Abstract
The Bayesian cohort model was introduced by Takashi Nakamura in 1982. His model succeeded in overcoming the identification problem in cohort analysis by setting up an assumption that successive parameters change gradually. Subsequently, he incorporated age-by-period interaction effects into his original model. This paper presents the Bayesian cohort models with different types of interaction effects, said to be difficult to realize, such as age-by-cohort, cohort-by-age, period-by-cohort and cohort-by-period as well as period-by-age. These new models are applied to analysis of liquor consumption in Japan. In addition, this paper shows interactions with other factors are also among candidates for components of appropriate models, taking up a life-stage cohort model as an example. The life-stage cohort analysis proposed in this paper is to make a study of household cohorts classified according to the birth year of the eldest child, regarded as aging as the eldest child grows older. The study reveals that interaction effects between age of the eldest child and number of children living together are essential to analyze a share of seafood in food consumption.
