how do you test parameter estimates against 1 in proc nlmixed?

[Ming]

Hi,

The proc nlmixed result generated few parameter estimates that are significant (p<0.0001), these are to show the estimates are significantly difference from zero.  How can test if these parameters are significantly different from 1?

sample results:

Parameter Estimates
Parameter	Estimate	Standard Error	DF	t Value	Pr > |t|	Lower	Upper
a	0.7181	0.06263	26	11.47	<.0001	0.5894	0.8468
n	-0.2904	0.06313	26	-4.6	<.0001	-0.4202	-0.1607
c	0.07872	0.03938	26	2	0.0562	-0.00222	0.1597
delta_a	0.5692	0.1104	26	5.16	<.0001	0.3424	0.796
delta_n	-0.1188	0.07111	26	-1.67	0.1068	-0.265	0.0274
delta_c	0.08615	0.04107	26	2.1	0.0458	0.00174	0.1706

I want to know if "n" is significantly different from 1 instead of 0.

thanks.

ming

[PGStats · Accepted solution]

I would try

ESTIMATE "n vs 1" n - 1.0;

PG

[Ming]

Thank you so much for your help!!!

[JacobSimonsen]

An other way to do this is calculating a likelihood ratio test. You get the -2 log(L) value in the output. Then you just need to run the model Again with n replaced by "1". Below I tried do so on an example from the documentation where I test β₂=1.

data pump;
   input y t group;
   pump = _n_;
   logtstd = log(t) - 2.4564900;
   datalines;
5  94.320 1
1  15.720 2
5  62.880 1
14 125.760 1
3   5.240 2
19  31.440 1
1   1.048 2
1   1.048 2
4   2.096 2
22  10.480 2
;

ods output fitstatistics=m0;
proc nlmixed data=pump;
   parms logsig 0 beta1 1 beta2 1 alpha1 1 alpha2 1;
   if (group = 1) then eta = alpha1 + beta1*logtstd + e;
   else eta = alpha2 + beta2*logtstd + e;
   lambda = exp(eta);
   model y ~ poisson(lambda);
   random e ~ normal(0,exp(2*logsig)) subject=pump;
run;

ods output fitstatistics=m1;
proc nlmixed data=pump;
   parms logsig 0 beta1 1 alpha1 1 alpha2 1;
   if (group = 1) then eta = alpha1 + beta1*logtstd + e;
   else eta = alpha2 + 1*logtstd + e;
   lambda = exp(eta);
   model y ~ poisson(lambda);
   random e ~ normal(0,exp(2*logsig)) subject=pump;
run;

data _NULL_; 
  merge m0(rename=(value=m0)) m1(rename=(value=m1));
  where (descr='-2 Log Likelihood');
  chisquare=(m1-m0);
  pvalue=sdf('chisquare',chisquare,1);
  put pvalue pvalue6.4;
run;