HI,
I have 3 groups of randomly assigned subjects and measuring dependent variable at 2 time points, before intervention (pre-test) and after intervention (post-test). I was thinking about doing ANCOVA with pretest value as co-variate. Or should I use rANOVA instead?
Should I compare pretest values with ANOVA to see if there are significant differences first?
My dependent variables are on continuous scale (mm, msec...)
And I was thinking in presenting change from baseline values for descriptive statistics. Percents or absolute values?
The test you'll use depends on what hypothesis you are trying to test. If the purpose of the experiment is to measure the change brought about by the intervention, and to see if that change differs between groups, then you would want to use a mixed ANOVA with time as a within-subjects (repeated measures) variable and group as a between-subjects variable, so that you can explicitly test the interaction.
It's nice to be able to show that there are no significant between-groups differences in the pretest, but if you do a mixed ANOVA then this becomes largely irrelevant: even if there are differences, the fact that you're testing the interaction allows you to try to identify cases where different groups showed different intervention effects. (That does not, however, explain why the effects differed. For example, if A>B at pretest and then A==B at post-test, that shows that B had a bigger treatment effect, but it doesn't tell you whether it's because of something you're interested in or because e.g. A was already at ceiling.)
As for your last question, usually when I see a design like this, what's presented (if anything) is the absolute changes, not the proportional changes. But you should also check related studies in your own area of research to see what is usually done.
As Stephen points, interaction is what you might probably looking for in your situation. And interaction can only be obtained in a repeated design.
I believe you can use ANCOVA with pretest value as covariate because using ANCOVA in your case will enable you to remove the initial difference as measured by the pre-test. This will enhance your efforts to equate subjects at the beginning of the experiment even though you have already used random assignment.
I was also thinking about ANCOVA. I want to see if inhalation of some aroma affects cognitive performance in healthy subjects. Subjects will do test before and just after intervention. I will also compare pretest value with one way ANOVA to see if the groups are similar.
You don't have to use ANOVA for the pretest. The idea behind random assignment is to satisfy the internal validity of the experimental design and equate subjects at the beginning of the experiment. Even though we use random assignment we want to be sure that subjects in our true experimental design have equal initial standing on all exterouse variables therefore we use per test and remove the effect of pretest from the post test by using ANCOVA
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA