This version adds Youden plots, with a range of robust covariance estimates for data ellipses and support for bivariate outlier marking on the plots.
In addition it adds support for logdensity to the dt.scaled function, for which the log argument was not active.
0.9-16 fixes a rarely-used dt.scaled calculation
The experimental version 0.9-14, available via the Files tab in the 2.15 branck of the /src and (for Windows) /bin repository, improves MC further (typically a factor of 5 or more over version 0.9-13 on examples with B=100000) by using eval internally instead of apply. An switch is provided to revert to apply for non-vectorisd functions.
The latest upload drastically improves uncertMC calculation time for medium to large MC runs. For a 10000-point MC run on the first help page example, system time on a 2.8GHz dual core Pentium improved from 19.5 sec. to 0.44 sec.
This gain arises from removal of unnecessary x-y correlation calculations, retaining only the pearson correlation for diagnostic plots.
This is a minor bugfix of 0.9-11, correcting a kplot feature that set plot limits inappropriately when observations were reported bu confidence bounds were missing.
Version 0.9-10 has the same functionality as version 0.9-09, but a range of code changes has been made across the package to remove inconsequential R CMD check warnings which impede CRAN submission. Users are encouraged to download version 0.9-10 for trials to make sure it is stable before moving to release.
This version is likely to be the first submitted to CRAN.
This modest update adds robust variants to Mandel's plots. Outliers are usually much more obvious when compared to robust estimates of dispersion, at the expense of some detail for non-outlyig groups at default plotting scales.
0.9-05 adds a function 'kpoints' to place extra points with CIs on an existing kplot.
0.9-04 adds degrees of freedom for the mandel-paule function, previously set to NA
This minor release adds a derSimonian-Laird location estimator and fixes a minor bug in the mandel-paule code.
Version 0.9 includes new code for Mandel's statistics for interlaboratory study and is built for R 2.12.2.
Version 0.8-36 is avaialable as .zip and tar.gz for R 2.9 and R 2.10.
0.8-36 includes a vangel-ruhkin estimator and implements a default plot method for ilab objects (including a bug-fix for a problem affecting do.pdf implementation for the ilab method). Read the help page _carefully_ for the v-r estimate; it does not take the same parameters as mpaule()!
The latest tested download is version 0.8-34, which includes Duewer plots for interlaboratory comparisons. This is the default download for all platforms.
In addition, version 0.8-35 is available via the 'all files' link. 0.8-35 includes an implementation of Vangel-Ruhkin estimation using maximum likelihood. This is yet to be tested so should be considered a preliminary release.
The current build is 0.8-29. WIndows .zip and tar.gz files are available on the downlod pages.
Although a pre-alpha release, this includes the three main elements intended for version 1:
- Uncertainty evaluation is substantially complete for expressions, formulae and functions, and can use symbolic or numeric derivatives (first only) and provides Monte Carlo evaluation. Correlation is supported, including correlation in Monte Carlo simulation for normally distributed variables. Print, plot and update methods for the uncertainty evaluation result are provided and again essentially complete.
- Location estimators: a few examples (MM-estimation, M-estimation and Mandel-Paule) are implemented (with a class for the return value) to give a flavour and allow comment.
- Support for interlaboratory metrology comparisons: Two standard plots are available in the form of default plotting functions, and both are substantially complete. In addition, an 'ilab' class is implemented, with constructor and extraction/subset methods. The 'ilab' class is fairly advanced, but discussion as to the contents and structure are welcome.