Bootstrap tests in correspondence analysis
For a two-way contingency table \(\pi_{ij}\) (\(i\)=1,\dots,\(I\); \(j\)=1,\dots,\(J\)), the canonical representation \[ \pi_{ij}=\pi_{i+}\pi_{+j}(1+\sum_{m=1}^M\lambda_m x_{im}y_{jm}) \] is considered, where \(M=\min(I,J)-1\), \(\lambda_1\geq\lambda_2\geq\dots\geq\lambda_M\). It is proposed to test the hypothesis \(C_k\;:\;\lambda_{k+1}=\lambda_{k+2}= \dots=\lambda_M=0\) with the help of the least squares estimators for \(\lambda'\)s, \(x'\)s and \(y'\)s, and the bootstrap procedure to estimate the significance of \(\lambda_j\). Different test statistics are discussed. The author proposes the test statistics \(G_k=l_{k+1}^2/(\sum_{m>k} l_m^2)\), where \(l_k\) are the LS estimators of \(\lambda_k\). Results of Monte-Carlo simulations and an example of real data analysis (mental health and parietal socioeconomic status correlation) are presented.
- scientific article; zbMATH DE number 1138583 (Why is no real title available?)
- Bootstrap confidence regions for correspondence analysis
- Bootstrapping analogs of the one way MANOVA test
- Which Bootstrap for Principal Axes Methods?
- Tests for redundancy of some variables in correspondence analysis
- Bootstrap tests for independence in two-way ordinal contingency tables
This page was built for publication: Bootstrap tests in correspondence analysis
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4258928)
