Jen posted on Tuesday, December 15, 2015 - 9:46 am
I have taken a look at Web Note 16, which was very helpful, but just want to check on a particular scenario.
We are analyzing two types of events, and hoping to use a multigroup model to compare identical structural models related to each event type.
Participants in our study could report on one or both event types. Participants thus have either 1 or 2 observations each total. Within each 'group' (event type), no participant has more than one observation.
We are interested in controlling for a few demographic variables that are at the person level (e.g., sex, age, etc.), but otherwise don't care about the 'between' level in this model: we are just interested whether theoretical variables relate to one another differently for event type 1 vs. 2. These theoretical variables and the outcomes were all measured with respect to the specific event type.
1) Can I just use a multigroup model with TYPE = COMPLEX and CLUSTER = subject? Mplus does not give me any error when I do this.
2) Alternatively, is it required that I use KNOWNCLASS with TYPE = TWOLEVEL MIXTURE? This also seems to work fine but is less ideal because of fewer fit indices and more complex syntax.
If required, we could randomly select just one event per person, but it seems a shame to throw out good data...
Web Note 16 is looking at groups like male/female within a cluster. I don't see event type as such a male/female division. Clusters need to have independent samples, so no person can be in two different clusters.
I wonder if your situation should instead be viewed more in line with a wide setting like two time-points. Some people are observed at only one time point but others are observed at both and that correlation due to the same person needs to be modeled, e.g. by a correlation of variables or factors across time. So not multiple group or multilevel.