Message/Author 


Hi, I'm wondering if there's a way to plot adjusted probabilities within MPlus for the values of a categorical outcome. From my reading of the user guide it seems that "plot3" should be able to do this, but when I try it I only see four options in the graph module: histograms, scatterplots, item characteristic curves, and information curves. Am I missing something, or do I need to do this in another program? Thanks a lot. 


Mplus does not provide the plot that you want. 

Seth Frndak posted on Tuesday, September 08, 2015  2:15 pm



I have a similar question. Instead of using the PLOT function to graph the adjusted/conditional probabilities in a categorical growth model, would it be possible to use the estimates produced in the Mplus output to plot conditional growth by hand? 


Here is an example based on UG example 8.4, changed to a nonmixture run. After I run this example, I choose View Plots and select the plot “Sample proportions and estimated probabilities”. Then I choose “Plot estimated probabilities, conditional on a set of covariates”. Then I click Next and choose variable U1. Then I click Next. I click on “Name a set of values”. I will call this X=0. Then I doubleclick on X in the Values box and set it to 0. Then I click Finish. Now the plot shows the estimated probabilities conditional on X=0 for the 5 categories of U1. From the Plot menu, choose “Save plot data” and select a filename. I call this file “u1onx=0.dat”. This file contains the estimated probabilities for the 5 categories of U shown in the plot. The first column holds the xvalues. The second columns holds the estimated probabilities. Do this for U2, U3 and U4. Then you can combine the 4 plot data in an Excel file. Column 1 contains values for U1, column 2 contains values for U2, etc. Row 1 will be category 1, row 2 will be category 2, etc. Now you can plot the values across columns for each row. 

Seth Frndak posted on Wednesday, September 16, 2015  1:04 pm



This is exactly what I was looking for. Thank you for your detailed response! 

Seth Frndak posted on Wednesday, September 16, 2015  1:05 pm



If I were to make my predictors latent variables, is there still an option to plot the estimated probabilities? 


I don't think so currently, unless you predict from estimated factor scores used as covariates. But we are putting this plotting matter on our todo list. 

Seth Frndak posted on Wednesday, December 21, 2016  7:01 am



Dr. Muthen, I want to thank you for your earlier help on this matter. I have a few additional questions. Currently, we are using a twopart growth model to test for moderation between two observed variables on the latent growth factors of the twopart model. The trajectory of interest is in the dichotomous portion of the twopart model. We are using the plotting option you described previously (Sept. 10th 2015), to plot conditional trajectories based on high and low values of the observed variables, however we are being asked by a reviewer to test the simple slopes of these effects (point of statistical significance of the slope and intercept means at high/low values of the predictors). We were under the impression that this approach is not possible for these models because of the nature of numerical integration. If this is correct, could you explain why this is not possible? 


Simple slopes can be tested as exemplified in http://www.statmodel.com/download/SimpleSlopes.pdf That approach is relevant for the slope and intercept means. Regarding the numerical integration, that pertains to the estimated trajectories. 

Back to top 