m******3
发帖数: 62 | 1 我用SAS run
proc mixed data=mix method=reml noclprint;
class ID treatment;
model score=treatment treatment*time1 treatment*time2
/solution noint;
random intercept time1 time2/subject=id;
repeated/subject=id;
weight wgt;
run;
结果DF 为0， 无法得到p-vlaue. 那位大牛给s说说，问题在哪呢？
谢谢 |
c**********5
发帖数: 653 | 2 First all,if if time1 and time2 in the interaction term,shoud it in the
model first? |
o*****e
发帖数: 50 | 3 I think random effects should not appear in model statement for proc
mixed.
【在 c**********5 的大作中提到】
: First all,if if time1 and time2 in the interaction term,shoud it in the
: model first?
|
l****u
发帖数: 529 | 4 If time1 and time2 are random effects, treatment*time1 and treatment*time2
should also be random effects. Then both of them should
be put in the random statement. |
d**s
发帖数: 1255 | 5 it should be the repeated statement rather than random statement
【在 o*****e 的大作中提到】
: I think random effects should not appear in model statement for proc
: mixed.
|
d**s
发帖数: 1255 | 6 that's not true.
【在 l****u 的大作中提到】
: If time1 and time2 are random effects, treatment*time1 and treatment*time2
: should also be random effects. Then both of them should
: be put in the random statement.
|
l****u
发帖数: 529 | 7 It should be true. If not, could you explain why?
【在 d**s 的大作中提到】
: that's not true.
|
S******J
发帖数: 30 | 8 1. Most of the time you should not use Random and Repeated cards at the same
time.
2. You only need time1 and time2 in the Random card, not treatment*time1 and
treatment*time2.
3. You may want to specify the variance structure as well. |
l****u
发帖数: 529 | 9 大侠，能给解释一下吗？
他把time1 and time2 设为random effect，那么treatment*time也应当为random
effect
呀，或者至少不能放在model中啊。
多谢了
same
and
【在 S******J 的大作中提到】
: 1. Most of the time you should not use Random and Repeated cards at the same
: time.
: 2. You only need time1 and time2 in the Random card, not treatment*time1 and
: treatment*time2.
: 3. You may want to specify the variance structure as well.
|
s*****9
发帖数: 108 | 10 贴一部分OUTPUT出来看看吧，specify your correlation structure。 可能就OK了 |
|
|
S******J
发帖数: 30 | 11 Random coefficient estimates the random effects due to subjects on slope
with respect to each treatment, not across treatments. |
l****u
发帖数: 529 | 12 大侠，我学艺不精，还是不明白，请您谅解，并多指较。
If treat*time effect is kept in the model, the F test of treat is MStreat/
MSE. But actually, the F test of treat should be MStreat/MStreat*time. It
will influence the p value of treatment and the comparison between
treatments.
Thanks |
d**s
发帖数: 1255 | 13 you need Random and Repeated at same time when you model the variance
structure for residuals
same
and
【在 S******J 的大作中提到】
: 1. Most of the time you should not use Random and Repeated cards at the same
: time.
: 2. You only need time1 and time2 in the Random card, not treatment*time1 and
: treatment*time2.
: 3. You may want to specify the variance structure as well.
|
d**********r
发帖数: 24123 | 14 你的 treatment*time1 或者 treatment*time2 只有一个对应的sample了吧。
每个effects没有>=2 的 repition, 你的DF 就会为0了。 |
s*r
发帖数: 2757 | 15 most of follow-up responses contain wrong information. i would encourage
people to read this
http://www.stat.ncsu.edu/people/davidian/courses/st732/notes/ch |
S******J
发帖数: 30 | |
