Duality in testing multivariate hypotheses (Q3800910)

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 4068086
Language Label Description Also known as
default for all languages
No label defined
    English
    Duality in testing multivariate hypotheses
    scientific article; zbMATH DE number 4068086

      Statements

      Duality in testing multivariate hypotheses (English)
      0 references
      0 references
      0 references
      1988
      0 references
      convex cone
      0 references
      one-sided test
      0 references
      polar cone
      0 references
      linear equality constraints testing problem
      0 references
      likelihood ratio test
      0 references
      duality result
      0 references
      Let b be a \(k\times 1\) vector and \(\hat b=b+h\) where h is N(0,V) [i.e. normally distributed with mean 0 and covariance matrix V] and V a known \(k\times k\) positive-definite matrix. Let R be a \(p\times k\) matrix of full rank where \(p\leq k\) and r a \(p\times 1\) vector. Let further \(Q(x,M)=2^{-1}x'M^{-1}x\), for \(x\in R^ n,\) with M a positive- definite \(n\times n\) matrix. Then the linear equality constraints testing problem in primal form is H: Rb\(=r\) versus K: Rb\(\neq r\). H can be tested by a likelihood ratio test. The corresponding likelihood ratio statistic is defined by a simple quadratic program. In the paper a general duality result for this optimization problem is proven (theorem 1), which leads to Corollary 1: The test H: Rb\(\geq r\) versus \(b\in R^ k\) based on \(\hat b,\) which is N(b,V), is equivalent to H: \(\lambda=0\) versus K: \(\lambda\geq 0\) based on \[ \bar L=(RVR')^{-1}(r-R\hat b), \] which is \[ N(l,(RVR')^{-1}),\quad for\quad \lambda =(RVR')^{-1}(r-Rb). \]
      0 references

      Identifiers