I am trying to test a simple model that gets complicated very quickly by the fact that my data is clustered and has repeated measures (7 time points). I am working with version 3.0 of mPlus. I have six observed variables that I believe to load on a single latent variable, and want to test if this is indeed the case. I have collected these 6 variables at 133 different facilities across the country quarterly for two years. The facilities cluster regionally, as they are managed by 23 regional offices (each facility can only belong to one region).
I was able to successfully test for fit using only one time point, (code printed below). Now what I want to do is be able to test the LV1 BY OV1-OV6 model but using all 7 quarters of data in addition to regional clustering (I assume that the scores on these observed variables change over time, however, I want to test that the model does not).
So my question is how do I account for both time and region in the same model?
Sylvia J. Hysong
ORIGINAL, SINGLE QUARTER CODE FOLLOWS: VARIABLE: NAMES ARE region ov1-ov6 facno; USEVARIABLES ARE region ov1-ov6; CLUSTER IS region;
ANALYSIS: TYPE IS COMPLEX; ESTIMATOR IS MLR; ITERATIONS = 1000; CONVERGENCE = 0.00005;
One way to deal with it is shown in Example 9.12. Or you could use Example 6.1 and add the CLUSTER option to the VARIABLE command and TYPE=COMPLEX; to the ANALYSIS command. I would also suggest upgrading to the most recent version of Mplus. There have been many changes in the nine years since Version 3 came out.
Thank you for your feedback. Sorry for the delay in a response, this project got put on hold for a while and is just now starting up again. I don't know if I made this sufficiently clear in my original post or not, but my interest is not to model change over time in these data. My interest is just to test the LV1 BY ov1-ov6 model; it's just that I happen to have clustered and autocorrelated data because I have 7 quarters of it. So all I want to do is calculate accurate estimates for the measurement model. Does that make sense? How do I do that?
Sylvia J. Hysong
P.S. My data is currently in "tall and skinny format", that is multiple rows of data for each case (i.e., variables = facility, region, quarter, OVnum, score). To do the analysis correctly, do I need it in "short and fat" format, that is, one row of data for each case (variables = facility, region, ov1q1, ov1q2... ov6q3, ov6q4)?
Sounds like what you want to do I would call longitudinal factor analysis. And you have clustered data so could use either Type = Complex or Type = Twolevel - you probably want to use Type = Complex for simplicity since you have only 23 clusters. You would arrange your data in line with ex 9.12 in the V4 UG so that you have the 6*7 outcomes as columns for a facility (plus the other variables, including the cluster the facility is in). The model would be like
f1 by ov11-ov16; f2 by ov21-ov26; ... f7 by ov71-ov76;
where you can test various degrees of across-time equality constraints of measurement loadings and intercepts (see Chapter 16 of the UG for such equality testing.
Dana Wood posted on Sunday, February 22, 2009 - 8:02 pm
I want to estimate a latent growth model for three time points. My study participants are nested within schools; however, all participants change school over time. I was planning to use Type = Complex and the cluster option. Is there any way to specify that participants are nested in different clusters at each time point?