Cumulative chi-squared statistic as a tool for testing goodness of fit

From MaRDI portal





This paper deals at first with the one-way classification model \(Y_{ij}\sim N(\mu_ i,\sigma^ 2)\), \(i=1,\dots,a\), \(j=1,\dots,r\), \(Y_{ij}\) stochastically independent. The hypothesis to be tested is \(\mu_ 1=\mu_ 2=\dots=\mu_ a\) against \(\mu_ 1\leq\dots\leq \mu_ a\) or its two-sided alternative. The test-statistic is \(X^{*2}=r\| P_ a^{*'}\bar Y\|^ 2\), \(\bar Y=(\bar Y_ 1,\dots,\bar Y_ a)'\) and \(P^*_ a\) a suitable matrix. By using Chebyshev polynomials it is shown that \(X^{*2}\) is a weighted sum of independent \(X^ 2\)-variables with one degree of freedom. \(X^{*2}/a\) can be approximated by \(\sigma^ 2dX^ 2_{\nu}\), where \(\nu\) is not necessarily an integer. The second problem discussed in this paper is the model \(Y_{ij}\sim N(\mu +\alpha_ i+\beta_ j+\gamma_{i+j-1},\sigma^ 2)\), \(Y_{ij}\) stochastically independent. The hypothesis to be tested is \(\gamma_ 1- \gamma_ 2=\gamma_ 2-\gamma_ 3=\dots=\gamma_{a+r-2}-\gamma_{a+r- 1}\) against ordered alternatives. The test-statistic \(X^{+2}\) similar to \(X^{*2}\) is a weighted sum of independent \(X^ 2\)-variables with one degree of freedom and can again be approximated by a constant times a \(X^ 2\)-variable.




Cited in
(24)








This page was built for publication: Cumulative chi-squared statistic as a tool for testing goodness of fit

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3730863)