Age–period–cohort modeling

Abstract

Age–period–cohort analysis is an essential epidemiologic tool for analyzing trends over time in health outcomes. Age effects describe the common developmental processes that are associated with particular ages or stages in the life course and can explain trends in health outcomes if the age distribution of the population shifts over time. Period effects describe changes in the prevalence of health outcomes associated with certain calendar years across all age groups. Cohort effects describe changes in the prevalence of an outcome associated with certain age groups in certain years. Thus, cohort effects can be best conceptualized as a population-level interaction between age and period. In this chapter, we (1) review essential concepts in age–period–cohort effect estimation using examples from injury epidemiology; (2) provide examples of historical uses of age–period–cohort analysis; (3) illustrate the statistical problem in simultaneously estimating age, period, and cohort effects; (4) offer an example of a multi-phased method for quantifying cohort effects using data on suicide mortality in the USA; and (5) summarize and describe new directions and innovations in age–period–cohort analysis. The prevalence and incidence of fatal and nonfatal injuries have exhibited substantial trends over time (Martinez-Schnell and Zaidi, Statistics in Medicine 13(8):823–838, 1989). By examining these trends, we can gain insight into the causes of injury at the population level, for example: the effectiveness of public health prevention and intervention efforts for gun control, or the magnitude of change in social norms regarding driving practices, and can forecast the future burden of injury outcomes in the population. Quantitative evaluation trends over time in injury is aided by a comprehensive approach to age–period–cohort analysis, an analytic tool to partition trends into components that are associated with changes over time within a given age structure of the population, time period, and birth cohort. In this chapter, we review essential concepts and definitions in age–period–cohort analysis, provide examples of historical uses of age–period–cohort analysis, illustrate the statistical problem in simultaneously estimating age, period, and cohort effects, offer an example of a multi-phased method for quantifying cohort effects using data on suicide in the USA from 1910 to 2004, and summarize and describe new directions and innovations in age–period–cohort analysis. The prevalence and incidence of fatal and non-fatal injuries have exhibited substantial trends over time (Martinez-Schnell and Zaidi 1989). By examining these trends we can gain insight into the causes of injury at the population level, for example: the effectiveness of public health prevention and intervention efforts for gun control, or the magnitude of change in social norms regarding driving practices, and can forecast the future burden of injury outcomes in the population. Quantitative evaluation of trends over time in injury is aided by a comprehensive approach to age-period-cohort analysis, an analytic tool to partition trends into components that are associated with changes over time within a given age structure of the population, time period, and birth cohort. In this chapter we will review essential concepts and definitions in age-period-cohort analysis, provide examples of historical uses of age-period-cohort analysis, illustrate the statistical problem in simultaneously estimating age, period, and cohort effects, offer an example of a multi-phase method for quantifying cohort effects using data on suicide in the United States from 1910-2004, and summarize and describe new directions and innovations in age-period-cohort analysis.