Abstract:
|
The linearly dependent age, period and birth cohort (APC) are well known confounders and their fixed effects have been believed deeply for decades non-estimable due to the parameter identification problem in the APC multiple classification model. Yet the importance of modeling simultaneous age, period and cohort effects has been emphasized repeatedly in the literature. Although this difficult parameter identification problem seems to be intractable, recent works provide solid evidence to address the problem, making it more promising than ever before to resolve this long term unsettled controversy. In this paper, I will first review the identification problem, then highlight some recent works, including the intrinsic estimator (Fu 2000, Yang et al 2004, 2008, Fu 2015) and the smoothing cohort estimator (Fu 2008) and their properties, and point out why these novel approaches are promising. I will further explain what have been missed in previous work. Finally, I will present most recent work in hypothesis testing on the age, period and cohort effects between populations for statistical inference and illustrate with data in public health and economic studies.
|