I am working with 5000 cases/children referred to the child welfare system, each with 4 waves of data. I am going to model the relationship between the number of placements between waves and score on a behavior problems measure. Label the four number of placement variables np1, np2, np3, and np4 and the four behavior variables b1,b2,b3, and b4. My basic model is:
!placement on placement; np2 on np1; np3 on np2; np4 on np3; !behavior on behavior; b2 on b1; b3 on b2; b4 on b3; !behavior on placement; b1 on np1; b2 on np2; b3 on np3; b4 on np4; !placement on behavior; np2 on b1; np3 on b2; np4 on b3;
I need suggestions on how to integrate two types control variables into the analysis: 1) those constant across waves (gender, initial type of maltreatment, and ethnicity) and 2) those that change by wave (age group, type of placement residing in, and length of time resided in placement).
Do I simply enter the control variables into the above regressions? Would building a growth factor help? Should I build the controls into the model?
also, could i use a multiple groups strategy to examine whether a covariate/control has a significant effect on the overall model?
For instance (sticking with the model statement that i wrote out on the prior email). could i state that model statement for, say, boys, and then repeat the entire model statement for girls? would doing so provide a test of whether gender contributes significantly to the overall model?
I had pictured testing a categorical covariate as a grouping variable by comparing some sort of (probably?) chi-square based measure of model fit of the model with and without the covariate.
could you possibly provide some guidance on 1) how to test intercepts, 2) how to test regression coefficients, 3) how one would get an overall chi-square test (or other related test) of whether the covariate was contributing significantly.
If i simply repeated the model statement for the second group?, what does that test and not test (coefficients? intercepts?) or is such a statement way off track.
The Topic 1 short course handout shows how to use chi-square difference testing with multiple group analysis to test the difference of parameters across groups. There is also a video of the Topic 1 short course available. Difference testing using chi-square or the loglikelihood is also discussed in Chapter 13 of the Mplus User's Guide in the multiple group discussion.