Age, birth cohorts, and drinking: an illustration of the hazards of inferring effects from cohort data.

The ambiguity of many sets of cohort data is illustrated by cohort data on drinking alcoholic beverages from American national surveys conducted during the late 1950s, late 1960s, and late 1970s. Although the data approximate the pattern predicted by pure linear cohort effects, examination of "side information" reveals that they are likely to reflect positive period effects offset from one date to the next by negative age effects. Researchers are cautioned that all available "side information" should be considered before statistical cohort models are tested. Some researchers have tried to use tests of statistical models to answer questions which must be answered before appropriate statistical tests can be applied.