Ryan Grimm posted on Thursday, May 10, 2012 - 12:03 am
Hi Drs. Muthen and Muthen,
I want to model growth between K and 2nd grade in an intervention study. All students were assessed in the fall of each year, but only intervention students were assessed in the spring. Thus, I have missing by design for 3 of my 6 time points, where the missingness has a known missing pattern (e.g., students higher than a certain score have a missing value). How best would I model growth over time with data like this that accounts for this design?
If the cutoff is made based on the observed Fall score which is available for everyone, then missingness is MAR and ML will be fine where you include all available data, scoring non-assessed students' scores as missing. So your growth model pertains to everyone.
Ryan Grimm posted on Monday, May 14, 2012 - 10:56 pm
Okay, great. Thank you. By design then, 2/3 of respondents are missing spring test scores. Can I just use one growth factor to model for all people, or should I add an additional growth factor for those who received treatment? I want to compare slopes for those who received treatment and those who did not. Should I just include treatment status as a covariate?
I have a question in relation to missing data in two-part latent growth curve models. In one of my conditional model I get this warning: "NO CONVERGENCE IN THE MISSING DATA ESTIMATION OF THE SAMPLE STATISTICS.". I decreased the coverage and increased the numb of iterations but did not solve the problem.
I guess this is partially due to the fact that individuals who do not report any behavior are treated as missing for the frequency part of the model. So, my question is: because this missing are assumed to be at random, and moreover I have MCAR, can I ignore this message? Thanks for your help!