Problems Fitting Latent Growth Curve Model for BMI

Hi all,

I have a question regarding panel data--collected biennially--from the National Longitudinal Study of Youth. (I use STATA 13.0 for Mac.)

I've been trying--and failing--to fit a latent growth curve model for respondent Body Mass Index (BMI). Respondents were empaneled as children and adolescents throughout the 1980s and 90s, and were repeatedly interviewed into young adulthood. (Latest round of interviews were 2012.)

My problem is model fit. A completely unconditional model with no time-varying or time-invariant covariates fits reasonably well (but not great). When I enter ANY covariates, the model falls apart. I've tried using a latent quadratic or cubic term, but the model doesn't fit much better, at least by Acock's (2013) standards. (I should mention that maximum likelihood and ML with missing values are both problematic.)

Eyeballing the raw data and based on the literature, I can see why the model fit isn't very good: BMIs increase rapidly during childhood, then tend to level off and fluctuate (even decrease) later in life.

So my question is: Is it possible to model these BMI data using an LGM? Thanks!