On the Problem of Interactions in the Analysis of Variance

From MaRDI portal





The model considered is the two-way balanced classification \[ y_{ij}=\mu_{ij}+\nu_{ij}+\varepsilon_{ij};\;\mu_{ij}=\mu+\alpha_ i+\beta_ j,\;i=1,\dots,I,\;j=1,\dots,J, \] \(\varepsilon_{ij}\) independent and \(N(0,\sigma^ 2)\). The problem to be discussed is the construction of a simultaneous set estimate of the form \((y_{ij}-d,y_{ij}+d)\) for the \(\mu_{ij}\). The method \(R_ 0\) of pretesting \(\nu_{ij}=0\) and then possibly changing the model to \(\nu_{ij}=0\) is criticized. Besides a simultaneous confidence interval for the \(\mu_{ij}\), also a confidence region for the largest of the \(\mu_{ij}\) is constructed (Problem B). By constructing probabilistic bounds on \(\nu_{ij}\) a second kind of confidence region for both problems is obtained. Asymptotic and finite sample results are proved and discussed for both procedures. The paper itself is followed by three comments of which two (Shafer and Naiman) are on the line with the author while the paper by Casella and Hwang is controversial. A rejonder of the author forms the end of the paper. The reader can be recommended to read first the remark by Naiman. This comment discusses the ideas of the paper by considering the simpler one-way classification model.











This page was built for publication: On the Problem of Interactions in the Analysis of Variance

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3981138)