Shige Song posted on Friday, January 26, 2007 - 8:51 pm
I am trying to get an estimated growth trajectory out of a model with individually-varying time of observations but had some problems. To simplify the explanation, I will borrow an example from the Mplus manual.
Here is a modified version of Example 6.12 in the Mplus manual: ----------------------------- ... USEVARIABLE = y1-y4 a11-a14; TSCORES = a11-a14; ANALYSIS: TYPE = RANDOM; MODEL: i s | y1-y4 AT a11-a14; PLOT: TYPE = PLOT3; SERIES = y1-y4 (s); -----------------------------
According to the manual (Pp. 535), this should produce estimated means and estimated individual values, among many other type of plots. But I got only three type of plot: histograms, scatterplots, and observed individual values. If I choose individual values, I did not get lines but several points on the Y-axis.
If I understand the question, you want to plot what SPSS proc MIXED calls BLUP, and Stata postestimation (e.g., after XTMIXED) predict command calls "fitted". (Both give individual model based trajectories by including fixed and random effects in the prediction). I have used savedata and FSCORES to get a file with individual scores for the growth factors and the time points. I then compute the prediction for each occasion using the individual tscore and the individual's factor score on the growth parameters. Then it is simply a matter of choosing software that handles spagetti plots well. I like Stata's twoway. Both SPSS and Excel try to make a legend entry for each person and that is a mess. SAS is also useable.
This appears to give the emperical Bayes prediction. (See discussion by SInger & Wilett, p.132). The values I get from MPlus inthis manner are identical to values the equivalent SPSS model saves under then name BLUP.
It is actually more tedious to get the group-averaged trajectories, what Stata predict calls "xb" and SPSS calls BLUE. FOr these you need to apply the fixed effects for each person, the growth factor means, and the TSCORES.