a********y 发帖数: 474 | 1 suppose it's a survey data with sampling weight
treatment: A (0,1) ,B (0,1)
sex: male; female;
age groups: young, middle, old
all patients may use A, or B, or both A and B at the same time. So the
independent sample assumption does not hold.
Question: is treatment A significantly more adopted than treatment B (
overall, stratified by sex, age, etc)?
请问可以 dependent t-test 吗?
谢谢!! | T*******I 发帖数: 5138 | 2 The "independence" of a sample should mean that each individual (here is a
patient) in a sample is independent to all others. So, your sample can be
treated as three independent groups:
Use A only
Use B only
Use A and B
You can take the simple ANOVA with interaction A*B. I believe the model can
be
Effect = groups(include A, B and A+B) + A*B + adjusters( age, gender)
or stratified models by categorized Age groups, or Gender, etc.
At leaset, the effect models established here can help you to find
difference among the three groups, although it mignt not be the actual
difference among them.
Good luck.
【在 a********y 的大作中提到】![](/moin_static193/solenoid/img/up.png) : suppose it's a survey data with sampling weight : treatment: A (0,1) ,B (0,1) : sex: male; female; : age groups: young, middle, old : all patients may use A, or B, or both A and B at the same time. So the : independent sample assumption does not hold. : Question: is treatment A significantly more adopted than treatment B ( : overall, stratified by sex, age, etc)? : 请问可以 dependent t-test 吗? : 谢谢!!
| p********6 发帖数: 1339 | 3 I think lz is asking about comparing proportions of A and B, not some '
effect' between group A and B.
can
【在 T*******I 的大作中提到】![](/moin_static193/solenoid/img/up.png) : The "independence" of a sample should mean that each individual (here is a : patient) in a sample is independent to all others. So, your sample can be : treated as three independent groups: : Use A only : Use B only : Use A and B : You can take the simple ANOVA with interaction A*B. I believe the model can : be : Effect = groups(include A, B and A+B) + A*B + adjusters( age, gender) : or stratified models by categorized Age groups, or Gender, etc.
| p********6 发帖数: 1339 | 4 T-test is to compare means of a continuous variable in two groups.
What you want to compare are the proportions (of treatment A and B) so t-
test is not appropriate. The test you want to use is McNemar's test. Search
McNemar's test and stratified McNemar's test to get more details. | a********y 发帖数: 474 | 5 You are right!
【在 p********6 的大作中提到】![](/moin_static193/solenoid/img/up.png) : I think lz is asking about comparing proportions of A and B, not some ' : effect' between group A and B. : : can
|
|