I'm interested in learning more about DSEM. I'm currently working on a response burst design. In my non-DSEM way of thinking, I think that level 1 is moment, level 2 is wave/burst, and level 3 is person. I have hypotheses about level, variance, and autocorrelation of my level 1 variable as a function of wave, but no variables relating to person.
1. Can DSEM handle a three-level model? 2. Is a three level model even necessary if there are no predictor variables at level 3? 3. Could you point me to a sample script?
1. Currently only if one of the levels has rather few units, such as the example in Asparouhov's presentation at the end of Part 8 of the Hopkins workshop video and handout (see his handout Part 6, slides 48-51). But given your description of level 1 and 2, I am not sure you need 3 levels - see below comments.
2. You need a person level even if there are no person predictors as long as you have variation across persons.
Perhaps you can describe your "moments" and "bursts". If moments are a series of measurements close in time and bursts are different periods in time where those moments occur, perhaps a single level is sufficient for moments and burst - unless it is expected that different bursts have different moment parameter values.
We're interested in emotional variability in people in a particular type of long-term treatment. We were going to measure emotion at each moment and model the effect of burst. We were going to track people at baseline and every six months for two years.
So, if I understand correctly, I could make a two level model where level 2 is the person-level and a predictor at level 1 would be which burst we were using?
Hello, I'm wondering if anything has changed with respect to DSEM allowing for 3-level models since this last post? I'm hoping to analyse a similar "measurement burst" design, in which N = 150 participants were assessed at T = 56 occasions (8 times per day for 7 days) over W = 3 waves of measurement (6 months apart). Would you still recommend using a 2-level DSEM model with "wave" included as a within-level moderator? Or is it possible to run a 3-level model with Occasions (T) nested within Waves (W) nested within Participants (N)? Many thanks, Pete Koval
I think the most reasonable modeling approach with the current implementation is to arrange the data and construct a model as follows variable: names = Y1 Y2 Y3; where Y1 represents the data of the first burst, Y2 represents the data of the second burst and Y3 in the third. On the between level I would use something like that f by y1-y3@1; y1-y3 (1); (assuming you don't expect a change in the average between the burst - if you do then a growth model at least on the fixed parameters could be appropriate).
On the within level the time series models for Yi should be independent of each other and possibly with the parameters held equal across the three bursts.