R. Alonso Sanz, M. C. Pardo Llorente
We investigate diagnostic measures for assessing the influence of observations on the Cox regression model. A subject may have influence on the regression coefficients, the variance of the estimated coefficients, the fitted values, and/or the goodness-of-fit statistics. Pettitt and Daud (1989) developed Cook’s local measures of influence for the proportional hazards model to measure the impact of an observation on all above regression outputs except on the variance of the estimated coefficients. We proposed a global influence measure based on the volume of confidence ellipsoids for detecting influence of subjects on the efficiency of coefficient estimations in Cox regression. We conduct simulation studies to evaluate our measure and analyze a real data set to illustrate their use.
Palabras clave / Keywords: Cox regression, Influence
Programado
Sesión V05 Bioestadística
1 de junio de 2018 16:00
Sala 4
