e****t
发帖数: 766 | 1 suppose, each patient will be measured at 1,2,3,4 time, but there is missing
value at different timepoint.
we fit model as
proc mixed ;
class sub time trt ;
model sub = time trt time * trt;
repeated time / sub = sub type = un ;
lsmeans trt * time / pdiff ;
run;
when calculating effect size,
we use ls Means diff / (SE / sqrt (1/n1 + 1/n2))
here n1 and n2 should be patient number in different group.
or should be patient number in different group by visit (due to missing,
observations at each visit may be different from n1 and n2. )
I am thinking the patient number in different group make more sense.
so at each visit, observation number could be differnt. when calculating the
effect size at differnt visit, we rely on the patient numbers (constant ) | e****t
发帖数: 766 | 2 but i dont have any strong evidence on it.
can any one give any thoughts ? | s*r
发帖数: 2757 | 3 the denominator for effect size is sd instead of se | e****t
发帖数: 766 | 4 yes, i agree. just when you calculate back from se to sd.
how to decide n1 and n2 ?
if using SE / (sqrpt (1/n1 + 1/n2))
thanks | s*r
发帖数: 2757 | 5 i only see people use raw data to get sd
【在 e****t 的大作中提到】
: yes, i agree. just when you calculate back from se to sd.
: how to decide n1 and n2 ?
: if using SE / (sqrpt (1/n1 + 1/n2))
: thanks
| e****t
发帖数: 766 | 6 :( ls mean diff and SE is from statement:
lsmeans trt * time / pdiff;
for example, for time 4, trt 1 - trt 2, use this statement, we obtain
lsmeans diff and se.... | s*r
发帖数: 2757 | 7 i suddenly remember that i used mse to estimate the sd for the denominator |
|
