199935 Comparison of three age-period-cohort models of obesity prevalence in the United States

Monday, November 9, 2009: 4:30 PM

Katherine Keyes, MPH , Columbia University/New York State Psychiatric Institute, New York, NY
Rebecca L. Utz, PhD , Sociology, University of Utah, Salt Lake City, UT
Whitney R. Robinson, PhD , Department of Epidemiology, University of Michigan, Ann Arbor, MI
Guohua Li , Departments of Anesthesiology and Epidemiology, Columbia University, New York, NY
Background: There are divergent views on what constitutes a cohort effect and how it should be quantified and interpreted. Some age-period-cohort methods conceptualize age and period as potential confounders of the cohort effect, whereas others conceptualize cohort effects as a special form of age and period interaction. The purpose of the present study was to compare the results of three distinct age-period-cohort methods on obesity trends in the United States from 1971-2006.

Methods: Data were drawn from seven cross-sectional waves of the National Health and Nutrition Examination Surveys (NHANES). Obesity was defined as BMI≥30 for adults and ≥95th percentile for children. First-order effects of age, period, and cohort were estimated using a traditional constraint-based approach to identification. Second-order effects were estimated by calculating linear contrasts, and additionally by conducting median polish analysis.

Results: All three methods produced similar estimates of age and period effects, with a lower prevalence in early life and increasing prevalence across the survey years. A positive cohort effect for more recently born cohorts emerged based on the constraint-based model, but when cohort effects are considered a second-order estimate, no significant cohort effects were detected in the prevalence of obesity.

Conclusion: First-order estimates of age-period-cohort effects suffer from statistical implausibility due to the reliance on constraints, whereas second-order estimates are both conceptually meaningful and statistically estimable. Age-period-cohort analysts should explicitly state the conceptual definition of cohort under consideration, as well as match the analytic technique to the conceptual relationship among age, period, and cohort.

Learning Objectives:
To compare three different analytic techinques to estimate age-period-cohort effects.

Keywords: Obesity, Statistics

Presenting author's disclosure statement:

Qualified on the content I am responsible for because: I,along with the coauthors analyzed the data. I received a PhD in Sociology (demography) from the University of Michigan in 2004, where I first learned about age-period-cohort modeling.
Any relevant financial relationships? No

I agree to comply with the American Public Health Association Conflict of Interest and Commercial Support Guidelines, and to disclose to the participants any off-label or experimental uses of a commercial product or service discussed in my presentation.