scientific article; zbMATH DE number 4068093

From MaRDI portal
Publication:3800916





The paper deals with a linear regression model of the form \[ (1)\quad Y=b_ 0+x'b+s e \] with a response variable Y, a \(p\times 1\) vector of covariates x, an intercept \(b_ 0\), a vector of regression coefficients \(b'=(b_ 1,...,b_ p)\), \(s>0\), and an error e with a fully specified density f(e) on \(-\infty <e<\infty\). The inference is supposed to be based on the assumed model (1). However, the true distribution of Y given x is \[ (2)\quad Y=m_ 0+x'm+t w \] where \(t>0\) and w has a density g(w) on \(-\infty <w<\infty\). The maximum likelihood estimators assuming model (1) are denoted by \(\hat b_ 0\), \(\hat b\) and \(\hat s,\) respectively. It is shown that under quite general conditions on f and g, \(\hat b\) is consistent for m but \(b_ 0\) and \(\hat s\) are not in general consistent for \(m_ 0\) and t. The asymptotic efficiency of \(\hat b\) relative to that of the maximum likelihood estimator \(\hat m\) under model (2) for known g is calculated. Further the variance estimation of \(\hat b\) is discussed.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3800916)