Data Science on the Google Cloud Platform

Chapter 5. Interactive Data Exploration

In every major field of study, there is usually a towering figure who did much of the seminal work and blazed the trail for what that particular discipline would evolve into. Classical physics has Newton, relativity has Einstein, game theory has John Nash, and so on. When it comes to computational statistics (the field of study that develops computationally efficient methods for carrying out statistical operations), the towering figure is John W. Tukey. At Bell Labs, he collaborated with John von Neumann on early computer designs soon after World War II—famously, Tukey was responsible for coining the word bit. Later, at Princeton (where he founded its statistics department), Tukey collaborated with James Cooley to develop the Fast-Fourier Transform, one of the first examples of using divide-and-conquer to address a formidable computational challenge.

While Tukey was responsible for many “hard science” mathematical and engineering innovations, some of his most enduring work is about the distinctly softer side of science. Unsatisfied that most of statistics overemphasized confirmatory data analysis (i.e., statistical hypothesis testing such as paired t-tests¹), Tukey developed a variety of approaches to do what he termed exploratory data analysis² (EDA) and many practical statistical approximations. It was Tukey who developed the box plot, jack-knifing, range test, median-median regression, and so on and gave these eminently practical methods ...