Anonymous posted on Friday, July 15, 2005 - 5:42 am
I'm doing analyses on data about adolescents between the ages of 12 and 15; I've got three equidistant timepoints. I'm running multivariate LGM to examine whether the process of Y1 is related to the process of Y2. However, repeated measures (and univariate LGM) indicated 1) that the curves are not linear, 2) that significant gender differences exist in both the initial levels and the shapes of the curves. For y1: in the male sample there is a small decline between T1 and T2 and a sharp decline between T2 and T3; in the female sample y1 goes up between T1 and T2 en goes down between T2 and T3. For y2: in the male sample y2 goes down between T1 and T2 en goes up between T2 and T3; in the female sample there is a small decline between T1 and T2 and a sharp decline between T2 and T3.
I assume that your input includes the other parameters that have to be set in a growth model, such as intercepts@0 and factor means free (see the Version 3 User's Guide table comparing the BY specifications with the new growth language for a host of different growth model settings).
More importantly, it looks like you are doing a multiple-group analysis of boys and girls and that seems questionable. This is because their growth shapes are different so that the growth factors refer to different things - it is like trying to make group comparisons of factor means and variances when the loadings are different over groups. Seems like you want to do separate gender analyses.
Anonymous posted on Monday, July 18, 2005 - 7:10 am
Thank you very much, dr. Muthen! I hope you don't mind my rather 'silly questions' ... because I've got three extra questions.
Is this what you meant by the new growth language and the correct approach of the other parameters? Or is something missing? MODEL: i1 s1 | y11@0y12@1 y13 *; i2 s2 | y21@0y22@1 y23 *;
i2 ON i1; s2 ON i1; s2 ON s1;
And suppose I had run univariate LGC analyses on both y1 and y2 and it turned that there was a good fit when the third parameter in the growth curve was 4.2 and -1.7, respectively. Should I use the syntax above then, or the following syntax? MODEL: i1 s1 | y11@0y12@firstname.lastname@example.org; i2 s2 | y21@0y22@1y23@-1.7;
i2 ON i1; s2 ON i1; s2 ON s1;
And suppose there is a significant effect of i1 and/or s1 on s2 (the one with -1.7); how should I interpret this effect?
I am testing a multivariate growth model. The growth factors (intercept, slope and quadratic) of one of the two parallel processes are considered as the dependent variables. On these dependent variables the growth factors (intercept, slope and quadratic) of the other parallel proces is regressed as well as some covariates. I want to test whether these effects are similar for boys and girls. Therefore I did a multigroup analyses. However, I have a lot of convergence problems. What am I doing wrong? My model is:
TITLE: this is an example of a linear growth model for a continuous outcome DATA: FILE IS C:\artikelgroeicovinzetlk.txt; VARIABLE: NAMES ARE idsch id g lft etn verb tot ses talen y1-y5 ilk1-ilk4; USEVARIABLES ARE y2-y5 ilk1-ilk4 g ses verb talen; MISSING ARE ALL (100000000); GROUPING = g (0=boys 1=girls); ANALYSIS: TYPE = MISSING H1 MEANSTRUCTURE;
I noticed that it is not possible to estimate a linear effect for boys and a quadratic effect for girls because then the program Mplus gives an error that it doesn't recognize the quadratic term. Is there no way to model a linear and a quadratic growth processes between two groups in Mplus?
I first did univariate analyses and found that on both processes boys and girls differ at least in one of the three growth factors. Could this be an explanation why the multiple group analysis doesn't work? On a previous similar question in the discussion list you answered 'More importantly, it looks like you are doing a multiple-group analysis of boys and girls and that seems questionable. This is because their growth shapes are different so that the growth factors refer to different things - it is like trying to make group comparisons of factor means and variances when the loadings are different over groups. Seems like you want to do separate gender analyses.' Does this also apply to my case? And if so, if I analyze boys and girls separately how can I make a comparison between the effects? Do you have any suggestions?
With a parallel process latent growth curve model with two continuous outcomes (Y & Z) and 10 covariates (X1-X10), I am interested in examining how these covariates might change the size of the correlation between two slope factors Y & Z, across boys and girls.
Unconditional multiple-group models showed that the two groups had different means and variances for growth factors (both intercept and slope) of outcome Y as well as a different slope factor correlation.
Thus, I fit a conditional multiple-group model where the growth factors for outcome Y were freely estimated while those for outcome X were constrained to be equal across the two groups. Results showed that boys and girls had different sets of significant paths from covariates to slope factor(s) Y and/or Z.
If I would like to check how the size of the slope factor correlation would change by adding to a model a significant covariate one at a time, should the analyses be done for boys and girls separately because the two groups have different sets of significant covariates? Or, should I have fit the conditional model separately for boys and girls since some of the growth factors (mean and variance for both slope and intercept of outcome Y) were found to be different across the gender?
Dear Dr.Muthen, Thanks for your reply. I understand that with covariates in a model, it is the residual correlation between two slope factors.
However, since my interest was to see the influence of added covariates on the slope factor correlation, I tried to compare the correlation between the slope factors in a unconditional model to estimated slope factor correlations (reported under Tech4) in conditional models. Now I am wondering if the estimated correlation in the conditional model is comparable to that of the unconditional model.