Multivariate Two-Sample Tests Based on Nearest Neighbors (Q3753309)

From MaRDI portal

!

This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:

scientific article; zbMATH DE number 3990658
Language Label Description Also known as
default for all languages
No label defined
    English
    Multivariate Two-Sample Tests Based on Nearest Neighbors
    scientific article; zbMATH DE number 3990658

      Statements

      0 references
      1986
      0 references
      multivariate two-sample problem
      0 references
      infinite-dimensional approximations
      0 references
      distribution-free test
      0 references
      Euclidean norm
      0 references
      kth nearest neighbor
      0 references
      Asymptotic normality
      0 references
      consistency
      0 references
      optimal weight
      0 references
      simulation study
      0 references
      power
      0 references
      Multivariate Two-Sample Tests Based on Nearest Neighbors (English)
      0 references
      Let \(X_ 1,...,X_{n_ 1}\) and \(Y_ 1,...,Y_{n_ 2}\) be independent random samples in \({\mathbb{R}}^ d\) from unknown absolutely continuous distributions F(x) and G(x) respectively. The problem is to test the null hypothesis H: F(x)\(=G(x)\) against the alternative K: F(x)\(\neq G(x)\). Take \(n=n_ 1+n_ 2\), \(\Omega_ 1=\{1,...,n_ 1\}\), \(\Omega_ 2=\{n_ 1+1,...,n\}\) and label the pooled sample as \(Z_ 1,...,Z_ n\), where \(Z_ i=X_ i\), if \(i\in \Omega_ 1\) or \(Z_ i=Y_{i-n_ 1}\), if \(i\in \Omega_ 2\). Let \(\| \cdot \|\) be the Euclidean norm and define the kth nearest neighbor to \(Z_ i\), denoted by \(NN_ i(k)\), as that point \(Z_ j\) satisfying \(\| Z_{j'}-Z_ i\| <\| Z_ j-Z_ i\|\) for exactly k-1 values of j' (1\(\leq j'\leq n\), j'\(\neq i,j)\). Then the author proposed a test statistic \[ T_{k,n}=(1/nk)\sum^{n}_{i=1}\sum^{k}_{r=1}I_ i(r), \] where \(I_ i(r)=1\) if \(NN_ i(r)\) belongs to the same sample as \(Z_ i\) or \(=0\) otherwise. The larger values of \(T_{k,n}\) are significant. Asymptotic normality of \(T_{k,n}\) under H is established and the consistency of the test is proved in this paper. Weighted version of \(T_{k,n}\) is also considered and the optimal weight is found for specific alternatives. A simulation study shows that the combined test, which robustifies the optimally weighted test by \(T_{k,n}\) has reasonably good power.
      0 references

      Identifiers

      0 references
      0 references
      0 references
      0 references
      0 references
      0 references
      0 references