Appropriate statistical approach? I have two groups of participants, with injury 1 and injury 2. (categorical).
ALL participants are divided in two groups, active or passive. (categorical).
All participants have filled in three questionnaires, two about depression and one about anxiety, these are all continuous/numerical.
I have two separate research questions, but with a catch.
The first two research questions are:
1)Does the active condition result in a lower depression score. (two t-tests, one for depression1, one for depression2).
2)Does the active condition result in a lower anxiety score. (t-test for anxiety)
However, I also want to know if there is a significant difference between injury1 and injury2 in the effects of active on depression and anxiety.
So: "3)is the effect of activity on depression and anxiety significantly different for injury2 and injury2? "
And that is where I get stuck. My professor keeps pointing me towards logistic regression, but I just can't figure out what tests to run/combine to come to my answer.
Any help would be greatly appreciated. I use SPSS
If you have a suggestion on how to combine the two t-tests of question 1 into one test, that would be great too.
 A: Assuming the questionnaires result in a continuous score for anxiety and depression, then separate linear regression models for each score with a main effects for the active vs. passive condition for the injury type could be a simple choice for the first two questions, but only if you randomly selected whether the active or passive condition was applied to each patient (assuming this is some kind of treatment strategy). Perhaps there'd be some refinements like adding a pre-treatment score to the model (if it is available) or jointly modeling both variables, but that basic strategy could be reasonable enough in a randomized study.
If the assignment was not randomized and your question is about whether one condition instead of the other one causes a higher/lower score, you have the extra complication that people in one condition may have been quite different and would have had very different scores even without the conditions. Then, you get into the difficult territory of causal inference from observational studies/data. One strategy is to use data prior to them being in either condition to estimate the probability of getting into the conditions (propensity score), and to match by that and compare the matched pairs, adjust the analysis the linear regression I described above by this or to stratify that analysis by the propensity score.
For the third question adding a injury type by condition interaction term to the model could be a sensible approach (all the considerations above also apply here).
