3
$\begingroup$

I have 4 treatment groups:

1 control (placebo)
1 with X treatment
1 with Y treatment 
1 combination-treatment-group XY

and I have monitored the individuals (n=10) in each group by measurement of their tumor volume at baseline, day 1, 3, 7, 10 and 14 to see what treatment is best and how early effect can be seen. What statistics should I use?

The repetitive measurement at day 1, 3, 7, 10 and 14 is some sort of paired comparisons I guess and the comparisons between groups are unpaired. Since there are more than two groups I guess it is ANOVA I should use but will I then have to Bonferroni correct for the fact that I have "looked for the same difference" 5 times?--even though I would expect the difference to show a trend--I would expect the difference between an effective drug and placebo to become greater for each treatment day. I would also like to test if the combination treatment is better than single X and single Y on its own. Since I only have 10 individuals in each group I can not "effort" much Bonferroni correction.

I have done a lot of unpaired and paired t-tests but that might not be correct without any post hoc testing. I am a bit confused about the fact that I lose power because I test more things at once or compare the differences at more than one day. Will it be fair to leave all p-values uncorrected (no Bonferroni) and state that it is a hypothesis generating study testing when a predefined difference can be seen?

$\endgroup$
1
1
$\begingroup$

This is a relatively small sample. Maybe you could try hierarchical linear modelling, or repeated-measures ANOVA. I would probably try repeated-measures ANOVA first. You would have 5 time points and treatment group as a factor. See if there is a group*time interaction.

Multiple-comparisons would not need to be controlled for in a repeated-measures ANOVA, so do not worry about p values and dividing them or anything.

$\endgroup$
0
$\begingroup$

I would recommend running a repeated-measures design, such as a repeated-measures ANOVA like Behacad suggested. You will not need to correct for multiple comparisons at that point (ANOVA is meant to deal with this exact problem).

If you do run a rANOVA and you find significant differences, you will need to run contrasts/post hoc tests that account for multiple comparisons. There are other approaches than the Bonferroni, though, that better retain power. I recommend you read around to get a sense for which would be most appropriate, but here is a basic description to get you started: http://pages.uoregon.edu/stevensj/posthoc.pdf

For a textbook you can try Andy Field's Discovering Statistics with SAS (or R or SPSS, whichever you use).

$\endgroup$

Your Answer

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

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