HGLM modelling of dropout process using a frailty model (Q2488401)

The variable of interest \(y_{it}\) is observed for cases \(i=1,\dots,n\), at time points \(0=a_1<a_2<\dots<a_t<\dots\) up to the time \(T_i\) of the failure (dropout) for the \(i\)-th case (i.e., while \(a_t<T_i\)). The model assumed for the data is a linear Gaussian regression for \(y_{it}\) and Cox's proportional hazard rate model for \(T_i\) with shared random effects, i.e., \(y_{it}=x'_{2,it}\beta_2+\delta v_i+\varepsilon_{it}\), and the failure rate for \(T_i\) is \[ \lambda(t\,|\, v_i)=\lambda_0(t)\exp(x_1\beta_1+v_i). \] Here \(x_1\), \(x_2\) are covariates, \(\beta_1\), \(\beta_2\) are unknown parameters and \(v_i\) is the shared random effect, \(\delta\) is a fixed known number. \(T_i\) is not observed, but an interval \([a_t,a_{t+1}]\ni T_i\) is known. The authors consider a hierarchical maximum likelihood approach to the parameter estimation in this model. The obtained estimates are analysed via simulations.