NEWS

ic.infer 1.1-7 (2023-10-04)

updated CITATION to newer format
updated URLs to bht-berlin instead of beuth-hochschule
removed latin1 encoding
renamed internal functions orlm.forboot orlm.forboot.fixed to orlm_for... in order to avoid that they are interpreted as S3 methods

changed % to \% in example code in manual
changed title to title case
changed start of description (removed note because of "This package")
adapted namespace file to current requirements (S3Methods instead of export, importing functions from grDevices, stats and graphics)

Bug fix (serious bug, thanks to Leonard Vanbrabant for alerting me to this): df calculation in function ic.test was wrong for TP = 11: df were always too large because the length of the parameter vector was used for calculation instead of the number of constraints; this resulted in conservative p-values for TP = 11 in ic.test and consequently also for the overall test in summary.orlm; for the latter, the first parameter of the beta distribution was always 0.5 too large. The bug has been fixed, and notes have been added to the documentation files of functions ic.test and orlm.
added an explanation to the documentation of pbetabar, explaining which mixture is calculated (different from the expectations that users of likelihood ratio tests in linear models might have; the denominator always adds the error sum of squares from the unrestricted model only, and therefore, the second beta parameter in the mixture is constant)

updated CITATION file
removed assignment of object bodyfat into global enviroment (file aaa.R)
moved data files to sysdata.rda instead of data directory
moved vignette files to vignettes directory
updated vignette to include the JSS article in the references
moved packages from Depends to Imports in package description, adapted namespace file to mention all relevant imports; adapted example and test code to access objects from packages no longer attached using :: or require

makde pdf documentation file into a vignette
made bodyfat data available in the package (typed them in), i.e. got rid of function bodyfat()

removed LICENSE file that erroneously claimed a geographical restriction to non-US since version 0.8. ic.infer is not and has never been geographically restricted, but is subject to GPL >=2.

added contrast function contr.diff for increment coding of factors with ordered levels
added function make.mon.ui for easier creation of matrix ui in monotonicity situations
corrected orlm.Rd (quotes in code blocks)
added two data examples (bodyfat and grades)

orlm and or.relimp did not work properly for linear models on data frames without column names.
bootstrapping did not work on models with factors because of an undocumented and unnecessary stopping condition in internal function boot.orlm.

improved error message when using orlm or or.relimp with non-quadratic matrix
added tests directory for standard automated code testing
added examples for summary.orest to documentation
Bug fixes:
Fixed typo that made ic.test use asymptotic instead of finite df test, when applied to linear model object
or.relimp did not work for covariance matrices without column names

Namespace now exports summary function for class orest (erroneously yielded default summary before, although a special method had been added with version 0.8-2)
Added helpful error message for plot.orlm, when applied to object that was generated based on covariance matrix and therefore without residuals

Prohibit simultaneous occurrence of restriction on intercept and bootstrap (consequence of switching to covariance-matrix based calculation in bootstrap)
Bug fix for internal function orlm_forboot (index 96 1 replaced by index in call)
Bug fixes for ic.weights (even and odd sums were wrongly implemented and only worked for some numbers of restrictions)

ic.weights saves additional time on weight calculation by exploiting that odd and even weights each sum to 0.5.
orlm and or.relimp can now operate based on a covariance matrix only and do not necessarily require a linear model for input.
all.R2 now requests a covariance matrix instead of a linear model for input (it should not be necessary for users to directly call all.R2).
Bootstrapping in orlm is now based on the covariance-matrix-based calculation for option fixed=FALSE. Thus, issues are avoided in making bootstrapping work with functions that indirectly call lm. 95 The documentation has been updated and improved in some respects.