Dear All,
I have a question about testing for lack of difference between treatment
arms in a medical randomised controlled trial.
I have been summarising the result of some medical trials. One of the
"filters" has been to look for a statistically significant difference in
outcome between the two treatment arms.
However, someone has (quite correctly, I think) asked whether we would
ever consider a finding of no difference between two arms as being
important.
As I understand it, we normally set up the null hypothesis such that the
null hypothesis would be that there will be no difference in outcome
between the two arms. We then reject the null hypothesis (at a certain
level of certainty) if there is a difference in outcome between the two
groups beyond that explicable by chance.
However, given some prior knowledge, we might expect there to be some
difference, and the lack of difference would be highly unexpected (and
therefore of interest). To take a non-medical example, we would expect
the heights of a group of 10 year olds to be different to that of 5 year
olds: a lack of difference would be important.
Has anyone had to deal with something like this? Can we set up a null
hypothesis to be a certain difference between two groups (would that
simply be a subtraction of one from the other)?
I suspect that this can be done using Bayesian approaches, but this
isn't how the data in medical research papers is presented.
Thanks,
Matt
Dr. M. Williams
Clinical Research Fellow, CR-UK & UCL
|