Likelihood inference for linear regression models (Q3774753)

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 4034901
Language Label Description Also known as
default for all languages
No label defined
    English
    Likelihood inference for linear regression models
    scientific article; zbMATH DE number 4034901

      Statements

      Likelihood inference for linear regression models (English)
      0 references
      0 references
      1988
      0 references
      large-sample likelihood ratio methods
      0 references
      location-scale analysis of Darwin's data
      0 references
      asymptotic expansion
      0 references
      Bartlett adjustment
      0 references
      conditional inference
      0 references
      confidence limit
      0 references
      maximum likelihood
      0 references
      approximation error
      0 references
      nuisance parameters
      0 references
      log-likelihood-ratio statistic
      0 references
      unconstrained ML estimator
      0 references
      restricted ML estimator
      0 references
      conditional mean and variance
      0 references
      chi- squared approximation of the conditional distribution
      0 references
      0 references
      A linear regression model with p regression coefficients \(\beta_ j\) \((j=1,...,p<n)\) and n observations with error terms \(\sigma E_ i\) \((i=1,...,n)\) is considered. The \(E_ i\) are assumed to be i.i.d. with a known density f. Suppose that q of the parameters \(\beta_ 1,...,\beta_ p\), \(\vartheta =\ln \sigma\), are of interest, i.e. \(p-q+1\) and p-q of the parameters are nuisance parameters if \(\sigma\) is unknown or not, respectively. The parameters of interest are denoted by \((\omega_ 1,...,\omega_ q)=\phi\) and the set of nuisance parameters by \(\psi\). Further let \(\omega =(\phi,\psi)=\{\beta_ 1,...,\beta_ p,\vartheta \}\). Fixing the value of \(\phi\), the log-likelihood-ratio statistic for \(\phi\) is with the log likelihood function \(\ell (\omega)\) of \(\omega\) \[ W(\phi)=2\{\ell ({\hat \omega})-\ell (\phi,{\tilde \psi}_{\phi})\}. \] Here \({\hat \omega}\) is the unconstrained ML estimator of \(\omega\) and (\(\phi\),\({\tilde \psi}_{\phi})\) is the restricted ML estimator of \(\omega\) corresponding to the specified value of \(\phi\). For the case \(q=1\) and \(\omega =\vartheta\) the signed square root of the log-likelihood-ratio statistic is given by \[ R(\omega_ 1)=sgn(\omega_ 1-{\hat \omega}_ 1)\sqrt{W(\omega_ 1)}. \] Formulae for the conditional mean and variance of R are given. It is further shown that and how the error of order \(O_ p(n^{-1})\) of the chi-squared approximation of the conditional distribution of W can be reduced to an error of order \(O_ p(n^{-3/2})\).
      0 references

      Identifiers

      0 references
      0 references
      0 references
      0 references
      0 references