

Interactions and cubic terms 

Message/Author 

Anonymous posted on Monday, April 18, 2005  4:24 pm



I have a specific issue dealing with interactions. I am trying to figure out the syntax to interact a series of time variables with my slope and quadratic term in a LCGA model(e.g., 8.11 LCGA model). So, how would I do this? This way I can take into account exposure time. I want to use this as an iteraction term and not an outcome. So, I am sure I don't want to perform survival analysis.I may be overlooking the obvious, but any assistance in this area is greatly appreciated. Also, I was wondering if it is possible to go beyond quadratic terms and include cubed terms in these models and how would they be set up? Thanks 

bmuthen posted on Tuesday, April 19, 2005  6:10 pm



When you say "time variables" do you mean timevarying covariates? If so, you can have them interact with s and q by using the XWITH option. A quadratic growth factor can be specified in the Version 3 growth language by saying i s q c  .... 

Anonymous posted on Tuesday, April 19, 2005  11:51 pm



Thanks. When I am saying "time variables" I am saying that the variable "age" of respondent should interact with the slope (s) and quadratic (q) terms. If this can be accomplished with a XWITH statement, would you give me an example as "i s q" terms cannot be on the right side of the "" symbol? For clarity, are you saying the "c" is the cube term? If so, wouldn't this create an agreement problem in the program as "c" is used in the class statement? I am currently using version 3.12. Assistance with these issues will be great. I can then finish my grant report. Thank you so much for your time and thoughts. 


I assume that you are looking at growth related to a time variable other than age and that you want to control for age differences at different time points. If so, you should regress the growth factors on age or use individuallyvarying times of observation to capture an interaction between growth and age. The fourth term is the cubic growth factor so c is the cubic growth factor in this example. If you have already used c to represent a latent class variabe, then you need to use another name for the cubic growth factor. The names of the growth factors do not need to be i, s, q, and c. 

Anonymous posted on Wednesday, April 20, 2005  10:38 pm



Thank you for the information thus far. I've tried the methods that you have suggested. However, I am unable to get the desired results. That is, I have acquired some data that others have used with Nagin's model through SAS. For some reason, I am unable to arrive at the same results. When I say the same results, I mean the estimates for the intercept, slope, and quadratic term for the four classes that previous research has found. The model is a logquadratic model that produces four classes of violent arrests. The model adjusts the slope and quadratic term for age. I have tried the suggestions that you have offered in conjunction with example 8.11 in the user's guide. At this point I am out of ideas, would you have any other thoughts? 

bmuthen posted on Wednesday, April 20, 2005  11:05 pm



The Nagin/SAS model can be specified as a special case of the general Mplus model. To make sure you are setting up your model exactly as in Nagin's SAS program, why don't you send your SAS output, your Mplus input, output, and data as well as your license number to support@statmodel.com. You may also want to read the Muthen (2004) chapter in the Kaplan handbook which is available on our web site. 

Anonymous posted on Wednesday, April 20, 2005  11:57 pm



I will do so tomorrow. Thank you very much. 

Back to top 

