Backfitting Random Varying-Coefficient Models with Time-Dependent Smoothing Covariates

From MaRDI portal





A mixed-effects type linear regression model is proposed for longitudinal data. In this randomly varying coefficients (RVC) model \[ y_{ij}=\beta_{0i}(t_{ij})+x_i(t_{ij})\beta_{1i}(t_{ij}) +e_i(t_{ij}), \] where \(t_{ij}\) is the time of the \(j\)-th observation for the \(i\)-th object, \(y_{ij}\) is the response, \(x_i(t_{ij})\) is the value of the covariate, \(e_i(t)\) is an independent measurement noise process, \(\beta_{ki}(t)=\beta_k(t)+\gamma_{ki}(t)\), \(k=0,1\), where \(\beta_k(t)\) are some fixed population time-varying parameters and \(\gamma_{ki}\) are independent realizations of a mean zero stochastic process with some fixed covariance structure. A backfitting algorithm is described for parameter estimation. Asymptotic expansions for bias and variance of the estimate \(\hat\beta_1\) for \(\beta_1\) are derived. A bandwidth selection technique for \(\hat\beta_1\) based on cross-validation is considered. Results of simulations and applications to AIDS clinical study data are presented.




Cited in
(28)








This page was built for publication: Backfitting Random Varying-Coefficient Models with Time-Dependent Smoothing Covariates

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4828223)