Analyzing Latent Growth Curve Models in Mplus

QuantFish instructor Dr. Christian Geiser explains latent growth curve models and shows how to analyze these models in the Mplus statistical software.

#Mplus #Mplusforbeginners #statistics #CFA #SEM #longitudinal #growthcurve #mplusforbeginners #geiser #quantfish #statisticstutorials

Get Dr. Geiser's FREE MPLUS FOR BEGINNERS COURSE:

Dr. Geiser's BOOKS:

ON-DEMAND MPLUS COURSES:

GET IN TOUCH with DR. GEISER:

ABOUT THIS INSTRUCTOR:
Dr. Christian Geiser is a quantitative psychologist, author of two books on Mplus, and a leader in the development of latent variable techniques for complex data. With his accessible books and sought-after workshops, he has helped thousands of researchers and students around the world to achieve their analytic goals.

ABOUT QUANTFISH:

Some affiliate links may be included at no additional cost to purchasers.

[QuantFish]

How is it going with your growth curve analyses in Mplus? Let us know here in the comments!

[gertlang]

Dear Christian, thanks also for this introductory video! Currently I am facing problems with my own latent growth curve model (LGCM). I have a latent variable (BGF) with 3 continuous measurement indicators for each of 3 time points. Therefore I modeled for each time point a latent factor (BGF) with three measures (str1-3, pro1-3, erg1-3; note: the number indicates the respective time point) and tested in a multiple CFA (configural model). I also tested for measurement invariance (IV) across the 3 time points. There is partial strong IV across time points (all loadings are equal across time and 7 out of 9 incepts are equal) with quite good fit and high reliability. The estimation of the LGCM also went fine but by default Mplus fixes the intercept-factor (INT) to zero. In your book you freely estimate the mean of INT by adding the command line [INT]; but here it comes to estimation/convergence problems. Below I list you my model command:
MODEL:
BGF1 BY str1
pro1 erg1 (1-2); !Slopes/Ladungen über Zeit gleich
BGF2 BY str2
pro2 erg2 (1-2);
BGF3 BY str3
pro3 erg3 (1-2);

[str2 str3] (3); !7 von 9 Intercepts über Zeit gleich (partial strong IV)
[pro1 pro3] (4);
[erg1 erg2 erg3] (5);

pro2 with pro3; !Spezifikation einer Messfehlerkovarianz über die Zeit

[INT]; !this produces the problem, without the estimation works fine.

And here it comes to problems. Mplus warning is:

> THE MODEL ESTIMATION TERMINATED NORMALLY
> THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES COULD
> NOT BE COMPUTED. THE MODEL MAY NOT BE IDENTIFIED. CHECK YOUR
> MODEL. PROBLEM INVOLVING THE FOLLOWING PARAMETER:
> Parameter 18, [ INT ]
> THE CONDITION NUMBER IS -0.327D-12.
> FACTOR SCORES WILL NOT BE COMPUTED DUE TO NONCONVERGENCE OR
> NONIDENTIFIED MODEL.

Do you maybe have any idea what is going on here? For ease of interpretation it would be great if I could get an estimated mean of the INT instead of a zero.

Many thanks,
Gert

[frok]

Great explanation! Thank you for this!

[laurakm]

Thank you very much for the video! It would be great if you could make a video about Growth Mixture Modeling and Latent Class Growth Mixture Models.

[frankenbert]

Thanks for this instructive video. I was wondering how to post-hoc estimate power and effect sizes. I would be thankful for a recommendation or video on this topic ;)

[frok]

Hi Christian, this is a great explanation of how to run growth curve models in MPLUS. I was wondering if you could speak more about R-Square issues. You mentioned that values (R-square) for the repeated measures around 0.5 could indicate that there is an omission of situation-specific effects. Could you talk more about this and how to deal with this? Or what would be potential remedies?

[cynthiaortiz]

Hello,

Do you have a video explaining ALT models in Mplus?

[tomasarriaza]

If age is significant for theses trajectories, should time be set in terms of age? For example, I interviewed adolescents about their alcohol use over 4 years. At T1, adolescents were between 11 and 14, T2 between 12 and 15, and so on. In this example, Should the growth factor loadings be fixed from 0 to 3 or from 0 to 6 (ages 11 to 17)? If it is 0 to 6, how many observations per T and per participant should there be at least? Is there any problem related to missing data?

[longzhudong]

Thank you so much!
I am wondering if you could help me with this question that bothers a for a long time. In your example, there are individuals that have positive growth slope, also individuals with negative growth slope. Is there a way to test what causes such positive vs negative growth pattern? I understand we can add covariates and test if these factors slow down or increase the growth slope, but can I find out what causes/separate positive vs negative growth pattern?
Thanks!

[lyl]

Dear Christian, thank you for this video! I tried this model by following your syntax. The weird thing is mplus did not give me the model fit information (Chi-square, RMSEA, etc.), but it gave sample statistic, UNIVARIATE sample statistic, MODEL RESULTS, and MODEL COMMAND WITH FINAL ESTIMATES USED AS STARTING VALUES. Do you know the reasons or any possible solutions to get the model fit information in this case? Thank you a lot.

[noahtoh]

Hi Christian,

Thank you for the tutorial - it was clear and easy to follow. In the example that you have shown, the mean for the slope factor was positive, and the correlation between the intercept and slope was negative. In this case, individuals with higher starting levels of IQ have smaller increases in IQ, and those with lower initial values of IQ have greater increases over time.

I am wondering what is the interpretation for a negative mean value for the slope factor, and a negative correlation between the intercept and the slope. Does this mean that individuals with higher starting values of IQ experience steeper decline in IQ, while those with lower initial IQ have smaller decrease in IQ over time?

Thank you again!

[birdwatche]

Hi Cristian this is really helpful! Thank you! I have one question about the time score though. I want to fit a linear trajectory but my time is nonequidistant, measured at age 1, 3, 5, 9, 15. Do you know how I can deal with this?