CommentaryThe impossibility of separating age, period and cohort effects
Introduction
This commentary addresses this issue of statistically modelling separate age, period and cohort (APC) effects. The issue has been hotly debated for decades (Ryder, 1965), in sociology (Glenn, 1976; Mason, Mason, Winsborough, & Poole, 1973), medical science (Osmond & Gardner, 1989; Robertson & Boyle, 1986) and elsewhere. However, the publication of an article in this issue (Page et al., 2013) alongside more recent methodological developments in APC modelling (Tu, Smith, & Gilthorpe, 2011; Yang & Land, 2006; Yang, Schulhofer-Wohl, Fu, & Land, 2008), shows that there is still profound interest in modelling and discerning APC effects.
This commentary does not directly critique Page et al.'s paper in terms of its substantive conclusions; rather, we address the key methodological issues in modelling APC effects. However the implications of our argument call into question the results found by Page et al., and should act as a warning for others researchers wishing to disentangle APC effects in a meaningful way.
We first outline what APC effects are substantively, and describe the identification problem which makes them so difficult to model. We then outline some proposed solutions to the identification problem, including that used by Page et al., and explain why they will only work in very specific and arguably usually unrealistic circumstances. The commentary finishes with some recommendations for researchers wishing to model APC effects.
Section snippets
Age, period and cohort effects
The difference between age effects, period effects and cohort effects is well explicated by this fictional dialogue by Suzuki (2012, p. 452):
A: I can't seem to shake off this tired feeling. Guess I'm just getting old. [Age effect]
B: Do you think it's stress? Business is down this year, and you've let your fatigue build up. [Period effect]
A: Maybe. What about you?
B: Actually, I'm exhausted too! My body feels really heavy.
A: You're kidding. You're still young. I could work all day long when I was
Solutions to the identification problem, and why they do not work
Despite this scepticism, there have been numerous attempts to find a way around the identification problem. The most common, and that suggested by Mason et al. (1973), is to constrain certain parameters in a model to be equal. So, each age group, period group and cohort group is included in a regression model as a dummy variable, but 2 age groups (or period groups, or cohort groups) are combined together as if they were a single group. This breaks the exact collinearity and allows the model to
Recommendations
So far, there is little good news for the researcher hoping to find separate APC effects. However, we believe that theorising and finding age, period and cohort is often very important in social science, and that they can be modelled, so long as the assumptions that are being made by the model are justified by theory and stated explicitly.
It is often the case that we can assume that continuous period effects are non-existent. It seems to us that theory often indicates that progress over time is
References (16)
Time changes, so do people
Social Science & Medicine
(2012)- et al.
Another ‘futile quest’? A simulation study of Yang and Land's Hierarchical Age–Period–Cohort model
(2013) - et al.
Current practice in the modelling of Age, Period and Cohort effects with panel data: A commentary on Tawfik et al (2012), Clarke et al (2009), and McCulloch (2012)
(2013) Cohort analysts futile quest – statistical attempts to separate age, period and cohort effects
American Sociological Review
(1976)Cohort analysis
(2005)Age, period and cohort effects – a confounded confusion
Journal of Applied Statistics
(1979)- et al.
The cross-classified age–period–cohort model as a constrained estimator
(2013) Cohort variation in the membership of voluntary organisations in Great Britain, 1991–2007
(2012)
Cited by (120)
Has age replaced class in British Elections?A hierarchical age-period-cohort analysis
2023, Electoral StudiesAre generations a useful concept?
2023, Acta PsychologicaImproving health and sustainability: Patterns of red and processed meat consumption across generations
2022, Health PolicyCitation Excerpt :The literature makes several attempts to find a solution to this problem, one of the most common of which is to constrain certain parameters in a model so they are equal. ( See Bell and Jones [41] for details.) Each age and birth cohort group is included in a regression model as a dummy variable, but two age groups and cohort groups are combined into a single group (e.g., [42]).
Healthier over time? Period effects in health among older Europeans in a step-wise approach to identification
2022, Social Science and MedicineThe recession generation? Age-period-cohort dynamics of political trust in six countries severely affected by the 2008 crisis
2024, Frontiers in Political ScienceMeasuring mental health action competencies in school teachers: internal and external validity evidence
2024, Frontiers in Digital Health