ANCOVA or oneway-ANOVA on gain scores

#1
Hi everybody.

I'm curious to know which is the most correct/preferred method to analyse my data.

I've got a randomised clinical trial with three groups(placebo(n=24), low dose(n=21) high dose(n=21)). There is pre and post measures on a 40+ continuos variables.
I've performed ANOVA at baseline to identify differences at baselines and found a significant baseline difference on 3 variables of which 2 are important end-points.

My initial thought was to analyse all variables by performing ANOVA on gain scores. But now things are a bit more complicated with a couple af baseline differences on two important baseline end points.

Regarding outliers, normality, variance: There's normality, equal variance (all other assumptions are satisfied too) but there are a number significant outliers in my dataset which I've have no grounds to exclude due to intercurrent physiological problems or due to measuring problems (in other words they stay included).

1) Would you prefer to perform ANOVA on gain scores on the variables with no baseline difference and then do ANCOVA on the varibales that are significantly different at baseline.

or

2) Would you prefer to analyse the entire dataset using the ANCOVA (is there anything gained by doing so if groups aren't different at baseline)

3) How would you present ANCOVA findings to clinicians with no understanding of degrees of freedom etc? Table of estimated means±SEM, p-value and covariate value or something else???

I have a very rudimentary understanding of the models at hand and that they answer slightly different questions:

ANOVA: Is there a difference of effect of the given treatment between groups (regardless of baseline)

ANCOVA: Given that all groups are equal at baseline can we expect the same change in the groups with the given treatment.

Any insight would be much appreciated.

Kind regards

Thomas
 

Karabiner

TS Contributor
#2
I'd suppose that the best method is repeated-measures
(or "mixed") ANOVA here (for ANCOVA, see "Lord's paradox").

With kind regards

K.