Quasi-Likelihood Regression with Multiple Indices and Smooth Link and Variance Functions

From MaRDI portal





The authors propose a MUSE (multiple-index semi-parametric) regression model with responses \(y_i\) and predictors \(x_i\) where \[ {\mathbf E} y_i=E(y_i)=\sum_{k=1}^M g_k(x_i^T\beta_k), \quad \text{Var}(y_i)=\sigma^2(E(y_i)), \] where \(E\) is a known link function, \(g_k\) and \(\sigma\) are unknown smooth functions, and \(\beta_k\) are unknown parameter vectors. Identifiability conditions are derived. Quasi-likelihood estimates for \(g_k\), \(\sigma\) and \(\beta_k\) are constructed. A backfitting algorithm is proposed for the computation of the estimates. Bandwidth selection is discussed. Asymptotic behavior of the estimates is investigated. Applications to some biostatistical data are considered.











This page was built for publication: Quasi-Likelihood Regression with Multiple Indices and Smooth Link and Variance Functions

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4677101)