Abstract:
|
For at least 80 years researchers in a wide variety of fields have sought to uniquely identify age, period, and cohort (APC) effects, even though an infinite number of solutions exist due to perfect linear dependency. In this paper we introduce a new approach for partially identifying APC effects based on bounding feasible regions of the parameter space. Depending on the location of the solution line in the parameter space, relatively minimal constraints on the direction and magnitude of the linear trends can lead to substantively meaningful conclusions. Furthermore, bounds can be derived from mechanism-based models that specify the processes by which one or more of the linear components influence the outcome of interest, even when such models are misspecified. We present several empirical examples to illustrate our approach. We conclude with a discussion of Bayesian interpretations of bounding analyses as well as guidelines for further research on APC effects.
|