統計学輪講 第4回

日時 2021年4月27日(火)
15時45分 ~ 16時35分
場所 Zoomオンライン開催(URLはITC-LMSをご確認ください)
講演者 武石 将大 (経済D1)
演題 サブグループが存在するCoxモデルの推定に関する幾つかの漸近的結果
概要

In survival analysis literature, a Cox model that incorporates heterogeneity of covariate effect has been proposed in [1] and [2]. In this so-called change-plane Cox mode, the population is split into the main group and a subgroup with enhanced covariates effect. Then, parameter of interests are the one expressing covariates effect (regression parameter) and the one that concerns classification of the aforementioned two groups (classification parameter). Because the model possesses discontinuity, theoretical analysis is intractable, and thus, several asymptotic properties remains an unsolved problem. Against this background, this research proposes a new smoothed estimator and provide some asymptotic guarantee of the estimator. Notably, we derive asymptotic normality of regression parameter and shows that the convergence rate of classification parameter can be arbitrarily close to 1/n up to a log(n) approximation error.

[1]Pons, O. (2003). Estimation in a Cox regression model with a change-point according to a threshold in a covariate. Ann. Statist. 31, 442-63.
[2]Wei, S. and Kosorok, M. (2018). The change-plane Cox model. Biometrika. 105(4), 891-903.