Version 1.0.8

• Argument “length” of seq() fixed to “length.out”
• Fixed Rprintf() and REprintf() bad arguments in lmrob.c

Version 1.0.7

• The opt and mopt loss functions, known as rho functions, specified by the choice family = “opt” or family = “mopt” in the function lmrobdet.control are now calculated using polynomials, rather than using the standard normal error function (erf) as in versions of RobStatTM prior to 1.0.7. The numerical results one now gets with the opt or mopt choices will differ by small amounts from those in earlier package versions. Users who wish to replicate results from releases prior to 1.0.7 may do so by using the family arguments family = “optV0” or family = “moptV0”. Note that the derivative of the rho loss function, known as the “psi” function, is still an analytic function. For further details, see the Vignette “polynomialRhoFunctions”.

Version 1.0.6

• locScaleM produces a warning and returns 0 for the estimated scale when more than half of the input values are equal to each other.

Version 1.0.5

• Replace legacy S-compatibility deprecated macros DOUBLE_* in R_ext/Constants.h (included by R.h) by the standard C99 constants

Version 1.0.4

• For the function lmrobdetMM.RFPE, when the argument bothVals is set to TRUE, the names of the first element in the returned list has been changed to minRhoMM.C (from minRhoMM).

Version 1.0.3 (November 2021)

• The Shiny User Interface capability has been removed, and correspondingly the Shiny Interface to RobStatTM Vignette has also been removed. The reason for doing so is that the Shiny capability resulted in too many package dependencies. We anticipate that Greg Brownson (gregory.brownson@gmail.com), the creator of the RobSTatTM Shiny UI, will make that capability available in an independent package complement to RobStatTM.

• The package fit.models has been removed as a dependency, i.e., it is no longer listed as Depends. The reason for this is that including fit.models in RobStatTM created too many dependencies. The fit.model package stand-alone functionality works very well with RobStatTM. Correspondingly, the use of fit.models has been removed from the “Vignette for Command Line Use of RobStatTM.pdf” document, and is now provided in the separate vignette “fit.models using RobStatTM.pdf”.

• In src/lmrob.c add USE_FC_LEN_T and use FCONE when calling BLAS and LAPACK Fortran functions (https://cran.r-project.org/doc/manuals/r-devel/R-exts.html#Fortran-character-strings)

• Re-ordered the elements in the object returned by covRob and Multirobu, renamed the argument cor to corr, and now the correlation matrix (if the argument corr = TRUE) is returned in the element cor.

• Renamed the returned entry weights to wts and the order of other entries in the object returned by covClassic

• Removed comment on the help page of lmrobdetMM.RFPE referring to it being for internal use. This was not correct. The function can be used directly.

• lmrobdetMM.RFPE now includes the argument bothVals. If set to TRUE the function returns a list with the two terms (named minRhoMM and penaltyRFPE) that added together equal the RFPE (see equation (5.39) in Section 5.6.2 of the book “Robust Statistics: Theory and Methods (with R)”. If bothVals is FALSE then the function returns a scalar with the RPFE value.

• Objects returned by covClassic and covRob now include an element call with an image of the call that produced the object with all the arguments named (the matched call).

• The function lmrobLinTest has been renamed to lmrobdetLinTest

• Fixed a bug producing undesired behavior when an exact fit (more than half of the data lying perfectly on a line) was detected.

• The help page for lmrobdetMM was revised to describe all entries in the returned object.

Version 1.0.1:

• Fix an issue with exact fits if the M-scale estimate is (close to) 0.
• Includes help pages for all datasets.