Description ver(1.0): Using content generated by R's MANOVA function, creates high precision power and p-values for the Wilks Lambda Statistic using the Butler/Wood Algorithm (Butler, R.W. and Wood, A.T.A."Approximation of Power in Multivariate Analysis" in Statistics and Computing Vol. 15,pp. 281-287,2005 doi:10.1007/s11222-005-4071-x).In addition provides an S4 object that computes these same statistics useful in evaluating multiple dependent variables models created by other R programs. In a second and third part, computes the likelihood ratio statistic for block independence and equivalence of covariance matrices along with p-values using other Butler/Wood methods.
Description ver(1.1) adds the following sentence: In a fourth part, computes a sample size estimate for 1-way MANOVA using the Wilks Lambda statistic and Butler/Wood algorithm.
Ver(1.2) differs from version 1.1 in that I found a one instruction error in the Wilks Lambda code in version 1.1. This was in the routine p_kprs that computes the first derivative as a function of s using an analytical formula. It changes the results in the first five vignette examples, mostly with minor numerical differences. However, in example 2, Wolf Skull measurements with all the variables, the variable “location” displays significance whereas in Ver. 1.1 it did not. In example 3,the interaction term is now significant whereas in Ver. 1.1 it was not. In example 4, Soils Data, with all variables considered,contour along with depth displays significance unlike in ver. 1.1. In example 5, with only 7 variables, only the interaction term is not significant, unlike in ver. 1.1. In my defense, I wrote all this code (and documentation) myself. And there are thousands of lines of it.
The correction in p_kprs also allows the number of dependent variables that can be handled in the 1-way MANOVA study design section to be expanded from 7 to 10. I believe in the future it could be expanded further.
Also for ver(1.2): In reference to the functions statsBIf and statsBIncatf, I have developed a much better algorithm than the one now in the R package, an ensemble method. But it exists only in Python code at the moment.
This github repository also contains a PDF of Butler/Wood 2005. In addition a tar.gz which can be downloaded to separately install the R package from a local file without using the devtools program with the R GUI (that is, >devtools::install_github("chvrngit/wmpvaer")).
You can download the tar.gz file from the repository and then install the package using devtools and install_github. But remember if you go the devtools and install_github route ,and want the vignette, you have to use devtools::install_github("chvrngit/wmpvaer",build_vignettes=TRUE) instead and use >vignette("wmpvaerV", "wmpvaer") to view the vignette created.
JUST WANT THE VIGNETTE? YOU ARE IN LUCK- YOU DON'T HAVE TO CREATE A COPY OF IT YOURSELF. I have also included a file "wmpvaerVpdf.pdf" in this github repository. It contains the PDF vignette.
On November 10,2019, I was approved by Github to participate in the Github Sponsored Developer program. This means that if someone wanted to pay me money to add to wmpvaer and related open source Github computer programs on an ongoing monthly basis, Github would collect the money and send it to me. And for one year partly match the payment made. Without financial support progress will be limited otherwise. For those interested see https://github.com/sponsors.