Given this benefit of ZIP, we applied latent growth modeling (LGM) with ZIP regressions to model changes in HED among youth over time (i.e., Wave I, minimum age 12 years, through Wave IV, maximum age 31 years). The cohort-sequential design allowed the data to be restructured on age rather than data wave (Bollen & Curran, 2006). Over the four waves of Add Health data, respondents contributed to up to four observations of HED. Because of age overlap, information exists on a possible 20 ages, enabling us to model HED trajectories from ages 12 to 31 years. Although this approach resulted in a substantial amount of missing data, Mplus uses the direct maximum likelihood estimation, which overcomes potential biases due to missing data (Bollen & Curran, 2006).
