|
|
| Message/Author |
|
|
| Yue Yin posted on Tuesday, October 09, 2018 - 9:40 am
|
|
|
Hi,
I generated the dichotomous dataset using the probit in SAS. The formulas are:
a=1;
b=(-2 + (item-1)*4/(tot_items-1));
prob1=1 / (1 + exp(-1*a*(0-b)));
z1=a*(theta - b);
L = z1 ge rand('NORMAL');
/* Probit model */;
lambda=.707;
c=1/sqrt(1-lambda**2);
min_prob=.12;
tau=(probit(min_prob) + (item-1)*probit(min_prob)*-2/(tot_items-1));
prob2=1-probnorm(tau);
z2=(-tau + lambda*theta)*c;
P = z2 ge rand('NORMAL');
a2=lambda*c;
b2=tau/lambda;
then I use Mplus to check if the result are exact the same by using WLSMV, and MLR PROBIT separately.
But a,b,tau,lamda parameters are not exact the same as I set in SAS, they are only partially same. The Mplus code for WLMSV PROBIT is as follows, could you tell me why they are not exact the same?
VARIABLE:
NAMES are child y1-y12 p1-p12;
USEVARIABLES are p1-p12 ;
CATEGORICAL = p1-p12 ;
MISSING are . ;
ANALYSIS:
TYPE = GENERAL;
ESTIMATOR = WLSMV;
!LINK = PROBIT;
PROCESSORS = 2;
COVERAGE = .01;
parameterization=delta;
MODEL:
! Fix loadings for 1PL;
LATENT BY p1*.7 (a);
LATENT BY p2-p12*.7 (a);
LATENT@1; |
|
|
|
|
| I would suggest sending a SAS-related question to SEMNET. |
|
|
| Back to top |
|
|
