Longitudinal Influences of Neighborhood Socioeconomic Environment on Weight Change Among Normal Weight 18-Year-Olds: An Application of Missing Data Methods in Path Analysis

• Activity Number: 519
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Social Statistics Section
• Author(s): Jin-Wen Yang Hsu*+ and Deborah Rohm Young and Guangyu Zhang
• Companies: Kaiser Permanente and Kaiser Permanente and NCHS/CDC
• Keywords: longitudinal data ; missing data ; multiple imputation ; full information maximum likelihood ; path analysis

Abstract

It is common to have missing data in longitudinal studies. We compared multiple imputation (MI) and full information maximum likelihood (FIML) methods to handle missing data in a study of the longitudinal influences of neighborhood socioeconomic environment on weight change. Path analysis was used to test a model for longitudinal body mass index (BMI). We selected 10,433 healthy, normal weight 18-year olds in 2008 and followed them over a four year period through 2012. Sex, race/ethnicity, height, and weight were recorded in electronic medical records. Neighborhood education and income were determined from US 2000 Census block group data. Of the cohort, only 3649 (35%) had complete BMI records during the 4-year follow-up. In addition to MI and FIML, analysis of data from complete cases analysis (CC) was also performed for the purpose of comparison. The results showed that the neighborhood education had a significant effect on the change of BMI, but the influence of neighborhood income was not statistically significant. Results of MI, FIML and CC methods were compared in terms of parameter estimates, standard errors and goodness of fit indices.