|
I did not know this syntax was allowed.
Typically a "where" statement subsets the data. So I would expect that
in this instance, you are only running the first analysis on the subset
of data where dose in "A" or "B", whereas in the second, you're running
the analysis on all the data.
Check your degrees of freedom to see if this is the case. If it is,
then the SE for the latter analysis is based on many more data points,
hence, tighter CI.
Warren Schlechte
-----Original Message-----
From: Joakim Englund [mailto:joakim.englund@GMAIL.COM]
Sent: Tuesday, July 29, 2008 11:52 AM
Subject: Estimate statement in proc mixed
Hi,
I have run two types of proc mixed code. Variable 'Dose' has 10 levels.
*Alt 1:*
proc mixed data=mydata;
class dose;
model result = dose;
where dose in ('A', 'B');
estimate 'B-A' dose -1 1 / cl;
run;
*Alt 2:*
proc mixed data=mydata;
class dose;
model result = dose;
estimate 'B-A' dose -1 1 0 0 0 0 0 0 0 0 / cl;
run;
I thought both these methods tested B-A, but they do not give identical,
albeit similar results I've discovered. The later one gives more narrow
confidence limits and is coherent with the diffs option under the
lsmeans
statement.
Can anybody explain what the real difference is between the
alternatives.
Kind Regards,
Joakim
| 
