavior change, it was expected that social behavior increases relatively more in T1 to T2, while symptoms decrease at a stronger rate between T2 and T3.

which seems to me might perhaps be done with friedmans test and pairwise comparisons?

David L Morgan The total N in the database is about 2.000, but it is not yet clear what N will be per subgroup (after exclusion).

I will comment on mode of analysis: you have repeated measures data and that needs to be taken account of in the analysis ; for a quick walk through see http://www.bristol.ac.uk/media-library/sites/cmm/migrated/documents/repeated-measures.pdf

and for lots more detail with software, see

https://www.pilesofvariance.com/

The only true technical restraint on whether you have too many variables in a regression is the number of degrees of freedom. Since you have such a larger sample, that is not a problem.