In example 6.18 of the User's Guide (a cohort sequential model), variances of outcome scores observed at a given age are constrained to be equal across cohorts. E.g., variance of the outcome score observed at age 12 for cohort 1 is constrained to equal variance of the outcome score at age 12 for cohort 2, even if subjects in the 2 cohorts were 12 years old at different times. Given this, why were the corresponding model estimated means not also constrained to equality? E.g., shouldn't the model estimate of the mean outcome score at age 12 for cohort 1 be assumed to equal that of cohort 2, even though the 2 cohorts were 12 years old at different times?
Hi, I'm trying to run a cohort-sequential design based on age (11-16 years at baseline, 4 waves), for boys and girls separately (so two independent models with multi-group based on age). Ultimately, I would like to make parallel process models with online and offline behaviors using a cohort-sequential design, but I started my attempts using just one process.
Both the boys and the girls sample have approximately 600-700 participants, but the N of different age groups varies considerably, e.g. in the boys sample N = 77 for 11yr but N = 248 for 14yr etc.
In addition, I'm studying behaviors that are quite uncommon in some age groups and more uncommon for girls. For that reason I specified MLR as estimator.
I've started building my model using example 6.18, but it doesn't work yet. Freely estimating the processes for each age group seems to work and gives an adequate fit, but as soon as I'm trying to estimate one growth processes by constraining intercept and slope across cohorts, my model fits decrease considerably OR the model does not converge because the maximum number of iterations has been reached. I've tried to increase that number, but that does not help.
What could be the problem here? And what would be an alternative procedure? Constrain the overlapping time points between the cohorts instead of estimating one intercept and slope?