scientific article; zbMATH DE number 3911472

From MaRDI portal
Publication:3687500





Consider tests testing goodness of fit from a sample of size R about the probability vector \(\Pi =(\pi_ 1,...,\pi_ k)\) in a multinomial distribution. For real \(\lambda\), define the power divergence statistics \[ 2nI^{\lambda}=2\sum^{k}_{i=1}X_ i[(X_ i/E_ i)^{\lambda}- 1]/[\lambda (\lambda +1)] \] where \(X_ i\) and \(E_ i\) are the observed and expected (under the null hypothesis) numbers of observations in the \(i^{th}\) cell. If \(H_ 0\) does not completely specify \(\Pi\), then the \(E_ i\) are taken to be appropriate estimators. The family of the \(2nI^{\lambda}\) includes several well known goodness of fit test statistics. For example, Pearson's \(\chi^ 2\) is \(I^ 1.\) For \(H_ 0:\Pi =f(\theta)\), \({\hat \theta}\) is a BAN estimator of \(\theta\) and the \(E_ i\) are computed from \(\theta ={\hat \theta}\). Under certain regularity conditions, the asymptotic distribution of each 2nI is \(\chi^ 2_{k-s-1}\) where s is the dimensionality of \(\theta\). Under the same conditions, the Pitman efficiency of \(I^{\lambda_ 1}\) relative to \(I^{\lambda_ 2}\) is always 1. Tables are provided for comparison of Bahadur efficiencies.




Cited in
(only showing first 100 items - show all)








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 Q3687500)