Period Effects

Individual human subjects and sometimes other entities, such as cities or nations, can be regarded as existing in two time dimensions, subjective and objective. The subjective time dimension for an individual begins at birth, and parallels to birth can be found for other entities. The objective time dimension is measured with respect to events external to the individual or entity, dating from the beginning of the universe or, at a more modest level, the beginning of a research study. We call the subjective time dimension age and the objective time dimension history. Entities whose subjective time begins at the same objective time are said to belong to the same cohort.

In LONGITUDINAL RESEARCH, patterns of change in variables and patterns of relationships among variables are studied with respect to both age and history. The impact of age on variables or relationships among variables, ignoring historical time, is called a developmental effect or age effect. Examples of age effects include language acquisition and development of other cognitive skills. The impact of living at a certain historical time on variables or relationships among variables, ignoring age, is called a period effect or historical effect. Examples of period effects include the use of computers, cell phones, and other devices, which depend on the level of technology in the historical period rather than on the age of the individual in that period, and attitudes and behaviors shaped by political and military events, such as expressing more patriotic attitudes or displaying the American flag more frequently in the United States after the bombing of Pearl Harbor or the destruction of the World Trade Center. The impact of being a certain age at a certain period (in other words, of being born in a certain year) is called a cohort effect. One of the principal problems in longitudinal research is separating the impacts of age, period, and cohort (Menard, 2002). The problem arises because if we use age, period, and cohort as predictors and assume a linear relationship of each with the dependent variable, the predictors are perfectly collinear, related by the equation, period = age + cohort, when cohort is measured as year of birth.

Different approaches to the separation of period effects from age and cohort effects are reviewed in Menard (2002). As noted by Hobcraft, Menken, and Preston (1982), period effects provide a proxy—often a weak proxy—for an explanation that is not really a matter of “what year is it” but rather a matter of what events occurred in that year, or before that year, at some significant period in the past. As Menard (2002) suggests, to the extent that it is possible, period effects should be replaced in longitudinal analysis by the historical events or trends for which they are proxies. At the aggregate level, it may also be possible to separate period effects from age or cohort effects by examining age-specific indicators of the variables of interest. This is commonly done in social indicators research, for example, by using trends in infant mortality or rates of labor force participation for individuals 16 to 64 years old as indicators of trends in national health and employment conditions.

...