Prev: Mainframe with SAS Enterprise Guide
Next: Data Structure
From: Mene Nema on 15 Jun 2010 17:38
Dear Users,

I am trying to specify a simple structural equation equation model
using NLMIXED.

There are only two latent predictor variables (f1 and f2) with 3 and 2
indicators (x-variables) and one latent criterion variable "eta" with
4 indicators (y-variables). The multivariate/bivariate distribution of
the latent exogenous predictor variables (f1,f2) is a finite mixture
distribution of two bivariate normal distributions with unknown mean
and covariance structure. The latent endogenous variable "eta" is
regressed on "f1" and "f2". I wrote some code, but, unfortunately, SAS
returns the following error message: "Execution error for observation
1."

There must be something fundamentally wrong. And I don't have a clue.
I have seen some posts of Dale McLerran, but they did not work for my
structural equation model.

It would be great if somebody had a hint.

Thank you very much!

Best,
Mene

My SAS Code:

*****************************************************************************;
DATA testdata;
INFILE "c:\temp\ZIndicators3b.txt";
INPUT dummy id X1 X2 X3 X4 X5 Y1 Y2 Y3 Y4;
RUN;


TITLE "testdata";
PROC PRINT DATA=testdata(OBS=574);
RUN;

proc nlmixed data = testdata (keep = dummy id X1-X5 Y1-Y4);

***starting values;
parms
lamx21 = 0.9401 lamx31 = 1.0768 lamx52 = 1.1720
lamy21 = 1.0659 lamy31 = 1.0637 lamy41 = 1.0551
interx1 = 0.000 interx2 = 0.000 interx3 = 0.000 interx4 = 0.000
interx5 = 0.000
intery1 = 0.000 intery2 = 0.000 intery3 = 0.000 intery4 = 0.000
thd1 = 0.4582 thd2 = 0.5770 thd3 = 0.3726 thd4 = 0.2938 thd5 =
0.0321
the1 = 0.2594 the2 = 0.1588 the3 = 0.1623 the4 = 0.1757
gamma1 = 0.5284 gamma2 = 0.1754
phi1a = 0.5334 phi2a = 0.7022 phi1b = 0.5334 phi2b = 0.7022 rhoa =
0 rhob = 0
prob = 0.1 to 0.9 by 0.01 psi = 0.5035 mf1a = -.5 mf2a = -.5 mf1b
= .5 mf2b = .5;

bounds -1< rhoa < 1, -1< rhob < 1, phi1a>=0, phi2a>=0, phi1b>=0,
phi2b>=0, thd1-thd5>=0, the1-the4>=0, psi>=0;

phi21a=(phi1a*phi2a)**0.5*rhoa;
phi21b=(phi1b*phi2b)**0.5*rhob;

mu1a = interx1 + 1*f1a;
mu2a = interx2 + lamx21*f1a;
mu3a = interx3 + lamx31*f1a;
mu4a = interx4 + 1*f2a;
mu5a = interx5 + lamx52*f2a;

mu1b = interx1 + 1*f1b;
mu2b = interx2 + lamx21*f1b;
mu3b = interx3 + lamx31*f1b;
mu4b = interx4 + 1*f2b;
mu5b = interx5 + lamx52*f2b;

pi = constant('pi');

Px1a = 1 / (((2 * pi) ** 0.5) * thd1) * exp(-0.5 * ((X1 - mu1a) /
thd1) ** 2);
Px1b = 1 / (((2 * pi) ** 0.5) * thd1) * exp(-0.5 * ((X1 - mu1b) /
thd1) ** 2);

Px2a = 1 / (((2 * pi) ** 0.5) * thd2) * exp(-0.5 * ((X2 - mu2a) /
thd2) ** 2);
Px2b = 1 / (((2 * pi) ** 0.5) * thd2) * exp(-0.5 * ((X2 - mu2b) /
thd2) ** 2);

Px3a = 1 / (((2 * pi) ** 0.5) * thd3) * exp(-0.5 * ((X3 - mu3a) /
thd3) ** 2);
Px3b = 1 / (((2 * pi) ** 0.5) * thd3) * exp(-0.5 * ((X3 - mu3b) /
thd3) ** 2);

Px4a = 1 / (((2 * pi) ** 0.5) * thd4) * exp(-0.5 * ((X4 - mu4a) /
thd4) ** 2);
Px4b = 1 / (((2 * pi) ** 0.5) * thd4) * exp(-0.5 * ((X4 - mu4b) /
thd4) ** 2);

Px5a = 1 / (((2 * pi) ** 0.5) * thd5) * exp(-0.5 * ((X5 - mu5a) /
thd5) ** 2);
Px5b = 1 / (((2 * pi) ** 0.5) * thd5) * exp(-0.5 * ((X5 - mu5b) /
thd5) ** 2);

Px1 = Px1a * prob + Px1b * (1 - prob);
Px2 = Px2a * prob + Px2b * (1 - prob);
Px3 = Px3a * prob + Px3b * (1 - prob);
Px4 = Px4a * prob + Px4b * (1 - prob);
Px5 = Px5a * prob + Px5b * (1 - prob);

eta = gamma1*f1a + gamma2*f2a + zeta;
etb = gamma1*f1b + gamma2*f2b + zeta;

mu6a = intery1 + 1*eta;
mu6b = intery1 + 1*etb;

mu7a = intery2 + lamy21*eta;
mu7b = intery2 + lamy21*etb;

mu8a = intery3 + lamy31*eta;
mu8b = intery3 + lamy31*etb;

mu9a = intery4 + lamy41*eta;
mu9b = intery4 + lamy41*etb;

Py1a = 1 / (((2 * pi) ** 0.5) * the1) * exp(-0.5 * ((Y1 - mu6a) /
the1) ** 2);
Py1b = 1 / (((2 * pi) ** 0.5) * the1) * exp(-0.5 * ((Y1 - mu6b) /
the1) ** 2);

Py2a = 1 / (((2 * pi) ** 0.5) * the2) * exp(-0.5 * ((Y2 - mu7a) /
the2) ** 2);
Py2b = 1 / (((2 * pi) ** 0.5) * the2) * exp(-0.5 * ((Y2 - mu7b) /
the2) ** 2);

Py3a = 1 / (((2 * pi) ** 0.5) * the3) * exp(-0.5 * ((Y3 - mu8a) /
the3) ** 2);
Py3b = 1 / (((2 * pi) ** 0.5) * the3) * exp(-0.5 * ((Y3 - mu8b) /
the3) ** 2);

Py4a = 1 / (((2 * pi) ** 0.5) * the4) * exp(-0.5 * ((Y4 - mu9a) /
the4) ** 2);
Py4b = 1 / (((2 * pi) ** 0.5) * the4) * exp(-0.5 * ((Y4 - mu9b) /
the4) ** 2);

Py1 = Py1a * prob + Py1b * (1 - prob);
Py2 = Py2a * prob + Py2b * (1 - prob);
Py3 = Py3a * prob + Py3b * (1 - prob);
Py4 = Py4a * prob + Py4b * (1 - prob);

LL = log(Px1 * Px2 * Px3 * Px4 * Px5 * Py1 * Py2 * Py3 * Py4);

model dummy ~ general(LL);

random f1a f2a f1b f2b zeta ~ normal([mf1a, mf2a, mf1b, mf2b,0],
[phi1a,
phi21a,
phi2a,

0, 0, phi1b,
0, 0, phi21b, phi2b,
0, 0, 0, 0, psi]) subject = id;

run;
*****************************************************************************;
 | 
Pages: 1
Prev: Mainframe with SAS Enterprise Guide
Next: Data Structure