I am writing for advice regarding graphing significant interactions terms in equations using latent growth curve modeling (LGCM)
Here are some details:
We are using 4 waves of data from a national survey; these data were collected over a 14 year period at unequal intervals.
The key measures for these analyses include chronic health conditions, self-rated health, perceived financial status, poverty status, and life satisfaction. These variables are assessed consistently across the four data collections.
We are examining whether life satisfaction moderates the negative effects of perceived financial status and actual poverty level on our health related outcomes. We used LGCM within MPlus version 7.4 and assessed individual changes across time. In this analytical approach, repeated measures data were used as indicators of latent variables to model baseline levels (measured as intercepts) and rates of change (measured as slope factors) to describe differences between individual trajectories.
Do you have suggestions as the easiest way to calculate a range of representative values as a way of representing the interactions?
First, are your variables that interact time-invariant covariates predicting the growth factors? Have you looked at UG ex 3.18 to see how interactions can be plotted? See also Chapter 1 in our new book.
Thank you for your quick reply. No they are not time-invariant. Each of the variables was measured at each time point and was able to vary over time. For example, we are looking to see if life satisfaction over time can buffer the effects of changing perceived financial status has on health outcomes.
We have found significant main effects as well as well as a significant interaction term, we are just trying to determine how we can visually show this. Thanks!
Ok, so you are interested in the regressions at each time point of the outcome on the tvc's and their interactions. So say that y is the outcome, x is financial status, z is poverty, and w is the life satisfaction moderator,
y = b0+b1*b2*z+b3*w+b4*x*w+b5*z*w+e = b0+(b1+b4w)x+(b2+b5w)z+b3w+e
You can express (b1+b4w)x and (b2+b5w)z in Model Constraint using LOOP and PLOT as on our Mediation page:
Thank you for your response! One last question, is there an option for obtaining/saving predicted values from a model as is the case in other software programs for linear and logistic regression? And then one could access the saved file? And if yes, would it possible to run 4 sub-group models?
I am attempting to do this plot to look at the interaction, but I am getting errors in my gh5 file when trying to read it into R. Could you please check my syntax and let me know if there is something I am doing wrong?
Variable: names = ID Sex Age Region Educ LS1 LS2 LS3 LS4 FS1 FS2 FS3 FS4 ITN1 ITN2 ITN3 ITN4 Health1 Health2 Health3 Health4 HProb1 HProb2 HProb3 HProb4;
usevariables = LS1 LS2 LS3 LS4 FS1 FS2 FS3 FS4 HProb1 HProb2 HProb3 HProb4 Sex Age Region Educ;
Missing = all (999.00);
Analysis: TYPE = RANDOM; ALGORITHM = INTEGRATION; ESTIMATOR = ML;