Message/Author 

Jean Frisou posted on Sunday, August 06, 2006  6:34 am



I have some problem with a model constraint. The syntax is the same as in the example 5.20. The ouput file incates an error *** ERROR Unknown parameter label in MODEL CONSTRAINT: LA11 The part of program file is this one: DATA: FILE IS C: Etude RL/Mplus/modèle/ artram/datamod.dat; VARIABLE: NAMES ARE x11x16 x21x26 y21y24; USEVARIABLES ARE x11 x12 x15 x21 x22 x25; ANALYSIS: ESTIMATOR = MLMV; DIFFTEST = etatrait.dat; MODEL: ENGA1 BY x11@1(la11); ........... MODEL CONSTRAINT: NEW(rel11 rel12 rel15 rel21 rel22 rel25 ); rel11 = la11**2*vf1/(la11**2*vf1 + ve11); rel12 = la12**2*vf1/(la12**2*vf1 + ve12); rel11 = rel21;rel12 = rel22; OUTPUT: TECH4; SAMPSTAT; RESIDUAL; STANDARDIZED; Can somebody indicates me where is the error ??? Thank you for the help 


I can't see from what you have posted unless it is that you are labelling a fixed parameter. Please send your input, data, output, and license number to support@statmodel.com. 

Jill McClain posted on Thursday, September 10, 2009  5:20 pm



Hi Drs. Muthen. I am trying to use the model constraint command to generate predicted values for a dependent variable for set values of my independent variables (from a linear regression). I can do this by hand, of course, but I'm hoping that using labeled parameters in Mplus will properly propagate the errors so that my predicted values will have appropriate confidence intervals (please let me know if this is not the case). However, I cannot figure out how to label or otherwise include the intercept (other than simply entering the intercept value from the output, in which case the error won't be accounted for). Is it possible to label the intercept in the model command or otherwise indicate that I want to use the estimated intercept and its standard error in the model constraint calculation? Thanks. Jill 


If you have y ON x (b); the intercept is simply referred to as [y] (a); so that you can write Model Constraint: New(yhatxi); yhatxi = a + b*xi; where yhatxi gets a point estimate and a SE that takes into account the sampling error in a and b. 


Thanks very much. That worked perfectly. Oddly, though, the SE for yhatxi is substantially smaller when I use the labeled parameter "a" than when I simply insert the numerical value of a (which has no error as far as Mplus knows) into my constraint equation. Is that correct? 


Check Tech3  maybe the a and b estimates are negatively correlated. 


Hm, yes. There are actually 18 betas in the equation (this is a very large cohort study), and 17 are negatively correlated with the intercept. Is this a problem? Thanks for any insight you can offer! 


That's not a problem, but it explains the reduction in SE that you reported. 


Excellent. Thanks very much, as always! 


Dear Drs. Muthen Is it possible to label the threshold in a probit analysis, in the same way as you indicate in the response to the question by Jill McClain? y ON x (b); the intercept is simply referred to as [y] (a); so that you can write Model Constraint: New(yhatxi); yhatxi = a + b*xi; My attempt produced an error message. 


Yes, but probit gives a threshold tau, [y$1] (tau); instead of an intercept, and the probit replaces y, probit = tau + b*x; You can also translate the probit into a probability using the Phi function. 

anonymous Z posted on Thursday, July 28, 2016  10:13 am



Dear Drs. Muthen, I am using model constraint to create indirect effects, and meanwhile generate bias corrected bootstrap confidence intervals for indirect effects. The output gives only 95%CI for unstandardized results Lower 2.5% Estimate Upper 2.5% New/Additional Parameters O1 0.333 0.175 0.044 O2 0.395 0.175 0.054 Given the intervals excluding zero, can I just report that the indirect effects were significant or should I create standardized results? Thanks so much! Jing 


Bootstrapped CIs are unlikely to disagree for raw and standardized indirect effects. I typically decide on significance based on the raw and if I want to describe the effect size I compare it to the SDs of the X and Y (that is, I consider the standardized effect)  but without further discussing significance for the standardized. 


Dr. Muthen, thanks so much! 

Back to top 