
John Tukey worked at Princeton and Bell Labs and helped reshape practical data analysis through exploratory graphics, resistant methods and computing. These lessons distinguish his verified statements and contributions from later paraphrases. — Brillinger’s Tukey memoir.
Part 1: The Philosophy of Approximation
- Ask the right question: “Far better an approximate answer to the right question, which is often vague, than an exact answer to the wrong question, which can always be made precise.” — The Future of Data Analysis.
- Choose the real problem: An analysis should address the substantive question, not merely the problem a familiar procedure is convenient to answer. — Sanctification or Detective Work?.
- Treat data analysis as empirical science: Tukey argued that data analysis should be judged by experience and usefulness, not only by mathematical proof. — The Future of Data Analysis.
- Resist cookbook rigidity: Tukey warned that teaching only tidy, formulaic procedures can obscure the judgment and flexibility real data analysis requires. — The Future of Data Analysis.
- Judge methods by how they work: Brillinger records Tukey’s test for a good procedure as its performance in practice, even before it is fully understood. — Brillinger’s Tukey memoir.
Part 2: Exploratory Data Analysis (EDA)
- Approach exploration as detective work: “Exploratory data analysis is detective work—numerical detective work—or counting detective work—or graphical detective work.” — Exploratory Data Analysis.
- EDA needs flexible looking: Tukey emphasized looking at data, using visual techniques and remaining flexible about what to calculate instead of following only preconceived analyses. — Sanctification or Detective Work?.
- Learn what you can do first: “It is important to understand what you CAN DO before you learn to measure how WELL you seem to have DONE it.” — Exploratory Data Analysis.
- Find clues before judging them: “Unless the detective finds the clues, judge or jury has nothing to consider.” — Exploratory Data Analysis.
- Explore and confirm: Tukey argued that exploratory inspection and confirmatory analysis both matter; restricting every analysis to a preconceived method can hide unexpected findings. — Sanctification or Detective Work?.
- Let pictures surprise you: “The greatest value of a picture is when it forces us to notice what we never expected to see.” — Exploratory Data Analysis.
- Look for the unexpected: Simple plots can bring out patterns and surprises that a predetermined numerical summary might miss. — Exploratory Data Analysis.
- Show spread before formal tests: Tukey’s box-and-whisker display presents a distribution’s center, spread and outside values for exploratory inspection. — Exploratory Data Analysis.
Part 3: The Reality of Data
- Distinguish clues from conclusions: Re-examining the same data can yield valuable indications, but Tukey cautioned that such indications need further testing before being treated as established conclusions. — The Future of Data Analysis.
- Iterate through the data: Tukey rejected the idea that good analysis is always one straight, preplanned calculation; what one analysis reveals may call for another. — The Future of Data Analysis.
- Use resistant methods: Tukey advanced robust and resistant methods so a small number of extreme observations need not dominate an analysis. — National Academy memoir.
- Allow for nonnormal data: Tukey urged analysts to study methods under realistic, sometimes nonnormal variation rather than assume idealized distributions automatically fit the problem. — The Future of Data Analysis.
- Flag outside values with fences: In Tukey’s exploratory display, an inner fence lies one step—1.5 times the H-spread—beyond each hinge; points outside deserve attention, not automatic dismissal. — Exploratory Data Analysis.
Part 4: Coining Terms and Language
- Name the bit: Brockway McMillan recalled Tukey responding at a Bell Labs lunch, “Well, isn’t the word obviously bit?” — Origin of the Term Bit.
- Why bit won: Later alternatives “binit” and “bigit” appeared in the 1950s and 1960, but did not displace “bit”; they were not the terms Tukey rejected at that original lunch. — Origin of the Term Bit.
- Give software a name: Tukey used “software” in a 1958 printed discussion of programming’s importance alongside computer hardware; the original wording is not reproduced here. — Brillinger’s Tukey memoir.
- Coin technical language: Tukey’s coined terms included “cepstrum,” a spectrum-related term in time-series analysis. — Brillinger’s Tukey memoir.
- Advance the jackknife: Tukey helped develop and name the jackknife approach, which assesses an estimate through systematically omitted observations. — National Academy memoir.
- Keep the observations visible: A stem-and-leaf display groups values while retaining individual digits, giving the analyst a compact view without discarding the raw values. — Exploratory Data Analysis.
- Shannon credited Tukey for bits: In the opening footnote of his 1948 communication paper, Claude Shannon credited Tukey with suggesting the term “bits.” — A Mathematical Theory of Communication.
- Separate exploration from confirmation: Tukey’s exploratory-data-analysis framework distinguishes looking for patterns and surprises from later tests of what the data establish. — Exploratory Data Analysis.
Part 5: Computing and Algorithms
- Make Fourier analysis faster: Cooley and Tukey’s fast Fourier transform made discrete Fourier computation much more efficient, opening new possibilities for digital signal analysis. — National Academy memoir.
- Open up digital signal analysis: Brillinger describes the fast Fourier transform as a major advance for analyzing signals and time series on computers. — National Academy memoir.
- Work problems in his head: A Princeton remembrance reports that Tukey sometimes lay on his back at university scheduling meetings and called out solutions he had worked through mentally. — Daily Princetonian remembrance.
