y******6
发帖数: 47 | 1 I have a question about mixed models too. For example, Var1 is location(
cities), Var2 is treatment, Var3 is Time (season) Var4 is Year(2008,2009,
2010). I want to make comparsions to see whether 2008,2009 and 2010 have the
same mean or not within each treatment.
proc mixed data= method=type3;
where treatment='Treatment1';
class Location Time Year;
Model counts=time year time*year;
random location;
run;
Type3 tests of fix effects: year, p-value<0.05
While if I change the random from location to LOCATION LOCATION*YEAR
THE p-value>0.05. Covariance Parameter Estimates result is that: location
z=0.3829 location*year z=0.02.
For both two models described above, their AIC, SBC, AICC are quite similar
though.
My question is that how to choose the random effect since that will affect
my final conclusion significantly.
Thanks. |
|
