like.gms : Maximum Likelihood Estimation

Description

This application from the biomedical area tests the hypothesis
that a population of systolic blood pressure can be separated into
three distinct groups.


Reference

  • Bracken, J, and McCormick, G P, Chapter 8.5. In Selected Applications of Nonlinear Programming. John Wiley and Sons, New York, 1968, pp. 90-92.

Small Model of Type : NLP


Category : GAMS Model library


Main file : like.gms

$title Maximum Likelihood Estimation (LIKE,SEQ=25)

$onText
This application from the biomedical area tests the hypothesis
that a population of systolic blood pressure can be separated into
three distinct groups.


Bracken, J, and McCormick, G P, Chapter 8.5. In Selected Applications of
Nonlinear Programming. John Wiley and Sons, New York, 1968, pp. 90-92.

Keywords: nonlinear programming, maximum likelihood estimation, econometrics
$offText

Set
   i 'observations' / 1*31 /
   g 'groups'       / one, two, three /;

Table data(*,i) 'systolic blood pressure data'
                 1   2   3   4   5   6   7   8   9  10  11  12  13  14  15
   pressure     95 105 110 115 120 125 130 135 140 145 150 155 160 165 170
   frequency     1   1   4   4  15  15  15  13  21  12  17   4  20   8  17

   +            16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31
   pressure    175 180 185 190 195 200 205 210 215 220 225 230 235 240 245 260
   frequency     8   6   6   7   4   3   3   8   1   6   0   5   1   7   1   2;

Parameter
   y(i) 'pressure'
   w(i) 'frequency weight'
   c    'constant';

y(i) = data("pressure",i);
w(i) = data("frequency",i);
c    = 1/sqrt(2*3.14159);

display y, w, c;

Positive Variable
   p(g) 'proportion of population'
   m(g) 'population mean'
   s(g) 'population standard deviation';

Variable mlf 'maximum likelihood value';

Equation
   like
   pdef
   rank;

like..      mlf =e= sum(i, w(i)*log(c*sum(g, p(g)/s(g)*exp(-.5*sqr((y(i)-m(g))/s(g))))));

pdef..      sum(g, p(g)) =e= 1;

rank(g+1).. m(g+1)       =g= m(g);

Model
   ml1 'maximum likelihood - ordered'    / like, pdef, rank /
   ml2 'maximum likelihood - unordered'  / like, pdef       /;

p.l(g)  = 1/3;
m.l(g)  = 100 + 30*ord(g); s.l(g) = 15;
p.lo(g) =.1;
s.lo(g) =.1;

* reported solution below gives a nonoptimal solution
* p.fx('one')   = .365;
* p.fx('two')   = .475;
* p.fx('three') = .160;

option domLim  = 1e3;
ml1.workFactor = 1.5;

solve ml1 maximizing mlf using nlp;