Message/Author |
|
|
Dear professors Muthén, I am trying to run a Monte Carlo simulation for a three-level model, with repeated measurements at Level-1, patients at Level-2, and therapists as Level-3. I would like to be able to specify different cluster sizes both for Level-2 (i.e. patients attending different numbers of sessions) and for Level-3 (therapists treating different numbers of patients). Is this possible in Mplus? I haven't been able to find any examples of such analyses? Best wishes, Fredrik Falkenström |
|
|
See pages 783-784 of the user's guide. The NCSIZES and CSIZES options for TYPE=THREELEVEL are described there. See also the Monte Carlo counterparts of the THREELEVEL examples in the user's guide. |
|
|
Dear Linda, thank you very much. A quick follow-up, is there a limit for how many cluster sizes that can be specified? It seemed to me that I got problems when I got over 60, but it may be that I had done something wrong. Best, Fredrik |
|
|
Please send the output and your license number to supoprt@statmodel.com. |
|
|
Thank you so much for all your help! Do you know if there is any way of using the Monte Carlo simulation procedure to test for bias and power in comparing two models, e.g. using the Likelihood Ratio test or just the difference in AIC/BIC? Best, Fredrik |
|
|
You can use the RESULTS= option of the MONTECARLO command to save AIC/BIC/log-likelihood across the different replications. You can also use the same Model Population to generate the data in two different runs with two different Model commands. Once you get the results for the two models for all the replications you can manipulate them, ex. compare BIC etc, using excel spreadsheet or another program. |
|
|
Thanks! Fredrik |
|
|
Hi again, I am trying to estimate a misspecified model in which one covariate (treatment length) is used in the Model Population part but not in the estimation model. However, my model doesn’t converge when I do this, and I am thinking I may have done something wrong. My input is too long to paste here, but what I did was just to omitting this covariate from the estimation model. Is there some other way of doing this? Best, Fredrik Falkenström |
|
|
Are you fixing its coefficient at zero? |
|
|
I had not done that since I thought leaving it out would have the same effect, but now I tested to explicitly fix its coefficients at zero and that didn't seem to help. Fredrik |
|
|
Please send to support. |
|
Back to top |