Parameter Constraints and Impact on Parameter Estimation in ANOVA and Age-Period-Cohort Models

• Activity Number: 533
• Type: Topic Contributed
• Date/Time: Wednesday, August 3, 2016 : 10:30 AM to 12:20 PM
• Sponsor: Social Statistics Section
• Author(s): Kuikui Gao* and Shujiao Huang and Wenjiang Fu
• Companies: University of Houston and University of Houston and University of Houston
• Keywords: Centralization ; Robust estimation ; Trend estimation ; Unbalanced data ; Variance

Abstract

Age-period-cohort (APC) models have been studied extensively in social science and public health. It models data on an axp table of a rows of age groups and p columns of periods. It is a special case of two-way ANOVA with diagonal effects as special age by period interaction. In general, the parameters in one-way or two-way ANOVA models need constraint, either by parameter centralization or by setting a reference level. In this paper, we examine the impact of such constraints. We first exam the variance of parameter estimates and how it varies with the constraint. We then further introduce a sensitivity analysis to study the estimator's robustness in data analysis. We conduct simulation studies and demonstrate the results with real data in cancer research. We conclude that constraint on the parameters plays a crucial role in data analysis and inference, except for balanced data. For the APC model with the number of observations varies with the diagonal, a robust parameter estimation is desirable and caution should be used in selecting the parameter constraint.