SAS GLM LSMEANS Non-est？WHY?

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

SAS中GLM过程，为什么程序运行后sex的lsmeans显示Non-est？如果我把class变量变成两个，比如class sex  DHABIT;
或class sex age_g  ;程序运行后均能正常显示sex的lsmeans，可是三个class变量反而不行，请高手指教！！！

proc glm data=a;
class sex age_g  DHABIT;
model ADJ_BPAL=SEX AGE_G  DHABIT CREATI_1 AGE_G*SEX AGE_G*DHABIT SEX * DHABIT
AGE_G * SEX * DHABIT/solution;
means AGE_G SEX DHABIT;
lsmeans SEX AGE_G  DHABIT AGE_G*SEX AGE_G*DHABIT SEX * DHABIT
AGE_G * SEX * DHABIT/ stderr pdiff cl tdiff   ;
run;quit;

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏0 回帖

关键词：LSmeans means mean ans Est solution 程序

1# previal

The problem should be from non-estimable L matrix. That is, SAS finds that least square means are not estimable.
Putting a variable in CLASS statement has two consequences. The one is checking its availability even if it is not in the fitting model, and the other is treating as a categorical variable instead of continuous one. In your case, adding one more variable in the CLASS statement will tell if the added variable is classification-modelled. Obviously, 3 categorical variables will further thin or shrink the cells where to hold the events than two categorical variables. That is, the singularity may incur in the thinned cells more often; and thus a more likely resulting singular matrix will prevent SAS from calculating the estimates, here are LS means. This singularity also can result from the collinearity between variables. I would suggest that, to table the three variables and see if there are zero cells (you are in 3-way interaction, i.e., A*B*C).
Maybe more relevant way for your case to solve the problem is to enlarge the value of the option of singular= in LSMEANS statement. The default is 1E-4. However, the remedy may not work at all and thus eventually the more attention should go to your data even that works.
only my 2-cents. JingJu