On inference assessment in generalized linear models (Q1965958)

A generalized linear model is considered in which the observation is a vector of independent random variables \(y_i\) from an exponential family with means \(E(y_i)=\mu_i.\) The vector \(\mu\) satisfies the relation \(g(\mu)=X\beta\), where \(g\) is a known link function, \(X\) is a known matrix and \(\beta\) is an unknown parameter. The authors propose a diagnostic algorithm to assess the outliers and influential observations in this model. The algorithm uses a diagnostic measure based on cross-validation and a bootstrap procedure to determine significance of the value of this measure for each observation. Results of food-stamp data analysis are considered as an example.