A1509
Title: A kernel-based test for the proportional hazards assumption
Authors: Tamara Fernandez - Universidad Adolfo Ibanez (Chile) [presenting]
Nicolas Rivera - Universidad de Valparaiso (Chile)
Merle Munko - Otto-von-Guericke University Magdeburg (Germany)
Abstract: A novel kernel-based test is presented for assessing the proportional hazards assumption. The proposed test evaluates whether deviations from the Cox model, captured through functions in a reproducing kernel Hilbert space (RKHS), lead to a better approximation of the underlying hazard. Unlike standard kernel tests, our approach faces an additional challenge: it requires estimating the Cox model parameters under the null hypothesis as part of the testing procedure. This estimation step implies that conventional resampling strategies, such as Wild Bootstrap, are no longer valid. We outline the derivation of the test statistic and introduce a corrected wild bootstrap method that accounts for parameter estimation and provides valid inference.
