Relative performance of stein-rule and preliminary test estimators in linear models least squares theory (Q4728054)

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 4003300
Language Label Description Also known as
default for all languages
No label defined
    English
    Relative performance of stein-rule and preliminary test estimators in linear models least squares theory
    scientific article; zbMATH DE number 4003300

      Statements

      Relative performance of stein-rule and preliminary test estimators in linear models least squares theory (English)
      0 references
      0 references
      1987
      0 references
      Stein-rule
      0 references
      asymptotic dominance
      0 references
      asymptotic risk-efficiency
      0 references
      multiple regression model
      0 references
      preliminary test estimation
      0 references
      shrinkage estimation
      0 references
      sub- hypothesis testing
      0 references
      linear model
      0 references
      least squares estimator
      0 references
      F-ratio
      0 references
      Pitman-type alternatives
      0 references
      loss function
      0 references
      Mahalanobis-distance
      0 references
      In the linear model \(X=Ba+e\), where the \(n\times p\)-design matrix B is known and the random vector e has i.i.d. components with finite variance, the parameter vector \(a\) is partitioned in \(a=(a^ T_ 1,a^ T_ 2)^ T\). One is interested in \(a_ 1\) when \(a_ 2\) is unknown but close to zero. The authors consider the following estimators for \(a_ 1:\) Let \(\tilde a\) be the least squares estimator (LSE) for \(a\), \(\hat a_ 1\) the LSE for \(a_ 1\) when \(a_ 2\) is taken as zero. Make a preliminary test for the hypothesis \(H: a_ 2=0\) based on the F-ratio and take \(\hat a^ P_ 1\) as \(\tilde a_ 1\), if H is rejected, and as \(\hat a_ 1\), otherwise. Further \(\tilde a^ S_ 1\) is a shrinkage version of \(\tilde a_ 1.\) The estimators are compared w.r.t. their asymptotic risk under Pitman- type alternatives, where the loss function is the Mahalanobis-distance and the risk is calculated w.r.t. the asymptotic distribution of the estimators. The conclusion is that \(\hat a^ P_ 1\) and \(\tilde a^ S_ 1\) behave better than \(\tilde a_ 1\) and \(\hat a_ 1\).
      0 references

      Identifiers

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