Age-period-cohort models for individual-level data: An acceleration-based regression framework

Thesis

I develop a framework to analyse the relationship between an outcome of interest and an individual's age, period of observation, or birth cohort, using individual-level data. The framework is suitable for any continuous or binary outcome. I have created a package for the statistical software R which implements this framework.

It is well-known that linear relationships between age, period, or cohort and an outcome of interest are not separately identified. My framework instead focuses on non-linear relationships, described by age, period, and cohort (APC) accelerations.

My framework embeds the age, period, and cohort (APC) accelerations as parameters in a regression model. The regression approach makes it easy to include covariates and to test restrictions on the model. I develop a regression-based test of the APC acceleration model against a more general model, the time-saturated model.

My APC acceleration framework is suitable for repeated cross section and panel data. For repeated cross section data, a generalized linear modelling strategy is used which accommodates continuous and binary outcomes. For panel data, a generalized least squares strategy is used which accommodates continuous outcomes. I consider three panel settings: pooled ordinary least squares, random effects, and fixed effects.

I give three examples of applying the APC acceleration framework. First, I identify a new stylized fact about the role of birth cohort in obesity among English men. Second, I use the framework as a diagnostic tool to evaluate control variables in a model linking commute time and hospital in-patient stays. Third, I show that the framework can detect well-known relationships between wages and age and period.

Authors: Fannon, Z

• DOI: 10.5287/ora-b7aaaypxe
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: period; time; age; fixed effects; identification; R; econometrics; cohort
• Subjects: statistical software; population economics; health economics; econometrics