0
$\begingroup$

I have a couple of general questions so bare with me.

I have three groups, patients, relatives and healthy controls (HC). I determined an effect between patients and HC, although that effect wasn't significant between HC and relatives. However, there's an obvious trend: the data for the relatives sit in between the two the way you expect them too. What kind of test I should run to see if there's a statistically significant trend, from patients to relatives to control? This test should come out negative if patients > relatives=control but positive if patients > relatives > control. I am guessing a one way ANOVA won't do cause it will test if any two means are different. Is regression with categorical variables the answer?

The second question is somehow related. I have several measurements from each subject of different but related things. I want to test if there are group differences between 2 groups in regards to those measurements. I am guessing a usual t-test won't do the job cause there's some dependence between the data collected (the emasurements that come from each subject). Should I then do repeated measures ANOVA? The issue is that those measurements from every subject are also describing different things.

I appreciate your help on this.

Best regards.

$\endgroup$
2
$\begingroup$

Regarding your first question: Regression with categorical variables is the same as ANOVA. Ordinal independent variables are tricky - most methods treat them as either continuous or categorical, but neither is exactly right. PROC TRANSREG in SAS offers optimal monotone scoring, which is a good method. I am not sure if this is available in other packages, but it probably is.

Regarding your second question: You need to account for dependence in the data. However, RM ANOVA makes some unrealistic assumptions. I recommend a multilevel model.

$\endgroup$
2
  • $\begingroup$ Thank you Peter. Any ideas how should I go about a multilevel model? $\endgroup$
    – G Ch
    Sep 30 '15 at 18:11
  • $\begingroup$ They are complicated models. You could either take a course or else hire a consultant. $\endgroup$
    – Peter Flom
    Sep 30 '15 at 20:00

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.