Breaking: robScale uses the bias-corrections factors
by default. Testing shows that the improvement is measurable and
significant.
Breaking: robLoc and robScale refactored to
move lesser-used options into an opts list and have a consistent
parameter order across the package.
Function examples now better demonstrate their usefulness.
Refactored and updated unit tests.
Corrected and updated documentation.
Removed unused, unexposed functions medianR and madf
carried over from older development.
New exported functions madn and admn providing
high-precision small-sample bias-corrected versions of the MAD
and ADM.
New optional constants for bias reduction in robScale (via
usefctrs and madfctrs).
Major rewrite: Complete port of the computational core to
modern Fortran 90, including a new optimized quicksort-based median
algorithm for small n. This gives substantially better performance,
cleaner code, and improved NA handling in line with base R.
Breaking: All small-sample (n \le 9) and asymptotic
bias-correction factors for madn, admn, and robScale
(both known and unknown location cases) were recomputed using 200 million
Monte Carlo simulations (small n) and 16.5 million simulations per
sample size (asymptotic: 10 \le n \le 100). The new 4-decimal
“AA” constants have Monte Carlo standard errors between
2.0 \times 10^{-5} and 9.1 \times 10^{-5}. The
original Croux & Rousseeuw (“CR”) factors remain available as an
option in madn and robScale
(Issue #1; thank you
@filiporn).
Breaking: robLoc and robScale now use the new
madn/admn internally with improved very-small-sample logic.
The new “AA” factors are now the default.
Unit tests expanded and modernized.
Documentation, GitHub workflows, and package metadata updated.
Various spelling, punctuation, and metadata issues.
Corrected package metadata unit test.
Breaking: In robScale, the constant representing
the denominator for x (see equation 26)—which allows for
\beta = \int{\rho(u)\;d\Phi(u)} = 0.5—was calculated to greater precision
than the four digits shown in the paper. This changes results slightly,
which is why the major version has been incremented.
Updated documentation.
Updated unit tests.
Expanded package metadata tests.
Use native CRAN DOI.
Added rhub checking.
Add SPDX codes to source files.
Update ‘README’, GitHub actions, and unit tests.
Explicitly mention use case for package to be 3–8 samples.
Note that asymptotic constant for adm may not apply with few
samples.
Update ‘README’.
Update GitHub actions.
Make code more compliant with style guide.
‘CITATION’ updated to use bibentry per CRAN
request.
Explicitly use integers in count and lengths.
Use singleton logical comparisons for short-circuit.
Update ‘README’ for Best Practice.
Added Best Practice badge to ‘README’.
Converted test suite to tinytest to reduce dependencies.
Augmented citation with DOI.
Corrected adm documentation.
Cleaned up help files and ‘README’.
Cleaned up source code and added SPDX identifiers.
Moved continuous integration to GitHub Actions.
Corrected CRAN requests.
Considered mature enough for official release.
Changed tests to be testthat 3rd edition compliant.
Changed default maxit to 80.
Added AppVeyor.
Added more tests.
Updated ‘README’.
Minor corrections in DESCRIPTION for CRAN.
Initial version.