Design of Experiments in R
En la pasada edición useR 2011 realizada en la Universidad de Warwick pudimos disfrutar de una maravillosa charla de Ulrike Grömping sobre el Diseño de Experimentos en R y en concreto la “view” sobre DoE que ella misma mantiene. Tanto el video como las diapositivas se encuentran en el siguiente enlace, mientras que el abstract se puede ver en la propia web del Congreso.
The purpose of Design of Experiments (DoE) is to create an experimental scheme that will generate data “as informative as possible” for the purpose of a planned (series of) experiment(s). Basic ideas in experimental design have been proposed very long ago (Wikipedia mentions references from 18th century). Often, R.A.Fishers contributions in early 20th century are seen as the foundation of modern experimentation; he systematically introduced randomization, blocking, replication and factorial design and broadened insights into the applicability of analysis of variance (cf. e.g. the account by his daughter Joan Fisher Box 1980).
Fisher – being based at Rothamsted experimental station in the 1920s and early 1930s – was experienced in agricultural experiments. Agriculture has its own special challenges, e.g. spatial correlation between neighbouring fields. Design of Experiments is also used in many other fields of application, e.g. medicine, psychology, sensometrics or quality improvement. The talk gives an overview of what functionality is available in R for experimental design and analysis of experimental data in general (cf. also the Task View on Experimental Design and Analysis of Experimental Data, Grömping 2011) and then focuses on industrial experimentation (cf. e.g. Box, Hunter and Hunter 2005) and a series of R packages by the author (DoE.base, FrF2, DoE.wrapper, RcmdrPlugin.DoE) as well as other packages (AlgDesign, BsMD, DiceDesign, lhs, rsm) on which the former rely.
Particularly in connection with offering a GUI, it is also discussed what should be the trade-off between guiding or even forcing untrained useRs to do the right thing vs. offering substantial flexibility and leaving responsibility for appropriate choices with the useR.
Espero os resulte interesante.