David Brazel posted on Wednesday, December 13, 2017 - 4:07 pm
I have intensive longitudinal data measured on a twin sample. I'd like to fit DSEM models to these data but with a three level structure - measurements within individuals within twin pairs. The documentation doesn't appear to have any examples with a three level structure. Is it possible to fit this type of model in MPlus?
I also have an intensive longitudinal data in which participants (level 3) completed up to 7 assessments (level 1) per day for 14 days (level 2). I'd like to use DSEM models to examine the association between X and Y (both measured at level 1) at three levels: momentary, daily, and between-person level. (1) Is it possible to fit such a model in Mplus 8.0?
(2) Can I use the cross-classified time series analysis(example 9.38) to fit such a model?
I would recommend first to do the two-level model where you have 7*14 times of observations so the cluster variable would be individual with 98 observations in each cluster and using the tinterval command to specify the time of observation. This two-level model can then be extended to a cross-classified model as in example 9.38.
I also have three level data (3 observations per day for 14 days for 117 participants). I would like to use the TINTERVAL command since observation times differed for participants, but with so few observations there will be a high percentage of 'missing' data, which I know is problematic. Should I widen the interval (to every 2 hours instead of 1)? If there are 2 surveys within the 2 hours will one be excluded from the analysis?
Or is there a different method better suited to my data?
1) Use TINTERVAL of 8 hours - that will essentially transform the data so that the 3 observations will be assumed to have been taken 8 hours apart. Observations are not deleted but are spaced out according to the TINTERVAL value. You can then try 4 hours and 2 hours but as you say the amount of missing data may prohibit the estimation.
2) Ignore the timing within day and instead use a factor model F by Y1-Y3@1 (&1); for the three within day observations (similar to averaging the 3 observations) See User's guide examples 6.27 and 9.34.
Thank you for the quick response! I'm trying to model within-day changes, specifically how variables at time 1/2 (4 measures of emotion regulation) affect a variable at time 2/3 (an affect measure), so I'm not sure if the factor model would be appropriate.
For TINTERVAL, how would I set up the time variable in my data set? I have the date, time of day, day in the study, survey number, and wave number (1,2,3). Would I have to compute each participant's continuous time spent in the study?
I tried running it with tinterval, but the PSR's are not very good. I removed one of the random variables, but the PSR's were still not great. I'm not sure how else to improve the model. Do I keep removing random variables?
I have also tried pursuing the three-level model, but am not fully sure that my code is right. Here is what I have:
MODEL: %WITHIN% NA2 ON NA1; NA3 ON NA2;
A|NA2 ON SERQb1; B|NA3 ON SERQb2;
C|NA2 ON SERQc1; D|NA3 ON SERQc2;
E|NA2 ON SERQi1; F|NA3 ON SERQi2;
G|NA2 ON SERQm1; H|NA3 ON SERQm2;
%BETWEEN% NA1 WITH NA2 NA3 SERQb1 SERQb2 SERQc1 SERQc2 SERQi1 SERQi2 SERQm1 SERQm2; NA2 WITH NA3 SERQb1 SERQb2 SERQc1 SERQc2 SERQi1 SERQi2 SERQm1 SERQm2; NA3 WITH SERQb1 SERQb2 SERQc1 SERQc2 SERQi1 SERQi2 SERQm1 SERQm2; SERQb1 WITH SERQb2 SERQc1-SERQc2 SERQi1-SERQi2 SERQm1-SERQm2; SERQb2 WITH SERQc1-SERQc2 SERQi1-SERQi2 SERQm1-SERQm2; SERQc1 WITH SERQc2 SERQi1 SERQi2 SERQm1 SERQm2; SERQc2 WITH SERQi1 SERQi2 SERQm1 SERQm2; SERQi1 WITH SERQi2 SERQm1 SERQm2; SERQi2 WITH SERQm1 SERQm2; SERQm1 WITH SERQm2;