Age, Period and Cohort Effects

What it is

The framework rests on a piece of arithmetic. A person's birth year is the survey year minus their age, so any two of the three quantities fix the third. In one survey, taken in one year, period is constant and birth year is age read backwards. A gap between a 22-year-old and a 55-year-old is then at once a gap of age and a gap of cohort, and the survey cannot say which of the two, or of a period influence acting on both, produced it.

Repeated measurement is what separates them. Bobby Duffy sets out the method in his book Generations (Atlantic Books, 2021): plot each birth group's answers across many years, and read the lines. A movement that appears in every group in the same year is a period effect. A difference that follows one birth group as it ages is a cohort effect. A pattern that repeats in each group at the same age is an age effect. Duffy states the three-way separation as the core of his method and applies it across money, housing, work, health and attitudes.

Two designs do part of the work cheaply. A time-lag design compares people of the same age at different dates, which holds age fixed. The Twenge and colleagues study of American high school seniors in 1976, 1991 and 2006, published in the Journal of Management in 2010, is built this way. Holding age fixed leaves cohort and period tangled, and the authors acknowledge that cross-sectional studies cannot separate the effects and that period remains a possible reading of their differences. The second design follows the same cohort over time, which shows how that group changes with age but cannot say how a different group would have changed.

In effect

Duffy's own example is life satisfaction in Spain. In his series, recession dips appear in every generation at the same moments, which marks them as period effects, and no cohort gap persisted over time. The one exception he finds is that each new youth cohort enters the data happier than the last. Read from a single year's snapshot, the same data would have looked like a difference between the old and the young (p. 95).

The evidence on work attitudes gives the framework a larger test. David Costanza and colleagues pooled twenty studies and nearly twenty thousand workers in a 2012 meta-analysis and found the differences between Traditionals, Baby Boomers, Generation X and Millennials in job satisfaction, organisational commitment and intent to leave to be modest to negligible, with many effects near zero. A 2015 review by Costanza and Lisa Finkelstein reported little solid empirical evidence for generationally based differences at work and little theory to explain why they should exist.

Period effects show up inside cohort studies too. Kali Trzesniewski and Brent Donnellan, analysing about 477,000 American school seniors from 1976 to 2006, found small changes in most traits and reported that later seniors were more cynical and less trusting, which the authors noted mirrored the wider population. A change shared with older people at the same moment points to the period, not to the birth group.

The Flynn effect is the standing example of a cohort effect that is accepted. James Flynn reported large gains in the mean IQ of Americans between 1932 and 1978, with each later-born group scoring higher than the one before. Because the comparison is between groups at the same age on tests, it has the structure that a cohort claim needs, and it is rarely read as a trait of a generation.

The article Are Generational Differences Real draws these together into three questions for any claim about a generation. Would it survive being restated about people aged 24, whatever year they were born? Would everyone alive in that year show it? And was a group of the same age asked the same question in an earlier decade, and did it answer differently?

What it does not say

It does not say that cohort effects are absent. The Flynn effect and the leisure result in the Twenge study are cohort differences, measured at the same age in different decades. It says only that a cohort effect has to be shown by a design that can show it.

It does not say the three can always be separated cleanly. The arithmetic means that a model has to assume something, for example that one of the three is zero or that two move together, and different assumptions can give different answers. A time-lag design fixes age and then relies on the argument that period is not doing the work, which its authors state openly.

It does not say what causes a cohort difference. A birth group differing from another says nothing by itself about whether the cause is schooling, prices, technology or something else. Duffy argues that the cohort gaps that matter most are economic, in incomes and housing, and that argument is his, resting on survey and official series this page has not rechecked.

And it does not make generation labels meaningless in every use. The point of the framework is narrower: a label should be believed at the size a repeated measurement gives it, and no larger.

---