xuyu608 发表于 2010-4-30 04:41
请问广义线性混合模型(binary error and logit link function)需要避免多重共线性问题吗?我应用该模型处理一组数据时,发现对一个变量的预测与实际均值相反。我怀疑是多重共线性导致的;我的样本量相对可以(分别为102和64)。我不是很懂统计学知识,希望大家能给我些意见。在此谢谢了!
See the following example, the coefs are opposite due to the definition of events.
data t1;
do i = 1 to 1000;
CSI=rannor(1);
Complaint=ranuni(1)<=0.3;
Loyalty=ranuni(1)<=0.5;
Attitude=1+1* CSI-1*Complaint+1*Loyalty > rannor(1);
output;
end;
run;
proc logistics data=t1;
model Attitude (event='0') = CSI Complaint Loyalty/link=probit;
run;
proc logistics data=t1;
model Attitude (event='1') = CSI Complaint Loyalty/link=probit;
run;