- Bring statistics to election night: Brillinger describes Tukey’s participation in RCA and NBC election-night projection work, including rapid analysis of incoming returns. — Brillinger’s Tukey memoir.
- Recognize the importance of software: Tukey’s early use of “software” reflected attention to the programmed instructions and interpretive routines needed alongside computer hardware. — Brillinger’s Tukey memoir.
- Use signals to monitor tests: Tukey contributed to seismic methods used to monitor underground nuclear tests. — National Academy memoir.
- Compare groups without easy false alarms: Tukey’s honest significant difference procedure is part of his multiple-comparisons work, addressing error when comparing several group means. — National Academy memoir.
- Downweight extremes: Tukey’s biweight is among the resistant estimators associated with his work; it limits the influence of extreme observations. — National Academy memoir.
Part 6: Teaching and Students
- Teach before taking a statistics course: Brillinger records Tukey recalling that the first statistics course he attended was one he was there to teach. — Brillinger’s Tukey memoir.
- Teaching takes time: “Teaching data analysis is not easy, and the time allowed is always far from sufficient.” — The Future of Data Analysis.
- Speakers need not bring answers: Tukey recalled that speakers in his applied-statistics seminar need not know the answers to the problems they brought. — Conversation with John W. Tukey.
- Invite questions during the talk: Tukey recalled that seminar audiences could ask almost any question at almost any time. — Conversation with John W. Tukey.
- Make uncertain remarks safe: The seminar norm was that an exploratory remark should not later be held against the person who offered it. — Brillinger’s Tukey memoir.
- Treat statistics like an unsolved crossword: Brillinger likened Tukey’s statistical work to a crossword puzzle whose answer key may never be available; the analogy is a paraphrase, not Tukey’s verified verbatim statement. — Brillinger’s Tukey memoir.
- Build a statistics department: Tukey became the first chair of Princeton’s statistics department in 1966, not 1965. — National Academy memoir.
- Remove the answer key: Brillinger recalls Tukey giving students crossword books with the answers removed, an exercise in working without guaranteed solutions. — Brillinger’s Tukey memoir.
- Keep an unassuming personal style: Tukey’s long-time Bell Labs secretary recalled his familiar black golf shirt and khakis; a White House medal ceremony was an unusual occasion for him to wear a suit. — Daily Princetonian remembrance.
Part 7: Collaboration and Consulting
- Play in many fields: As reported by the New York Times, Tukey told a colleague that a statistician gets to “play in everyone’s backyard.” — New York Times obituary.
- Think with others’ brains: Brillinger records Tukey’s consulting aphorism: “A consultant is a man who thinks with other people’s brains.” — Brillinger’s Tukey memoir.
- Work between Princeton and Bell Labs: Tukey held long-running roles at Princeton and Bell Labs, bringing mathematical methods into applied collaborations. — National Academy memoir.
- Treat statistics as shared methodology: In his Princeton oral history, Tukey described statistical thinking as methodology useful across many subject areas rather than an isolated branch of mathematics. — Princeton oral history.
- Cross from chemistry into mathematics: Tukey recalled training in chemistry at Brown before moving into mathematics, a path that preceded his broad statistical collaborations. — Princeton oral history.
- Learn the substantive field: Brillinger describes Tukey’s habit of engaging with a collaborator’s actual scientific problem rather than treating statistics as a detached calculation service. — Brillinger’s Tukey memoir.
- Collaborate across disciplines at Bell Labs: Tukey’s Bell Labs work brought him into sustained contact with engineers and information theorists, including Claude Shannon. — Brillinger’s Tukey memoir.
- Value practical performance: Brillinger portrays Tukey as judging methods by how usefully they worked on real problems, not only their mathematical elegance. — Brillinger’s Tukey memoir.
- Keep discussing the problem: Brillinger’s first-hand account depicts Tukey as persistently engaged with colleagues’ questions and new applications well beyond formal meetings. — Brillinger’s Tukey memoir.
Part 8: Science and Public Policy
- Question the Kinsey sample: Tukey and fellow statisticians challenged the Kinsey studies’ sampling methods because a large nonprobability sample cannot substitute for a sound probability design. — National Academy memoir.
- Assess ozone risks under uncertainty: Tukey chaired a National Research Council committee examining the environmental effects of chlorofluoromethanes, including possible ozone depletion and uncertainty about regulatory timing. — Brillinger’s Tukey memoir.
- Join the 1972 environment conference: Tukey served on the U.S. delegation to the 1972 United Nations Conference on the Human Environment. — National Academy memoir.
- Work on wartime fire control: During World War II, Tukey worked on fire-control research that included B-29 problems and anti-aircraft applications. — Princeton oral history.
- Contribute to wartime code work: Brillinger records Tukey’s involvement in wartime mathematical work related to German Enigma communications. — Brillinger’s Tukey memoir.
- Advise on reconnaissance: Tukey served on a committee whose recommendation helped lead to the U-2 reconnaissance aircraft; this was advisory work, not evidence he designed the plane. — National Academy memoir.
- Apply statistics beyond the classroom: Tukey’s first-hand memoir traces applications from wartime defense and telecommunications to social research and environmental policy. — National Academy memoir.