Age-Period-Cohort Analysis

Prof. dr. Mohammad Hakimi, SpOG(K), PhD.

 Cohort Analysis
• Cohort analysis, which is a general
strategy for examining data rather
than a specific statistical technique,
has become increasingly popular in
the social sciences in the past few
decades as an abundance of data
appropriate for its application has
become available.
• Its popularity results partly from its
usefulness in addressing substantive
issues relating to aging and social and
cultural change, but many
researchers apparently have become
interested in cohort analysis primarily
because it presents an unusually
intriguing methodological challenge.
 Purpose of Cohort Analysis
• Cohort analysis has been used primarily for two
distinct purposes, although both have motivated
some studies.
– The more common of these is to assess the
effects of aging, usually of human beings but
sometimes of entities such as marriages.
– The second major use of cohort analysis is to
understand the sources and nature of social,
cultural, and political change.
 Definitions
• The term cohort originally referred to a group of
warriors or soldiers, and the term is now sometimes
used in a very general sense to refer to a number of
individuals who have some characteristic in common.
• Here and in other literature on cohort analysis,
however, the term is used in a more restricted sense to
refer to those individuals (human or otherwise) who
experienced a particular event during a specified
period of time. The kind of cohort most often studied
by social scientists is the human birth cohort, that is,
those persons born during a given year, decade, or
other period of time.
 Definitions
• The event commonly experienced by the individuals is
used as an adjective to identify the kind of cohort, as in
retirement cohort and graduation cohort, and if the
individuals are not human beings, another adjective may
be needed to identify them.
• The events that define cohorts may range from marriage
to joining an organization, from entering a graduate
program to becoming a parent for the first time.
• The individuals are usually human beings but may be
marriages, organizations, textbooks, movies, automobile
models, or other entities that came into being during a
specified period of time.
 Cohort Analysis
• The term cohort analysis is usually reserved
for studies in which two or more cohorts are
compared with regard to at least one
dependent variable measured at two or more
points in time.
 Cohort Analysis
• Examples of studies that would not ordinarily
be considered cohort analyses, even though
they involve cohorts in some way, are
– (a) a comparison of the political attitudes of young
adults, middle-aged persons, and elderly persons
with data from one survey conducted in
November 2002, and
– (b) a comparison of the attitudes of all members
of a high school graduating class of 1985 in that
year and in 1995.
 Cohort Analysis
• The former study is cross-sectional, that is, it is
conducted with data collected at one point in time, or,
more accurately, within a short period of time.
• The latter is a panel study, that is, it measures the
characteristics of the same individuals at more than
one point in time.
• Both kinds of studies can be very useful for some
purposes, but they have limited value for attainment of
the most common goal of cohort analysis, namely,
assessment of the consequences of growing older, or,
in other words, estimation of age effects.
 Cohort Analysis
• Differences by age shown by cross-sectional data may or
may not be age effects, because the people (or other
entities) of different ages are members of different cohorts
and may have been shaped by different formative
experiences and influences.
• In other words, differences among them may be cohort
effects.
• To illustrate, persons born in the United States in 1920
spent their late adolescence and early adulthood in the
Great Depression, whereas persons born just 10 years later
spent the same stages of life in a period of unprecedented
prosperity and economic growth. This contrast is believed
to have created lifelong differences between the cohorts in
economic and political attitudes and behavior.
 Cohort Analysis
• Panel studies can provide evidence of compositional effects
due to differential mortality, and they also have the
advantage of being able to deal with individual-level
change as well as aggregate-level change in aging cohorts.
• However, they alone cannot provide convincing evidence of
age effects, because not all changes that occur in
individuals as they grow older are the results of aging.
• Especially in modern societies, people grow older not in a
static society but in a changing one, and influences from
social and cultural change impinge on persons as they grow
older, bringing about changes in attitudes, behavior, health,
and emotional states or offsetting effects that would result
from aging in a static society.
 Cohort Analysis
• These period effects are confounded with age effects
in the data from panel studies.
• To illustrate, in the United States, the birth cohorts that
were in young adulthood in the 1970s had, as a whole,
become more conservative in several respects by the
late 1980s.
• One cannot tell how much, if any, of this shift resulted
from influences associated with aging; the fact that the
society as a whole changed in the same direction as
the cohorts suggests that much, if not most, of the
intra-cohort trend was brought about by period
influences.
A: Sepertinya aku tidak bisa melepaskan perasaan lelah ini.
Kurasa aku telah menjadi tua. [Age effect]
B: Apa menurutmu itu stres? Bisnis turun tahun ini, dan kamu
telah membiarkan kelelahanmu menumpuk. [Period effect]
A: Mungkin. Bagaimana dengan kamu?
B: Sebenarnya aku juga lelah! Tubuhku terasa sangat berat.
A: Kamu bercanda. Kamu masih muda. Aku bisa bekerja
sepanjang hari ketika seusiamu.
B: Oh, benarkah?
A: Ya, orang muda akhir-akhir ini dengan cepat mengeluh.
Kami tidak seperti itu. [Cohort effect]
Bell A & Jones K (2016) Age, period and cohort processes in longitudinal and
life course analysis: a multilevel perspective. In: Burton-Jeangros C, Cullati S,
Sacker A & Blane D (eds). A Life Course Perspective on Health Trajectories and
Transitions. Springer International Publishing, pp. 197-213.
 Age Period Cohort Effect
• Age period cohort (APC) analysis plays an
important role in understanding time-varying
elements in epidemiology.
• In particular, APC analysis discerns three types
of time varying phenomena: Age effects,
period effects and cohort effects.
 Age effects
• Age effects are variations linked to biological and
social processes of aging specific to individuals.
• They include physiologic changes and
accumulation of social experiences linked to
aging, but unrelated to the time period or birth
cohort to which an individual belongs.
• In epidemiological studies age effects are usually
denoted by varying rates of diseases across age
groups.
 Age Effects
Age Effects
Patterns of disease frequency which
are due to the aging of the
population.

