|
Abstract:
|
The age-period-cohort (APC) models have been applied to demography, economics, public health and sociology. Due to the linear dependence of the age, period and birth cohort, the APC classification model yields multiple estimators, leading to indetermination of parameter estimation and temporal trend. To achieve a unique estimator, an equality constraint has often been used to estimate the parameter and the temporal trend. However, such constraints lead to arbitrary parameter estimation and yield biased estimates almost surely. Given that many APC studies have been analyzed using the constraint method and published without the original data, bias correction is desirable but challenging, particularly after the original data become unavailable. The challenge is how to achieve consistent estimation by the intrinsic estimator with only biased estimates. In this paper, we present a bias correction method to correct the bias based on only the biases estimates without requiring the original data. We demonstrate our method with two data sets, one with original data and one without, and show that both yield the intrinsic estimator and achieve sensible trend estimation.
|
