4
$\begingroup$

I wish to examine the effect of an intervention on a set of ~60 mutually correlated features (dependent variables), measured at baseline and post-intervention in ~70 subjects. At both baseline and follow-up each feature was measured under two different conditions. I want to determine if which features were significantly effected by the intervention.

I don't believe a repeated measures ANOVA is appropriate due to the large number of mutually correlated features. I also don't believe multiple t-tests are acceptable due to the risk of Type I error (particularly at a 0.05 significance level).

Could using the change scores (i.e. change in each feature between baseline and follow-up) in a Lasso regression be a valid method?

What is the most appropriate statistical approach? Grateful for any help.

Improve this question
$\endgroup$
4
  • $\begingroup$ Do you mean to say "predictors" here? These sound like dependent variables. Your predictor is time (pre and post intervention), right? $\endgroup$
    – sammosummo
    Commented Aug 23, 2016 at 10:15
  • $\begingroup$ @user1637894 I mean predictors in the sense of features, I have edited to clarify. $\endgroup$
    – BGreene
    Commented Aug 23, 2016 at 10:18
  • $\begingroup$ I think either "dependent variables" or "measures" are more appropriate than "features" $\endgroup$
    – sammosummo
    Commented Aug 23, 2016 at 10:21
  • $\begingroup$ I think the correct technique here is the MANOVA. $\endgroup$
    – sammosummo
    Commented Aug 23, 2016 at 10:22

1 Answer 1

Reset to default
2
+50
$\begingroup$

You might consider a repeated-measures MANOVA. The MANOVA is analogous to the ANOVA, but with more than one dependent variable. Just like the ANOVA, you can have several independent variables (predictors). The procedure is easy to implement in statistical software packages, such as SPSS. One word of caution mentioned in the linked article is that SPSS and similar packages do not conduct the proper post-hoc tests by default.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Thank you for this answer. The method in Matlab to do this is manova with a repeated measures design $\endgroup$
    – BGreene
    Commented Aug 30, 2016 at 13:28

Your Answer

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

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