Example: Rates of coronary artery

disease by age.
 Period effects
• Period effects result from external factors that
equally affect all age groups at a particular
calendar time.
• It could arise from a range of environmental,
social and economic factors e.g. war, famine,
economic crisis.
• Methodological changes in outcome definitions,
classifications, or method of data collection could
also lead to period effects in data.
 Period Effects

Patterns of disease frequency which are

due to the period of time in which the
people live.
Example: Elevated death rates from
pneumonia and influenza in the winter.
 Cohort effects
• Cohort effects are variations resulting from the
unique experience/exposure of a group of
subjects (cohort) as they move across time.
• The most commonly defined group in
epidemiology is the birth cohort based on year of
birth and it is described as difference in the risk
of a health outcome based on birth year.
• Thus a cohort effect occurs when distributions of
disease arise from an exposure affect age groups
differently.
 Cohort Effects
Patterns of disease frequency due to an exposure
occurring to a group of people at about the same time
in their lives (and who are therefore of the same age,
and, who, as they age, represent a cohort).

Examples:
• Birth cohort - all people born in 1900.
• Veteran cohort - all men (of similar ages) who served
in the military together.
• Employment cohort - all individuals (of similar ages)
hired by a company at about the same time.
 Cohort effects
• In epidemiology, a cohort effect is
conceptualized as an interaction or effect
modification due to a period effect that is
differentially experienced through age-specific
exposure or susceptibility to that event or
cause.
 Cohort effects
• In contrast to this conceptualization of cohort effect as
an effect modification in epidemiology, sociological
literature consider cohort effect as a structural factor
representing the sum of all unique exposures
experienced by the cohort from birth.
• In this case, age and period effect are conceived as
confounders of cohort effect and APC analysis aims to
disentangle the independent effect of age, period and
cohort.
• Most of the APC analysis strategies are based on the
sociological model of cohort effect, conceptualize
independent effect of age, period and cohort effect.
 Detecting Cohort Effects
• Cohort effects are a special interaction of
age and period.
• The three effects are interrelated. Once two
are specified the third is determined by the
other two.
• Cohort effects are often hard to detect
because analysis is usually by age and
period.
• Cohort effects are usually seen as an
elevated risk which changes with age over
time.
 Diseases With Cohort Effects
Many important disease time-trends can
better be understood by cohort analysis.
This includes:
• peptic ulcer
• tuberculosis
• congenital rubella prostate cancer
(possibly)
 Identification problem in APC
• APC analysis aims at describing and estimating
the independent effect of age, period and cohort
on the health outcome under study.
• The different strategies used aims to partition
variance into the unique components attributable
to age, period, and cohort effects.
• However, there is a major impediment to
independently estimating age, period, and cohort
effects by modeling the data which is know as the
“identification problem” in APC.
 Identification problem in APC
• This is due to the exact linear dependency among
age, period, and cohort: Period – Age = Cohort;
that is, given the calendar year and age, one can
determine the cohort (birth year).
• The presence of perfectly collinear predictors
(age, period and cohort) in a regression model
will produce a singular non-identifiable design
matrix, from which it is statistically impossible to
estimate unique estimates for the three effects.
Sexual Quality of Life and Aging:
Disentangling age, period, and
cohort effects
Miriam K. Forbes1, Nicholas R. Eaton2, & Robert F. Krueger3
1 Departments of Psychiatry and Psychology,
University of Minnesota. Supported by National
Institute of Drug Abuse training grant: T320A037183.
2 Department of Psychology, Stony Brook University.
3 Department of Psychology, University of Minnesota
 Background
A fulfilling sex life is important for wellbeing in adulthood1, but sexual quality
of life (SQoL) has a negative relationship with age2.
Research that aims to understand aging needs to consider three types of
effects:
1. Age effects (changes that affect all cohorts as they age,
independent of time period);
2. Period effects (changes at a particular time that affect all ages and
cohorts uniformly; and
3. Cohort effects (differences that characterize a generation that are
independent of the process of aging)3.
These effects are intertwined and confound the interpretation of apparent
age-related changes, particularly in cross-sectional research4. Research to
date has not separated these effects in the relationship between sexual
quality of life and aging.
 Aim
• To disentangle age, period, and cohort effects
to tell how much of the change in SQoL over
