A Distribution-Free Rank Test for Ordered Alternatives in Randomized Complete Block Designs (Q3333900)

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 3866393
Language Label Description Also known as
default for all languages
No label defined
    English
    A Distribution-Free Rank Test for Ordered Alternatives in Randomized Complete Block Designs
    scientific article; zbMATH DE number 3866393

      Statements

      0 references
      0 references
      1984
      0 references
      Monte Carlo study
      0 references
      Jonckheere test
      0 references
      Page's L
      0 references
      Hollander test
      0 references
      joint symmetry
      0 references
      distribution-free
      0 references
      ordered alternatives
      0 references
      strict inequality
      0 references
      signed rank test
      0 references
      randomized complete blocks
      0 references
      orthogonal contrasts
      0 references
      small sample distribution
      0 references
      Wilcoxon two-sample signed rank test
      0 references
      consistency
      0 references
      asymptotic normality
      0 references
      A Distribution-Free Rank Test for Ordered Alternatives in Randomized Complete Block Designs (English)
      0 references
      Let \(X_{ij}=\mu +\beta_ i+\tau_ j+\epsilon_{ij}\), \(i=1,2,...,n\), \(j=1,2,...,k\), where \(\beta_ i's\) and \(\tau_ j's\) are block and treatment effects, respectively, subject to restrictions \(\sum \beta_ i=\sum \tau_ j=0\). \(\epsilon_{ij}'s\) are independent and identically distributed random variables having an absolutely continuous symmetric distribution. The objective is to test the null hypothesis \(H_ 0:\tau_ 1=\tau_ 2=...=\tau_ k\) against the ordered alternatives \(H_ 1:\tau_ 1\leq \tau_ 2\leq...\leq \tau_ k\) with at least one strict inequality. The authors have suggested a signed rank test for randomized complete blocks with \(k=3\) and \(k=4\). To obtain the test statistic, in each block, k-1 orthogonal contrasts of the observations are defined; i.e., for \(k=3\), \(Y_{i1}=X_{i2}-X_{i1}\) and \(Y_{i2}=2X_{i3}-X_{i1}- X_{i2}\); and for \(k=4\), \(Y_{i1}=X_{i3}-X_{i1}\), \(Y_{i2}=X_{i4}-X_{i2}\), and \(Y_{i3}=X_{i4}-X_{i3}+X_{i2}- X_{i1}\). Let \(R_{ij}=rank| Y_{ij}|\) among \((| Y_{11}|,...,| Y_{n,k-1}|)\), then the test statistic is \(W=\sum^{n}_{i=1}\sum^{k-1}_{j=1}sign(Y_{ij})R^ t_{ij}.\) The authors have shown that under \(H_ 0\) the small sample distribution of W is the same as that of the Wilcoxon two-sample signed rank test. Furthermore, necessary and sufficient conditions are given for consistency of the test. In order to provide the user of this method with a choice of ''asymptotically best'' (in Pitman sense) tests, the authors make asymptotic relative efficiency comparisons with some parametric and nonparametric procedures. The advantages of this method are: ease of computation, availability of tables of critical values for small sample test, and asymptotic normality of the test for large samples, as is appropriately pointed out by the authors. The main disadvantage is that it can be used for randomized complete block designs with at most 4 treatments and one observation per cell.
      0 references

      Identifiers

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