NLMIXED的问题求教

panaceapan

数据是binary，有subject effect (random), rater effect (fixed) 和两者的interaction. 想test rater effect 和 interaction =0。

Nlimixed 我不是很熟，下面是我写的。哪位能帮我看看

   teta=beta0 +beta1*rater +b1+b2*rater; ？

proc nlmixed data=agree;

    parms beta0=1.5 beta1=0 d11=0.5 d22=0;

    d11new=exp(d11);* make sure we have nonzero variance;

    d22new=exp(d22);

    teta=beta0 +beta1*rater +b1+b2*rater;

    expteta=exp(teta);

    p=expteta/(1+expteta);

    model outcome~binary(p);

    random b1 b2 ~ normal([0,0], [d11new,0,d22new]) subject=id;

    contrast “test” beta1,d22new;

run;

另外一种做法是用glimmix,但是我不知道怎么同时test rater effect和interaction.

proc glimmix data=agree;

class id rater;

model outcome(event=“1”)=rater/dist=binary solution;

random id id*rater;

run;

谢谢！