求助，nlmixed出现问题：The final Hessian matrix is not positive definite [推广有奖]

用nlmixed做了一个随机效应模型，运行完以后，sas给了下述警告：

The final Hessian matrix is not positive definite, and therefore the estimated covariance matrix is not full rank and
         may be unreliable.  The variance of some parameter estimates is zero or some parameters are linearly related to other parameters.

程序如下：

proc nlmixed data=mylib.quan8;
parms b0=0 b1age2=-0.2238 b2age3=-0.6074  b3age4=-0.8399
b4ind2=0.071 b5ind3=0.0915 b6ind4=0.0801 b7ind5=0.1386 b8ind6=0.1808
b9hh2=0.074 b10hh3=0.1408 b11hh4=0.1435  b12hh5=0.2304
b13edu1=0.0836 b14edu2=0.2339 b15edu3=0.2316 b16edu4=0.2262
b17gen=-0.123 b18insur=0.0679 b19res=0.1839 b20=0 b21=0 b22=0 b23=0
sd=1 i1=1 i2=1;
bounds i1>0,i2>0;
eta=b0+b1age2*age2+b2age3*age3+b3age4*age4
+b4ind2*indinc2+b5ind3*indinc3+b6ind4*indinc4+b7ind5*indinc5+b8ind6*indinc6
+b9hh2*hhinc2+b10hh3*hhinc3+b11hh4*hhinc4+b12hh5*hhinc5
+b13edu1*edu1+b14edu2*edu2+b15edu3*edu3+b16edu4*edu4+b17gen*female
+b18insur*insurance+b19res*resource+b20*time1+b21*time2+b22*time3+b23*time4+u;

if (u48a=1) then p=probnorm(-eta);else if (u48a=2) then
p=probnorm(i1-eta)-probnorm(-eta);
else if (u48a=3) then
p=probnorm(i1+i2-eta)-probnorm(i1-eta);
else p=1-probnorm(i1+i2-eta);
if (p>le-8) then ll=log(p);
else ll=-le100;
model u48a~general(ll);
random u ~ normal(0,sd*sd) subject=sample;

请问一般是什么问题才会出现上述情况？应该怎样解决？谢谢！

[此贴子已经被作者于2009-2-22 20:44:23编辑过]