A Course in Statistics with R

Chapter 4Exploratory Data Analysis

Package(s): LearnEDA, e1071, sfsmisc, qcc, aplpack, RSADBE

Dataset(s): memory, morley, InsectSprays, yb, sample, galton, chest, sleep, cloud, octane, AirPassengers, insurance, somesamples, girder

4.1 Introduction: The Tukey's School of Statistics

Exploratory Data Analysis, abbreviated and also simply referred to as EDA, combines very powerful and naturally intuitive graphical methods as well as insightful quantitative techniques for analysis of data arising from random experiments. The direction for EDA was probably laid down in the expository article of Tukey (1962), “The Future of Data Analysis”. The dictionary meaning of the word “explore” means to search or travel with the intent of some kind of useful discovery, and in similar spirit EDA carries a search in the data to provide useful insights. EDA has been developed to a very large extent by the Tukey school of statisticians.

We can probably refer to EDA as a no-assumptions paradigm. To understand this we recall how the model-based statistical approaches work. We include both the classical and Bayesian schools in the model-based framework, see Chapters 7 to 9. Here, we assume that the data is plausibly generated by a certain probability distribution, and that a few parameters of such a distribution are unknown. In a different fashion, EDA places no assumptions on data-generating mechanism. This approach also gives an advantage to the analyst of making an appropriate guess of the underlying ...