The hardware and bandwidth for this mirror is donated by dogado GmbH, the Webhosting and Full Service-Cloud Provider. Check out our Wordpress Tutorial.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]dogado.de.

RATest 0.1.12

Bug fixes (Khmaladze permutation tests)

This release corrects several errors in the Khmaladze-transformation permutation tests PT.Khmaladze.fit() and PTQTE.Khmaladze.fit(). Users of these functions in earlier versions should re-run their analyses. A detailed erratum documenting each issue, its mechanism, and the corrected behavior is available separately.

• PT.Khmaladze.fit() — empirical process. The two-sample empirical process was constructed so that the test statistic did not depend on the treatment sample. It now evaluates both empirical CDFs at the control quantiles (generalized-inverse convention) as intended.

• PT.Khmaladze.fit() — score function. The extended score is now estimated on the correct (outcome) scale and with the correct normalization (-psi from quantreg::akj, i.e. f'/f, rather than f').

• PT.Khmaladze.fit() / PTQTE.Khmaladze.fit() — evaluation grid. The compensator grid was set to three times the total sample size, which disabled the score-direction projection of the martingale transform. The grid is now tied to the control sample size, restoring the intended behavior.

• PTQTE.Khmaladze.fit() — observed statistic. The observed treatment assignment was not included among the permutations, forcing the rejection rate to equal the nominal level regardless of the data. The observed assignment is now included, so the test has power against alternatives.

New features

• PT.Khmaladze.fit() gains an akj.control argument, an optional named list of tuning parameters passed to quantreg::akj for score estimation (alpha, kappa, h, iker1). The default (standard akj settings) is recommended: in simulation, alternative settings that reduce the score’s estimation error were found to lower test power.

Notes

• Under the designs used in the associated papers the corrected tests have approximately nominal size. Under heavier-tailed designs, finite-sample size distortion can appear at small samples and diminishes as the sample size grows.

These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.
Health stats visible at Monitor.