Cohort's Analysis in eCommerce

Abstract

Cohort analysis treats an outcome variable as a function of cohort membership, age, and period. The linear dependency of the three temporal dimensions always creates an identification problem. Resolution of this problem requires external knowledge that is often difficult to acquire. Most satisfactory is the introduction of variables held to measure the dimensions that underlie at least one of age, period and cohort. Such measured, substantive variables can provide direct tests of cohort-based explanations. A Promising path for future technical development is a hierarchical Bayes approach, which treats appropriately defined cohort, age, and period contrasts as randomly distributed and allows for their dependence on substantive, measured variables. Models that include age, period, and cohort can also include interactions between these dimensions, but not all such interactions are identified. This extends the realism of cohort models, since many phenomena seem to require specifications that allow for interactions between two or more of age, period, and cohort. Panel studies and cross-sectional studies with retrospective information not only support cohort analyses, they engender them. These longitudinal data structures do not, however, provide the basis for a solution to the identification problem.[5]