time is due to aging, and how much is due to
the particular period in history being analysed,
or the specific cohort-related time and
circumstances in which an individual aged4.
 Method
• Sample: We used longitudinal data (n = 6369)
collected across three periods (1995, 2004, and
2013) from the nationally representative Midlife in
the United States (MIDUS) study
• Participants were born between 1920 and 1974,
and ranged in age from 21 to 93.
• SQoL: Participants rated the sexual aspect of
their life these days from ‘the worst possible
situation’ (0) to ‘the best possible situation’ (10).
 Figure 1 illustrates the trends in the observed means of
SQoL (0-10), which include a decline across all age groups
(Panel 1), and possible period effects (Panel 2).
Table 1 shows the best linear mixed-effects model of the
age, period, and cohort effects, where a ten year increase in
age was associated with a 4.4% decrease in SQoL.
Figure 2 highlights the age by period moderation effect from
Table 1: Older adults had a steeper decline in SQoL between
the periods of assessment, compared to younger adults.
 Discussion
• The age effect was the strongest of the three
time-related effects in our models: Older
adults had lower SQoL, and more rapid
decline in SQoL over time.
 Discussion
• There were also evident period effects, as SQoL
declined over the 18 years of the study independent of
the effects of age. One explanation for this result could
be the proliferation of internet pornography since the
mid-90s. During this time there have also been
population-level changes in sexual behavior,
expectations about sexual interactions, and sexual
satisfaction that mirror the effects of viewing
pornography5,6. However, ultimately the period effect
was not significant after accounting for by the age by
period moderation.
 Discussion
• The cohort effect did not add to the predictive
power of the models for SQoL. This implies that
there were no important differences in SQoL
characterised by the historical birth cohorts
specifically. Any differences that were present
were captured by the age and period effects,
which means that generational differences were
either linearly related to age, or were masked by
societal shifts that affected all generations
equally5.
 Conclusions
• The results are consistent with research that
has found a negative relationship between
aging and SQoL.
• These findings extend our understanding of
this relationship, and highlight the importance
of future research to understand why SQoL
declines with age.
 References
1 Woloski‐Wruble, A. C., Oliel, Y., Leefsma, M., & Hochner‐Celnikier, D. (2010). Sexual
activities, sexual and life satisfaction, and successful aging in women. The Journal of
Sexual Medicine, 7(7), 2401-2410.
2 Fleeson, W. (2004). The quality of American life at the end of the century. How
Healthy Are We? p. 252-272.
3 Blanchard, R. D., Bunker, J. B., & Wachs, M. (1977). Distinguishing aging, period and
cohort effects in longitudinal studies of elderly populations. Socio-Economic Planning
Sciences, 11(3), 137-146.
4 Yang, Y., & Land, K. C. (2013). Age-period-cohort analysis: New models, methods, and
empirical applications. CRC Press.
5 Field, N., Mercer, C. H., Sonnenberg, P., Tanton, C., Clifton, S., Mitchell, K. R., ... &
Jones, K. G. (2013). Associations between health and sexual lifestyles in Britain:
Findings from the third National Survey of Sexual Attitudes and Lifestyles (Natsal-
3). The Lancet, 382(9907), 1830-1844.
6 Hald, G. M., Kuyper, L., Adam, P. C., & Wit, J. B. (2013). Does viewing explain doing?
Assessing the association between sexually explicit materials use and sexual behaviors
in a large sample of Dutch adolescents and young adults. The Journal of Sexual
Medicine, 10(12), 2986-2995.
Rosenberg PS, Check DP, Anderson WF. A Web Tool for
Age-Period-Cohort Analysis of Cancer Incidence and
Mortality Rates. Cancer Epidemiology, Biomarkers
Prevent 2014:23:2296

• Age–period–cohort (APC) analysis can inform

registry-based studies of cancer incidence and
mortality, but concerns about statistical
identifiability and interpretability, as well as the
learning curves of statistical software packages,
have limited its uptake.
• APC methodology can now be carried out
through a freely available user-friendly Web tool.
The tool can be accessed at
http://analysistools.nci.nih.gov/apc/.
Class Exercise
 Deaf-Mutes in New South Wales,
Australia
Age 1911 1921 1933
0-4 16 17 11
5-9 59 72 95
10-14 111 86 89
15-19 64 57 141
20-24 65 117 98
25-29 60 59 69
30-34 54 67 140
35-39 57 62 71

Do you see the birth cohort effect in these data?

The data show a peak in deaf-mutes among 10-14
years olds in 1911, among 20-24 year olds in 1921,
and among 30-34 year olds in 1933. This is clearly
the same cohort that is aging.

Later research showed that a rubella epidemic took

place in New South Wales at the same time that this
cohort was in utero. The effects of congenital rubella
infection in causing sensori-neural hearing loss were
not discovered until decades later, but these data
were early evidence of the effects of the epidemic.