“Exploratory data analysis” (EDA) (Tukey 1977) is the term that John W. Tukey gave to the practice of exploring the data with visualizations and summaries. The careful data analyst should do EDA before going on to the more familiar mode of making statistical models, and testing them.
I remember coming across these ideas when I was was a PhD student, in 1996. I bought one of Tukey’s books in a second hand bookshop.
In retrospect, I hardly heard about EDA after that. I didn’t use it myself.
I just came across a 2003 article (Curtis and Araki 2003) recording its quiet disappearance:
The purpose of this research was to analyze recent statistics textbooks in the behavioral sciences in terms of their coverage of exploratory data analysis (EDA) philosophy and techniques. Twenty popular texts were analyzed. EDA philosophy was not addressed in the vast majority of texts. Only three texts had an entire chapter on EDA. None of the authors used the term “confirmatory factor analysis” or discussed model building or cross-validation. Seven texts contained references to published work by Tukey, but these references were mainly for specific techniques, most typically the stem-and-leaf display and box-and-whiskers plot, which were presented in 15 and 9 texts respectively. The paper ends with recommendations for integrating EDA into the fields of psychology and education.
EDA has now made a triumphant return, as part of “data science”. As one example, see the introduction to R for data science.
I find this interesting, as part of the pattern that Donoho describes (Donoho 2015), of the tendency for academic statistics departments to study clever and satisfying mathematical models, instead of practical techniques for data analysis.
Perhaps EDA was too hard to teach, or even too hard to do, before we learned that data analysis must use coding.