Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: sampsi
Phil Schumm <email@example.com>
Re: st: sampsi
Fri, 19 Aug 2011 02:54:42 -0500
On Aug 18, 2011, at 3:18 PM, Eurich, Dean wrote:
> Ok trying to do a sampsi calculation with 2 groups with 1 baseline and 1 follow-up measure. It is basically a before after analysis with a control group. We are interested in the change score from baseline to the 8 week follow-up in predicting death for example
> We have pilot data in 16 people: group 1 are healthy people; Group 2 are people with Luo Gerhig's disease (values are MRI scans)
> Baseline MEAN Values are as follows:
> Group 1 =3D0.61 (SD =3D 0.03)
> Group 2 =3D 0.55 (SD =3D 0.03)
> After 8 weeks we found the following MEANS:
> Group 1 =3D 0.61 (SD 0.03)
> Group 2 =3D 0.539 (SD 0.03)
> So we have a mean change of 0.011 with a SD of change =3D 0.019
> For my sample calculation I am totally confused on what to use for my m1 and m2 and SD.
For testing the null hypothesis that the mean change from baseline to follow-up is the same in both groups, you want
sampsi `=0.61-0.61' `=0.55-0.539', sd1(0.03) pre(1) post(1) r01(0.8)
which gives 63 subjects per group for 90 percent power at the two-sided 0.05 level. This, of course, is equivalent to
sampsi `=0.61-0.61' `=0.55-0.539', sd1(0.019)
(i.e., a simple t-test on the change scores).
Note that while it's great that you have the benefit of preliminary data, your sample size in this case is very sensitive to your estimated correlation between the baseline and follow-up measurements. For example, dropping it from 0.8 to 0.7 increases your sample size from 63 per group to 94 per group. A correlation of 0.8 is pretty high (though not impossible), and so I'd probably want to use a somewhat lower correlation for my power calculation, since there is undoubtably some error around the estimate.
I'm a bit concerned about your statement that "we are interested in the change score from baseline to the 8 week follow-up in predicting death," because this sounds like a very different hypothesis, which would require a different sample size calculation.
* For searches and help try: