restriktor constrained statistical inference

Version 0.1-80.711

• Released on CRAN: December 5, 2017

• Fixed bug; the print.goric() function gave the wrong relative evidence in case of the complement.

• Changed; the first robust Wald test has been removed. In linear models the first and second Wald tests are equivalent.

• Changed; the test-statistics for robust M- and MM-estimators have been updated. The robust Wald test now uses the full parameter vector. Only the robust score test uses the block that corresponds to the inequality-constrained parameters in the covariance matrix of the M- and MM-estimator.

• Changed; the standard deviation from the summary.rlm() function is not correct. It uses a robust version of the sum of squares. Therefore, it has been changed to the method given in Yohai (1987, p. 648, Eq. 4.2, 4.3 and 4.4).

• Added support (not fully tested yet); These new argument bound for the goric() function allows for testing 'about equality' constraints.

• Added support; information about the weights is printed in the restriktor() output and the summary() output for objects of class rlm.

• Known issues:

  • same as 0.1-70.

• Bugs/glitches discovered after the release:

  • just released.

Version 0.1-70

• Released on CRAN: August 24, 2017

• Fixed bug; the parametric and model-based bootstrap resulted in an error for the Wald and score test-statistic (class rlm only).

• Fixed bug; an error was given if mix.weights = "none" and no constraints were specified.

• Changed; the test-statistics for robust M- and MM-estimators have been updated. For most bounded-influence estimators the error distribution should be symmetric, otherwise the intercept and center of the error distribution are confounded. Therefore, to compute the test-statistics we now use the block that corresponds to the inequality-constrained slope parameters in the covariance matrix of the M- and MM-estimator.

• Added support; the parametric bootstrap for shifted/relocated cones is now supported for the gaussian family.

• Added support; the goric() function can now compute the complement of one order-constrained hypothesis.

• Added support; order-constrained estimates for the multivariate normal linear model. No standard errors are available (yet).

• Added support; likelihood ratio test-statistic for the multivariate normal linear model. No Wald- or score-test statistic is available (yet).

• Known issues:

  • the parametric bootstrap for shifted/relocated closed convex cones is incorrect if the family object is not gaussian.

• Bugs/glitches discovered after the release:

  • The print.goric() function gives the wrong relative evidence in case of the complement. The goric weights are correct, so the values can easily be calculated by hand.

Version 0.1-55

• Released on CRAN: February 22, 2017

• Known issues:

  • the parametric bootstrap for shifted/relocated closed convex cones is incorrect if the parametric bootstrap is used. For now, an error is given. This will be fixed in the next release.

• Bugs/glitches discovered after the release:

---