Generalized Pearson-Fisher chi-square goodness-of-fit tests, with applications to models with life history data (Q688383)

Suppose that \(X_ 1,\dots,X_ n\) are i.i.d. \(\sim F\), and the authors wish to test the null hypothesis that \(F\) is a member of the parametric family \(F= \{F_ \theta;\;\theta\in\Theta\}\) where \(\Theta\subset R^ q\). The classical Pearson-Fisher chi-square test involves partitioning the real axis into \(k\) cells \(I_ 1,\dots,I_ k\) and forming the chi- square statistics \[ \chi^ 2_{\widehat\theta}= \sum^ u_{i=1} (\nu_ i- nF_{\widehat\theta}- (I_ i))^ 2/nF_{\widehat\theta}(I_ i), \] where \(\nu_ i\) is the number of observations falling into cell \(I_ i\) and \(\widehat\theta\) is the value of \(\theta\) minimizing \(\chi^ 2_{\widehat\theta}\). The authors obtain a generalization of this test to any situation for which there is available a nonparametric estimator \(\widehat F\) of \(F\) for which \(n^{1/2}(\widehat F- F)@>d>> W\), where \(W\) is a continuous, zero-mean Gaussian process satisfying a mild regularity condition. Essentially, they estimate by the value that minimizes a distance between the vectors \((\widehat F(I_ 1),\dots,\widehat F(I_ k))\) and \((F_ \theta(I_ 1),\dots,F_ \theta(I_ k))\), and then form a chi-square type test statistic, based on the difference between \[ (\widehat F(I_ 1),\dots,\widehat F(I_ k))\quad\text{and}\quad (F_{\widehat\theta}(I_ 1),\dots,F_{\widehat\theta} (I_ k)). \] The authors prove that this test statistic has asymptotically a chi-square distribution with \(k-q-1\) degrees of freedom.