Functional principal component analysis in age–period–cohort analysis of body mass index data by gender and ethnicity

ABSTRACT

In this paper, we propose a two-stage functional principal component analysis method in age–period–cohort (APC) analysis. The first stage of the method considers the age–period effect with the fitted values treated as an offset; and the second stage of the method considers the residual age–cohort effect conditional on the already estimated age-period effect. An APC version of the model in functional data analysis provides an improved fit to the data, especially when the data are sparse and irregularly spaced. We demonstrate the effectiveness of the proposed method using body mass index data stratified by gender and ethnicity.

KEYWORDS:

• B-splines
• conditional expectation
• longitudinal data
• mixed-effects model
• random effects

Acknowledgements

We thank the editor and the referees of an earlier version of this paper, who gave helpful suggestions on clarifying and explaining our ideas.

Disclosure statement

No potential conflict of interest was reported by the authors.