test2x2 no longer computes the sums of squares from the closed-form 2x2 formulae. The model log(Var) ~ GRP/SUBJ + PRD + TRT is now fitted with the sasLM package: the analysis of variance comes from sasLM::T3test (Type III sums of squares, GROUP tested against SUBJECT(GROUP)), the geometric least square means from sasLM::LSM, and the confidence interval from sasLM::CIest. All reported values are unchanged apart from floating-point noise.
The analysis of variance table now lists Df before Sum Sq, matching SAS PROC GLM and sasLM. The rows and their contents are unchanged.
test2x2 now stops with a message if the TRT column holds anything other than 'R' and 'T'.
The rtf package is back on CRAN, so BE depends on it again. The vendored copy of its functions (‘R/rtf_utils.R’, added in 0.2.5) is removed, along with the R.oo and R.methodsS3 imports it needed.
The dependency on the archived rtf package was removed. The subset of rtf functions required by be2x2 is now vendored internally in ‘R/rtf_utils.R’ (originally authored by Michael E. Schaffer, GPL (>= 2)); BE now imports R.oo and R.methodsS3 for the RTF object system.
The example dataset NCAResult4BE moved from internal ‘R/sysdata.rda’ to ‘data/’ with LazyData.
NAMESPACE now uses explicit exports (instead of a broad pattern) and registers the internal RTF S3 methods.
be2x2 no longer opens on-screen graphics devices when writing an RTF report.
Corrected the mean-and-SD panel of plot2x2 (error-bar whisker and label position); no test result is affected.
Documentation typos and inaccurate \value/\details sections fixed; internal formula duplication removed.
powmse now uses the standard TOST power p1 - p2 (previously p1 - p1*p2, which overstated power); this also affects powcv, ssmse, sscv, ssscv and the sample-size rows of test2x2. Some determined sample sizes increase by about one subject per sequence.
hodges nonparametric 90% confidence-interval limits corrected (off-by-one in the order statistics; also removes an error at the smallest samples). The test is renamed to Wilcoxon Rank-Sum (Mann-Whitney) and its p-value is now two-sided.
scaledBound applies the 50% CV cap only to the EMA/MFDS bound (k = 0.76); the US FDA bound (k = 0.893) is no longer capped.
trimws is used for the input data.frame
Some quotation mark UTF-8 characters in the rtf file are changed to ASCII code
Some typo fixed
Too low power or too large sample size are reported better.
sssCV and scaledBound for scaled average BE now support US FDA criteria.
sscv and ssmse return power value as an attribute.
Interval criteria for scaled average BE can be calculated with scaledBound function
Asterisks are shown in the ANOVA table
Negative between-subject variance will be shown as 0.
Beutified ANOVA result in test2x2 and be2x2
Minor update due to the upgrading of R system
2x2x3 design is supported. See the help of powmse.
be2x2 supports output to be saved in a rtf file
First version