On Testing Monotone Tendencies (Q3313101)

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 3842968
Language Label Description Also known as
default for all languages
No label defined
    English
    On Testing Monotone Tendencies
    scientific article; zbMATH DE number 3842968

      Statements

      0 references
      0 references
      1983
      0 references
      testing monotone tendencies
      0 references
      likelihood ratio tests of homogeneity
      0 references
      decreasing on average
      0 references
      maximum likelihood estimator
      0 references
      order constraints
      0 references
      unbiasedness
      0 references
      normal means
      0 references
      On Testing Monotone Tendencies (English)
      0 references
      The monotone tendencies of normal means subjected to some order restrictions are tested and the maximum likelihood estimators (MLE) of the mean parameter vector and their distributions are investigated. Suppose \(Y_ i\), \(i=1,...,k\), are independent normal random variables with means \(\theta_ i\) and variances \(1/w_ i=a_ i\sigma^ 2, a_ i\) known constants. Let \(W_ i=\sum^{i}_{j=1}w_ j (W_ 0=0)\), \({\bar \theta}{}_ i=W_ i^{-1}\sum^{i}_{j=1}w_ j\theta_ j\), \(\bar y_ i=W_ i^{-1}\sum^{i}_{j=1}w_ jy_ j\), and \(a^+=\max \{a,0\}\), \(a^-=\min \{a,0\}\). The MLE of the vector \(\theta =(\theta_ 1,...,\theta_ k)^ t\) subject to the constraints \(H_ 1:{\bar \theta}_{i-1}\geq(=,\leq){\bar \theta}_ i, i=2,...,k\), is given by \({\hat \theta}_ i=\psi_ i+\sum^{k}_{j=i+1}w_ jW_ j^{- 1}(y_ j-\bar y_{j-1})^{\nu_ j}, i=1,...,k\), with \(y_ i\) unrestricted estimates of \(\theta_ i\) and \((\psi_ i,\nu_ i)=(\min(\bar y_ i,y_ i),+)\), (ȳ\({}_ i,1)\), or (max(ȳ\({}_ i,y_ i),-)\) if the ith constraint in \(H_ 1\) is \(\geq\), \(=\), or \(\leq\), respectively. The null-hypothesis \(H_ 0:\theta_ 1=\theta_ 2=...=\theta_ k\) against alternatives \(H_ 1-H_ 0\) is tested by means of the likelihood ratio statistic \[ Q = 1 - \Lambda^{2/k} = \sum^{k}_{i=1} w_ i(\theta_ i - \bar Y)^2 / \sum^{k}_{i=1} w_ i(Y_ i - \bar Y)^2 \] which has a null-hypothesis distribution given by \[ P(Q\geq t) = \sum^{k-1}_{m=0} \binom{k-1}m \left(\frac12\right)^{k-1} P\left(B_{(k-m-1)/2,m/2} \geq t\right) \] for all \(t\), where \(B_{\alpha,\beta}\) denotes a standard beta random variable with parameters \(\alpha\) and \(\beta\). This test is unbiased. If \(\sigma^2\) is known, an unbiased likelihood ratio test of \(H_0\) against \(H_1- H_0\) can be made by \[ R = -2 \ln \Lambda = \sum^{k}_{i=1} w_ i({\hat\theta}_ i-\bar Y)^2 \] the distribution of which is given by \[ P(R\geq t) = \sum^{k-1}_{m=0} \binom{k-1}{m} \left(\frac{1}{2}\right)^{k-1} P \quad (\chi_ m^2 \geq t) \] where \(\chi^2_ m\) denotes a standard chi-squared random variable with \(m\) degrees of freedom. Also statistics \(Q'\) and \(R'\) for unknown or known \(\sigma^2\), respectively, are given by means of which \(H_1\) as a null-hypothesis can be tested against not \(H_ 1\). For unknown \(\sigma^ 2\) more than one observation \(X_{il},...,X_{in}\) instead of a single \(Y_ i\) are necessary to estimate \(\sigma^ 2\) in a normal population.
      0 references

      Identifiers