0

I have 300 patients with heart problems (2 groups/types of problem) that have completed the questionnaire SF-36 (8 scales), before and after the surgery. Some of them have been attending sessions with a therapist (psychologist) before the surgery. We want to see if the patients that attended sessions have better quality of life using repeated measures ANOVA. Also, we want to see if the type of problem is significant. The scales of the questionnaire SF-36 before and after the surgery, are not normally distributed (KS test, histograms, seems to be more skewed than normal, but not all).

Can i use repeated measures ANOVA (any recommendation for bootstrap)? If NO, is there an alternative model/method that i can use (for example, non parametric method) ?

Peter Flom
  • 94,055
  • 35
  • 143
  • 276
Didie
  • 11
  • 3
  • It is more important that the residuals are normally distributed than that the variables themselves are nornally distributed. What software are you using? – llewmills Dec 09 '18 at 11:13
  • See this discussion. https://stats.stackexchange.com/questions/6350/anova-assumption-normality-normal-distribution-of-residuals – llewmills Dec 09 '18 at 11:16
  • Thank you for your response. Neither the residuals are normally distributed in most of the cases. I am both using SPSS and R. – Didie Dec 09 '18 at 11:38

1 Answers1

0

RM ANOVA makes assumptions that are rarely justified (especially the assumption of sphericity). I would suggest a multilevel model. Both linear and nonlinear MLMs are available (since you mention that you use R, check the lme4 and nlme packages).

Peter Flom
  • 94,055
  • 35
  • 143
  • 276
  • Thank you a lot for your suggestions. Lets say that my variables are: Y1: the first measure of a scale Y2: the second measure of a scale FactorA: Type of Problem (2 levels) FactorB: Sessions (2 levels, YES or NO) MyData – Didie Dec 09 '18 at 11:59
  • I don't know; I am mostly a SAS user. You could ask on an R list. – Peter Flom Dec 09 '18 at 12:13