1
$\begingroup$

We have a randomised controlled trial, with two groups of the same size (drug and placebo). A variable was measured before and after substance administration. To assess a treatment effect, two approaches were used:

  1. a repeated measure ANOVA, with time as within-subjects factor and group as between-subjects factor.
  2. a ANCOVA, with pre-treatment variable values as a covariate, pos-treatment as dependent variable and group as fixed-effect.

These two approaches led to different results:

  1. RM-ANOVA shows a main effect of time (p<0.0001), an interaction effect (p=0.013) and no group effect.
  2. ANCOVA shows no effect for group (p=0.094), neither for the baseline measure (p=0.092).

Can I conclude that there is a treatment effect based on the interaction effect? Or, since the ANCOVA shows no effect, this interaction effect could be only due a difference in baseline scores? How to report this result?

$\endgroup$
  • $\begingroup$ Did you include an interaction in your ANCOVA with your baseline? What variable was measured before and after substance administration? The dependent variable? If so, why not simply take the difference between the before and after $y$-values and regress this difference on study group (treatment or placebo) to simply everything? $\endgroup$ – StatsStudent Jul 25 '16 at 4:40
  • $\begingroup$ RM for the BL + post-BL values makes very strong assumptions on the joint normality of the values that is typically violated, if you needed values below or above some threshold (for the response variable or one correlated worth it) to get into the trial. $\endgroup$ – Björn Jul 25 '16 at 8:16
  • $\begingroup$ @StatsStudent: I am measuring the scores of a depression scale (HAM-D), which is my dependent variable. To the ANCOVA, I used the following command on SPSS: UNIANOVA HAM_D7 BY Treatment WITH HAM_BL /METHOD=SSTYPE(3) /INTERCEPT=INCLUDE /EMMEANS=TABLES(Treatment) WITH(HAM_BL=MEAN) COMPARE ADJ(SIDAK) /PRINT=ETASQ HOMOGENEITY DESCRIPTIVE /CRITERIA=ALPHA(.05) /DESIGN=HAM_BL Treatment. I don't understand how to perform this analysis doing a regression. Can you explain? Thanks $\endgroup$ – Fer Jul 25 '16 at 15:06
  • $\begingroup$ @Björn: I didn't understand your comment. The RM ANOVA is not a valid way to test between treatments response? $\endgroup$ – Fer Jul 25 '16 at 15:12
  • $\begingroup$ If you include the baseline as a measurement rather than a covariate, and eg a minimum HAM-D was required for trial inclusion, then I would be suspicious of a RM model. $\endgroup$ – Björn Jul 25 '16 at 22:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.