GitHub - janlisec/eCerto: An R package to conduct statistics for certification of reference materials. (original) (raw)

eCerto

eCerto is an R-package providing functions to perform statistical tests required during the production of a certified reference material.

The production of certified reference materials (CRMs) is a core task of the Bundesanstalt für Materialforschung und -prüfung (BAM). Various statistical tests are required and applied depending on the task and recorded data to ensure that reported values of CRMs are appropriate. Many of them are calculated according to the procedures described in ISO GUIDE 35:2017.

The eCerto package contains a Shiny-App which provides functionality to load, process, report and backup data for the statistical evaluation of analyses conducted during CRM production.

The app also allows to perform long term stability monitoring (or post certification monitoring) to ensure the CRM values remain within specification.

Another app module allows to perform statistical tests for analytical method validation, computing the working range and linearity of a method.

Installation (as a local package)

To run eCerto locally you can install the package fromCRANor the development version from GitHub and start the app with:

CRAN

install.packages("eCerto")

GitHub (alternative for the latest version)

install.packages("devtools") devtools::install_github("janlisec/eCerto")

start the app

eCerto::run_app()

Apart from run_app(), the eCerto package currently provides only some example data sets and few exported functions.

Live App

Before installation you can test the most recent app version hosted atBAM using your own or the provided example data.

eCerto is frequently updated. Should you experience problems using the_most recent_ version from the link above, please:

• report a GitHub Issue and
• try the (hopefully) more stable version also hosted atBAM.

Detailed documentation

A description of eCerto showing a re-evaluation of previously published CRM data can be found in this scientific article. Most of the functionality is also described in the online help.

This is a list of the implemented statistical tests and functionality:

• Homogeneity
  • mean, standard deviation, n
  • ANOVA P-value (bonferroni adjusted)
  • uncertainty contribution
  • boxplot of specimen
  • z-scores over all analytes and specimen
  • downloadable HTML report
• Stability
  • linear model parameters, their errors and significance
  • uncertainty contribution for specified shelf life
  • annotated graphical representations of the linear models and their uncertainties
  • Arrhenius calculations to evaluate stability using different temperature levels
  • downloadable HTML report
• Certification (Inter Laboratory Comparison Study)
  • testing outliers for mean and variance of labs (Scheffe, Dixon, single/double Grubbs, Cochran)
  • testing the lab mean distribution (Bartlett, ANOVA/KS, Skewness/Agostino, Kurtosis/Anscombe)
  • sequential filtering of outliers possible
  • uncertainty contribution of ILC study and combined uncertainty (relative and absolute)
  • annotated graphical representation of analyte data
  • downloadable HTML report per analyte or for the CRM
• Long Term Stability
  • simple post certification testing as an option
  • long term monitoring with sequential updating of remaining shelf life
  • downloadable PDF report/certificate
• Method Validation
  • working range
  • linear model parameters and statistical evaluation of the residuals (Mandel, Neumann, KS)
  • residual comparison against quadratic models
  • LOD/LOQ
  • Trueness, Precision, Measurement uncertainty as statement boxes for reference
  • downloadable HTML/Word report