Latent Growth Curve without the same ... PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
 Somtawin posted on Wednesday, April 25, 2001 - 7:24 pm
Can I employ latent growth curve model to study the academic growth of schools taking the past mean of achievement score of each school, not using the same student for five consecutive years ?
 Linda K. Muthen posted on Thursday, April 26, 2001 - 8:33 am
In principle, that is possible, although you would need to consider whether the school can be considered the same across years when the student body changes. This is a substantive issue that other more substantive researchers might have better comments on.
 Anonymous posted on Wednesday, July 07, 2004 - 7:32 am
I have data on children spanning two years and six waves, three in the last year of kindergarten, three in the first year of elementary school. Data are about the quality of the teacher-child relationships. I want to estimate a model that captures all data waves, despit the fact that the teacher in the first three waves (kindergarten) is someone else than the teacher in elementary school (the last three waves). So, though the same measures are used at all waves, the scores at first and last waves might not be compared directly. Nevertheless, I guess there might be a solution... How can I include all six waves of data in a single LGC model framework?
 Linda K. Muthen posted on Wednesday, July 07, 2004 - 9:04 am
You could have two sequential growth processes --one for each teacher. Or you could do one growth process. I would do both and then compare them.
 Anonymous posted on Wednesday, July 07, 2004 - 11:52 pm
OK, that's of course what we could do, do both an compare. However, what's on my mind since last night, is whether a piecewise LGC model (maybe with differential slopes also, like in Willet, Singer & Martin, 1998) also makes sense in this context? That is, when measures (scales, observers, etc.) are the same, but only target persons (teachers) differ from kindergarten to elementary school. So, we allow the intercept and slope to vary in the second part of the model, compared to the first. Does this make sense to you, looking at the transition from kindergarten to elementary school as an intervention?
 Linda K. Muthen posted on Thursday, July 08, 2004 - 6:39 am
I think this is the same as having two sequential growth processes.
 Gareth Morgan posted on Tuesday, March 25, 2008 - 12:51 pm
I have data on children's weekly vocabulary scores from a curriculum based assessment and i would like to model their growth in vocabulary over time. However, while the number of vocabulary words is the same for each week, the actual vocabulary words are different for each week. Thus, the scores go up and down from week to week. I was wondering if i could use a running total from one week to next of number of correct vocabulary words and use that in LGM framework. Would this work? If not, do you have another suggestion?
 Linda K. Muthen posted on Tuesday, March 25, 2008 - 1:49 pm
Growth curve modeling assumes the same outcome is measured repeatedly. I don't think a running total would work. Is there some standard test you can use each week in addition to their regular tests? See the paper, The Metric Matters, by Mike Seltzer.
 Annie Desrosiers posted on Thursday, July 03, 2008 - 8:35 am
I have 25 time points from 1975 to 2000 with different cases each year. Thus, it's not the same subjects at each year. I have approximately 3000 subjects each year. Year after year, the same set of questions (on delinquency) was assigned at 3000 students. I want to see how delinquency progresses across time.

Can I do LGM on these subjects like example 6.1 even if itís not the same cases at each time? Each year represents an independent sample. If not, what would be the best strategy of analysis in MPlus?
Thank you.
 Linda K. Muthen posted on Thursday, July 03, 2008 - 10:33 am
You need to have the same subjects across time to estimate a growth model. I'm not sure how people use data such as yours.
 Eugenia Parrett Gwynn posted on Thursday, July 19, 2012 - 6:37 am
I have data on children's relationships with their teachers from kindergarten to sixth grade. I did individual growth modeling to determine if relationsihp quality declines over time--it did. Then I tried to use those growth models (with all 7 time points--ie., kindergarten to 6th graade) to predict an outcome at age 15 using latent growth curve modeling but the models would not converge. I finally dropped the first 3 time points of the growth models (i.e., k-2nd grade) and I was able to get the lgcm to fit. When doing the latent growth curve models, did just the intercept and slope for each child predict the outcome or did the model also include the values of teacher-child relationship quality at each time point as well as each child's intercept and slope?
 Linda K. Muthen posted on Thursday, July 19, 2012 - 9:19 am
If you want help with the non-convergence, please send the output and your license number to

Typically a distal outcome is regressed on the growth factors.
 Eugenia Parrett Gwynn posted on Thursday, July 19, 2012 - 1:42 pm
I use the computer lab on campus so I am not sure of the license number. I guess what I am getting at is when the outcome is regressed on the growth factors, are the growth factors based on an average intercept and an average slope for each child or do the growth parameters include data values for each child at each time point?

I'm sorry. Someone keeps asking me this and I don't know the answer.
 Linda K. Muthen posted on Friday, July 20, 2012 - 10:43 am
Growth factors are random effects. They have means and variances. They include values for each child.
 RuoShui posted on Monday, November 18, 2013 - 4:56 pm
Dear Dr. Muthen,

I am looking at emotional well-being over 4 waves. The subjects are the same across waves but the informants (teachers) are different at each wave. I am wondering whether I can justify using latent growth curve modeling?

Thank you very much.
 Linda K. Muthen posted on Tuesday, November 19, 2013 - 8:58 am
This question is more appropriate for a general discussion forum like SEMNET.
 Lucy Markson posted on Thursday, March 05, 2015 - 6:57 am
I would like to conduct LGM using Child Behavior checklist data over 3 years. Although the CBCL is used with the same participants at all waves, the items used vary due to age appropriateness of the CBCL. Is there a recommended way of adjusting for this using Mplus for LGM?

Many thanks
 Bengt O. Muthen posted on Thursday, March 05, 2015 - 9:39 am
If you have item-level data and have some items in common at adjacent time points it is possible to model this so that you have a comparable scale at all time points. You hold equal parameters for the items that are the same at the different time points.
 Jill Rabinowitz posted on Friday, January 26, 2018 - 10:21 am
Hi there,

I am running a longitudinal latent class analysis and my indicators are not the same across time. To be more specific, my first 5 indicators refer to frequency of alcohol use in the past year measured from ages 14-18 and the next 8 indicators refer to frequency of the alcohol use in the past month from ages 19-26. Given that the scales are different for the indicators (i.e., past month vs. past year use), would it be okay to just z-score the indicators to put them on the same scale? Is there an alternative approach you would recommend?

 Bengt O. Muthen posted on Friday, January 26, 2018 - 4:08 pm
I wouldn't get into z-scores but just accept that the measures are different at the two tome points. To some extent they convey similar information. You can still consider transitions between time 1 and time 2 classes even when the classes have different meaning.
 AT Jothees posted on Monday, February 05, 2018 - 12:09 pm

I have a generic question. Is it possible to examiner age-period-cohort effect in latent higher order growth curve models? If so, is there any mplus example I read and understand ?

Many thanks in advance


 AT Jothees posted on Monday, February 05, 2018 - 12:20 pm

I have a generic question. Is it possible to examine age-period-cohort effect in latent higher order growth curve models? If so, could you please suggest mplus example I can read and understand

Many thanks in advance


 Bengt O. Muthen posted on Monday, February 05, 2018 - 1:27 pm
You may want to ask about this on SEMNET. I assume you are aware of the Mplus User's Guide ex 6.18. That can certainly be expanded to higher order growth curves.
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message