 Plotting a conditional Latent Curve M...    Message/Author  Chris Greenwood posted on Monday, August 05, 2019 - 1:20 am
Hi,

I am working with a Latent Curve Model with a number of predictors.

What I am wanting to do is plot the estimate curves, whilst varying one of the predictors (i.e., -1sd, mean, +1sd).

Is there a simple way this can be done?

My current thoughts are to use MODEL CONSTRAINT to re-calculate the growth curves parameters (i.e., intercept, slope, etc.) by varying one predictor (the other predictors kept as the mean).

On this however, I cant figure out how to call the intercept of the intercept (not the mean of the intercept).

I can pull out all the values I need in the savedata command, but would be nice if I could do all the manipulation within Mplus.

Any help would be greatly appreciated.

Thanks,
Chris  Bengt O. Muthen posted on Monday, August 05, 2019 - 5:11 pm
The intercept of the intercept growth factor is called [i]. So just like the mean when there are no predictors of i.  Chris Greenwood posted on Monday, August 05, 2019 - 5:32 pm
Thank you for the quick response.

Also model constrain code is working as anticipated.  Chris Greenwood posted on Monday, August 05, 2019 - 11:25 pm

I am noticing a slight difference in the values from my loop plots between two approaches for deriving the curves.

1. I am bringing into the model the means for each of the predictors and using them in model contraint
[var] (m1);

2. I am not bringing the means for each predictor into the model, but I am pulling them out by saving TECH4 and then entering them into MODEL CONTRAINT as a new variable.
NEW(m1);
m1 = xxxxxxxx;

In both models, the variance for all predictors has been called into the model.

The means of the plots created from these 2 approaches are identical, but the confidence intervals are different (although highly correlated).

Do you know why this discrepancy is occuring? Is it possibly just a decimal place rounding being saved in the TECH4 .csv file?

Thanks again.  Bengt O. Muthen posted on Tuesday, August 06, 2019 - 3:54 pm
Approach 2 does not take into account the sampling error in the means so the SEs are underestimated. Imagine a Monte Carlo simulation where you analyze repeated draws of data - in Approach 1 the mean varies over the draws while in Approach 2 it doesn't.  Chris Greenwood posted on Tuesday, August 06, 2019 - 5:01 pm
Thanks so much!

I am in a situation where I can't call the means of the predictors into the model because it causes the model to crash.

When i plot the estimated growth curves (at levels of one of the predictors), would it be best just to stick to the mean curves, and ignore the confidence intervals, since they will be underestimated.

Thanks again for the feedback.

I have included the model constraint code I am using below :
p = intercepts of growth paramters
i1-q4 = regression coefficients
v = variances
m = means from TECH4

Model Constraint:
PLOT (mean );
NEW (ilow slow qlow clow
m1 m2 m3 m4);

LOOP(x, 0, 12, 1);
m1 = 1.2017545;
m2 = 1.5085389;
m3 = 3.2190728;
m4 = 0.15971218;
ilow = p1 + i1*(m1-SQRT(v1)) + i2*m2 + i3*m3 + i4*m4 ;
slow = p2 + s1*(m1-SQRT(v1)) + s2*m2 + s3*m3 + s4*m4 ;
qlow = p3 + q1*(m1-SQRT(v1)) + q2*m2 + q3*m3 + q4*m4 ;
clow = p4 + c1*(m1-SQRT(v1)) + c2*m2 + c3*m3 + c4*m4 ;

low = ((ilow) + (slow*x) + (qlow*x*x) + (clow*x*x*x));  Bengt O. Muthen posted on Wednesday, August 07, 2019 - 4:31 pm
If you have a crash you should send the input and data to Support.    Topics | Tree View | Search | Help/Instructions | Program Credits Administration