COHORT ANALYSIS

A cohort is a set of individual items (usually persons) that have in common the fact that they all experienced a given event during a given time interval. For example, the "U.S. marriage cohort 1995–1999" consists of all persons who got married in the United States in the period from 1995 to 1999. In demography birth cohorts are of particular importance and frequently are referred to simply as cohorts (e.g., "the 1960 cohort" or "cohort 1960," indicating all persons born in 1960).

Cohort analysis is the study of dated events as they occur from the time of the event that initiated the cohort. For example, one can analyze the first births of marriage cohort 1995–1999 or the mortality of birth cohort 1960 (and compare this with the mortality of, say, birth cohort 1930). Cohort analysis often is contrasted with period analysis, the study of events occurring in multiple cohorts at a particular historical time, such as during a specified calendar year.

Applications

There are two main applications of cohort analysis. The first could be termed cohort analysis in its own right: the study of how behavior develops over the life course, with the initiating event (e.g., marriage) serving as a key explanatory factor, marking the start of the exposure to risk of the dependent event of interest (e.g., marital fertility). Since the 1980s powerful statistical techniques have been available that allow, using micro-level data, a much more detailed study of how behavior develops over the life course: Besides the event initiating the cohort, a wide range of additional explanatory variables (including time-varying ones) can easily be included.

The second main application is to study temporal variation at the level of the aggregate population through changes in life course behavior over successive cohorts. For example, research seeking to explain the baby boom of the 1960s may focus on the fertility level of the cohorts that were of reproductive age during the 1960s. The underlying idea is that aggregate demographic events cannot be properly understood without paying attention to the conditioning life course situation of the individual members of the population. It is in this sense that the term cohort analysis is especially well known in demography, in particular because of the pioneering work of the demographer Norman B. Ryder in the 1950s and 1960s.

Ryder stressed the crucial importance of the flow of successive cohorts into the population (which he termed demographic metabolism) for adapting modern society to changed external conditions. Cohorts differ because they have experienced certain key historical events (e.g., economic conditions, the introduction of the contraceptive pill) at different and sometimes critical ages. History determines a cohort's destiny. Because of this, it is important to differentiate by cohort when one is studying aggregate behavior. An example in demography is the presence of cohort effects in mortality (e.g., Barker 1994): Experiencing a famine or war at younger ages has a permanent impact on survival for the cohorts that are involved.

Demographic Translation

Ryder was also concerned with the relationship between time series of fertility measures on a period basis and those on a cohort basis. Fertility is most commonly measured by demographers in terms of the schedule of age-specific fertility rates (ASFRs) and derived summary statistics, notably the total fertility rate (TFR)–the sum of the ASFRs–and the mean age at childbearing (MAC). ASFRs can be arranged in a Lexis surface, with the period (calendar year) on the horizontal axis and age on the vertical axis. Each ASFR belongs to a period (vertical section) and a cohort (diagonal section). As a consequence, summary statistics such as the TFR can be calculated in two ways: on a period basis, summing ASFRs vertically, and on a cohort basis, summing ASFRs diagonally. If the level (quantum) and timing (tempo) of fertility are constant over time, period and cohort indicators are exactly equal. However, if level and/or timing are not constant, period and cohort indicators are not identical. For example, if sub-sequent cohorts have their children at increasingly higher ages (fertility postponement; that is, MAC rises over time), the annual number of births is depressed and the period TFR becomes smaller than the cohort TFR for all the cohorts involved.

Ryder investigated the mathematical relationships between such period and cohort time series of fertility indicators, establishing what is now known as demographic translation theory. A famous translation formula is TFR period = TFRcohort /(1 + annual change in MAC cohort), linking period and cohort TFRs under the conditions of a constant quantum of cohort fertility but with the cohort tempo shifting linearly over time. Using this formula, one can calculate the drop in period fertility that results from a postponement of childbearing that does not alter ultimate family size.

Some researchers believe that such translation formulas can be used to estimate cohort fertility from period fertility. The inherent problem in calculating cohort fertility indicators is that one has to wait until the cohort has finished childbearing: For cohorts still of reproductive age, one observes only part of their fertility career (i.e., up until now). It is tempting to try to use the full period information to make statements about these cohorts' future fertility. Unfortunately, such attempts are hazardous. Any procedure used in an attempt to estimate cohort quantum from period quantum is based on simplifying assumptions, the justifiability of which can only be verified empirically: by comparing the estimated cohort fertility with the actual cohort fertility. But if actual cohort fertility is known, the translation procedure is no longer needed.

Hypothetical Cohort

Age-specific indicators of demographic behavior, such as fertility rates and mortality rates, that are all measured during a single period refer to different cohorts. Nevertheless, one can ask what would happen to a cohort if over its lifetime it were to behave according to the age-specific indicators observed during this particular period. For example, it is possible to calculate the average life span of a fictitious group of persons surviving according to the age-specific mortality rates observed in the United States during the year 2002. Such calculations on period data are then interpreted as if they applied to a cohort. Such a cohort is known as a hypothetical cohort or synthetic cohort. Hypothetical cohorts can be very useful analytically but should never be confused with true cohorts, which experience age-specific rates that are typically not independent from one year to the next.

Period versus Cohort?

The work of Ryder and others has initiated a heated and unresolved debate between followers of the cohort approach and adherents of the period approach. In their extreme forms these two approaches as they are applied to fertility can be described as follows:

Cohort approach: Each cohort, shaped by the historical conditions under which it reaches reproductive age, follows its own fertility career. Year-by-year changes in fertility are caused by new cohorts replacing old cohorts in the reproductive age span. Period fertility measures are just the average of the underlying cohort fertility measures.

Period approach: Aggregate fertility is driven by current conditions. If conditions change, period fertility changes also. Cohorts shape their fertility career as they go through time. Cohort fertility measures are just the average of the underlying period fertility measures.

As is always the case with extreme positions, the truth lies in between. The extreme cohort position ignores the fact that cohorts (in fact, individual persons) do not start their reproductive career with cast-iron fertility targets but instead modify their fertility behavior as period conditions change. The extreme period position ignores the fact that family formation is a lifetime enterprise, and as a consequence, period effects affect cohorts differently, depending on the life course position the cohorts currently hold and the fertility choices they have made. For example, a period effect such as the introduction of reliable contraceptives will have a much larger effect on the fertility of cohorts currently 20 years old than on the fertility of cohorts currently 40 years old.

Indeed, a birth cohort is not only a set of individuals born during the same period in the past but also a set of individuals each of whom experiences a period effect at the same stage of the life course (current year = birth year + age). This double significance of the cohort concept alone should make it clear that both the period perspective and the cohort perspective are needed to understand aggregate fertility or any other type of demographic behavior.

Period measures of fertility indicate how many children are born each year and, consequently, how the age structure of the population changes over time. Cohort measures of fertility indicate the extent to which individual members of the population reproduce themselves. Although both sets of measures are taken from the same Lexis surface and therefore refer to the same babies and mothers, the exact relationship between period and cohort measures depends on so many factors (notably, shifts over time in the age pattern of fertility) that it is sensible to treat them as two fundamentally different concepts of the quantum of fertility.

COHORT ANALYSIS

Applications

Demographic Translation

Hypothetical Cohort

Period versus Cohort?

BIBLIOGRAPHY