Selection of mixed copula for association modeling with tied observations

The link between Obesity and Hypertension is among the most popular topics which have been explored in medical research in recent decades. However, it is challenging to establish the relationship comprehensively and accurately because the distribution of BMI and blood pressure is usually fat tailed and severely tied. In this paper, we propose a data-driven copulas selection approach via penalized likelihood which can deal with tied data by interval censoring estimation. Minimax Concave Penalty is involved to perform the unbiased selection of mixed copula model for its convergence property to get un-penalized solution. Interval censoring and maximizing pseudo-likelihood, inspired from survival analysis, is introduced by considering ranks as intervals with upper and lower limits. This paper describes the model and corresponding iterative algorithm. Simulations to compare the proposed approach versus existing methods in different scenarios are presented. Additionally, the proposed method is also applied to the association modeling on the China Health and Nutrition Survey (CHNS) data. Both numerical studies and real data analysis reveal good performance of the proposed method.