Generalized martingale-residual processes for goodness-of-fit inference in Cox's type regression models (Q1359430)

Let \({\mathbf N}= (N_1,\dots, N_n)\) \((n\geq 1)\) be a multivariate counting process. Let \(\beta_0\) be a column vector of \(p\) unknown regression coefficients, \(\lambda_0\) be an arbitrary and unspecified baseline hazard function, \(Y_i\) be a predictable \(\{0,1\}\)-valued process indicating that the \(i\) th individual is at risk when \(Y_i=1\) and \(Z_i\) be a \(p\)-variate column vector of processes which are assumed to be predictable and locally bounded. Let \(w\) be some weight function. Define the general version of Cox regression of the form \[ \lambda_i(t)= Y_i(t)\exp [\beta_0^T Z_i(t)]\lambda_0(t), \quad t\in[0,\tau], \] and consider the maximum weighted partial likelihood estimator \(\widehat{\beta}_w\) of \(\beta_0\). Let \[ \widehat{M}_i^w(t)= \int_0 tw(Z_i(s),s) dN_i(s)- \int_0^t w(Z_i(s),s) Y_i(s)\exp [\widehat{\beta}_w^T Z_i(s)] d\widehat{\Lambda}_w(s), \] where \(\widehat{\Lambda}_w(s)\) is a weighted version of the standard estimator for the cumulative baseline hazard function. Further, define \[ \widehat{\mathbf M}_i^w(t)= \int_0^t \varphi_i(s) d\widehat{M}_i^w(s)\quad (i=1,\dots,n) \] for some functions \(\varphi_i\). Let \[ \psi(\widehat{\beta}_w,\cdot)= n^{-1/2} L(\widehat{\beta}_w,s)^T d\sum_{i=1}^n {\mathbf M}_i^w(s) \] where the process \(L(\beta_0,\cdot)\) is assumed to be a linear combination. The authors show that the process and its modified version converge weakly to some Gaussian processes. Among others, as an application, the formal construction of the Kolmogorov-Smirnov and Cramér-von Mises-type goodness-of-fit tests is considered.