Re: Why the WARNING occurs?

xg405012 · Posted 02-16-2017 05:40 PM

Hi,

When I run the nlmixed model, I keep getting warning like this

NOTE: By default, formatted unique SUBJECT= variable values are used. In releases prior to

SAS/STAT 13.1, by default, SUBJECT= variable values were not examined for uniqueness. To

revert to the previous behavior, specify the NOSORTSUB option in the PROC NLMIXED statement.

NOTE: Convergence criterion (GCONV=1E-8) satisfied.

NOTE: At least one element of the gradient is greater than 1e-3.

NOTE: Execution error while processing ESTIMATE statement.

NOTE: The SAS System stopped processing this step because of errors.

WARNING: The data set WORK.DATA1 may be incomplete. When this step was stopped there were 382068

observations and 33 variables.

WARNING: Data set WORK.DATA1 was not replaced because this step was stopped.

WARNING: The data set WORK.DATA2 may be incomplete. When this step was stopped there were 382068

observations and 33 variables.

WARNING: Data set WORK.DATA2 was not replaced because this step was stopped.

Here is the code:

%macro focalvar (focalvar);

proc nlmixed data=base qpoints=5;

/* define initial values and bounds */

parms alpha0=-1 alpha1=1 alpha2=1 alpha3=1 alpha4=1 alpha5=1 alpha6=1 alpha7=1 beta0=1 beta1=1 beta2=1 beta3=1 beta4=1 beta5=1 beta6=1 beta7=1 delta0=1 delta1=2 delta2=2 delta3=2 delta4=2 delta5=2 delta6=2 delta7=2 var1=1 var2=1 cov12=.5 k=2;

bounds var1 var2 >=0;

/* Part I log-likelihood */

teta=alpha0 + a + alpha1*during + alpha2*exposed2 + alpha3*during*exposed2+alpha4*&focalvar+alpha5*&focalvar*during+alpha6*&focalvar*exposed2+alpha7*&focalvar*during*exposed2;

expteta=exp(teta);

p=expteta/(1+expteta);

if new_rate=0 then loglik=log(1-p);

/* Part II log-likelihood */

if new_rate=1 then do;

mu=beta0 + b + beta1*during + beta2*exposed2 + beta3*during*exposed2+beta4*&focalvar+beta5*&focalvar*during+beta6*&focalvar*exposed2+beta7*&focalvar*during*exposed2; /* Mean of gen. gamma dist. */

sigma=exp((delta0 + delta1*during + delta2*exposed2 + delta3*during*exposed2+delta4*&focalvar+delta5*&focalvar*during+delta6*&focalvar*exposed2+delta7*&focalvar*during*exposed2)/2); /* Scale of gen. gamma dist. */

eta=abs(k) ** (-2);

u=sign(k)*(log(rate)-mu)/sigma;

value1=eta *log (eta) - log(sigma) -.5 * log(eta) - lgamma(eta);

loglik=log(p) + value1 + u *sqrt(eta) - eta * exp(abs(k)* u);

end;

/* fit the model above */

model rate ~ general(loglik);

random a b ~ normal([0, 0], [var1, cov12, var2]) subject=household_key;

/* generate empirical Bayes estimates for random effects

and store them in SAS datasets data1 and data2; */

predict a out=data1;

predict b out=data2;

estimate '1' alpha1+log(1+exp(alpha0))-log(1+exp(alpha0+alpha1)) +beta1 + log(k**2)/k*( exp((delta0+delta1)/2)-exp(delta0/2) ) +

log( gamma(1/k**2+exp((delta0+delta1)/2)/k) / gamma(1/k**2+exp(delta0/2)/k) );

estimate '2' alpha2+log(1+exp(alpha0))-log(1+exp(alpha0+alpha2)) +beta2 + log(k**2)/k*( exp((delta0+delta2)/2)-exp(delta0/2) ) +

log( gamma(1/k**2+exp((delta0+delta2)/2)/k) / gamma(1/k**2+exp(delta0/2)/k) );

estimate '3' alpha3+log(1+exp(alpha0+alpha1+alpha2))-log(1+exp(alpha0+alpha1+alpha2+alpha3)) +beta3 + log(k**2)/k*( exp((delta0+delta1+delta2+delta3)/2)-exp((delta0+delta1+delta2)/2) ) +

log( gamma(1/k**2+exp((delta0+delta1+delta2+delta3)/2)/k) / gamma(1/k**2+exp((delta0+delta1+delta2)/2)/k) );

estimate '4' alpha4+log(1+exp(alpha0))-log(1+exp(alpha0+alpha4)) +beta4 + log(k**2)/k*( exp((delta0+delta4)/2)-exp(delta0/2) ) +

log( gamma(1/k**2+exp((delta0+delta4)/2)/k) / gamma(1/k**2+exp(delta0/2)/k) );

estimate '5' alpha5+log(1+exp(alpha0+alpha1+alpha4))-log(1+exp(alpha0+alpha1+alpha4+alpha5)) +beta5 + log(k**2)/k*( exp((delta0+delta1+delta4+delta5)/2)-exp((delta0+delta1+delta4)/2) ) +

log( gamma(1/k**2+exp((delta0+delta1+delta4+delta5)/2)/k) / gamma(1/k**2+exp((delta0+delta1+delta4)/2)/k) );

estimate '6' alpha6+log(1+exp(alpha0+alpha2+alpha4))-log(1+exp(alpha0+alpha2+alpha4+alpha6)) +beta6 + log(k**2)/k*( exp((delta0+delta2+delta4+delta6)/2)-exp((delta0+delta2+delta4)/2) ) +

log( gamma(1/k**2+exp((delta0+delta2+delta4+delta6)/2)/k) / gamma(1/k**2+exp((delta0+delta2+delta4)/2)/k) );

estimate '7' alpha7+log(1+exp(alpha0+alpha1+alpha2+alpha3+alpha4+alpha5+alpha6))-log(1+exp(alpha0+alpha1+alpha2+alpha3+alpha4+alpha5+alpha6+alpha7)) +beta7 + log(k**2)/k*( exp((delta0+delta1+delta2+delta3+delta4+delta5+delta6+delta7)/2)-exp((delta0+delta1+delta2+delta3+delta4+delta5+delta6)/2) ) +

log( gamma(1/k**2+exp((delta0+delta1+delta2+delta4+delta7)/2)/k) / gamma(1/k**2+exp((delta0+delta1+delta2+delta3+delta4+delta5+delta6)/2)/k) );

%mend focalvar;

%focalvar (cate1);

%focalvar (cate2);

%focalvar (cate3);

%focalvar (cate4);

%focalvar (cate5);

%focalvar (cate6);

%focalvar (cate7);

%focalvar (cate8);

run;

Except for first two macros, others run successfully. It is so weird....

Thanks!!!!!

mkeintz · Posted 02-16-2017 05:54 PM

Let's see the program from which this originated. Do you have OPTIONS MPRINT enabled so that you can show us the offending sas code reprinted on the log?

--------------------------
The hash OUTPUT method will overwrite a SAS data set, but not append. That can be costly. Consider voting for Add a HASH object method which would append a hash object to an existing SAS data set

Would enabling PROC SORT to simultaneously output multiple datasets be useful? Then vote for
Allow PROC SORT to output multiple datasets

--------------------------

xg405012 · Posted 02-18-2017 08:56 AM

Thanks! Here is the code. It may seems a little bit completed: