On Becoming Indispensable

The voice on my radio at 5 a.m. one day last month was that of NPR’s Noel King on Morning Edition:

Her guest, New York Times columnist Binyamin Appelbaum, elaborated:

Noel King continued:

This last sentence reminded me of a statement made by Vijay Nair, D. A. Darling Professor Emeritus at the University of Michigan and head of statistical learning and advanced computing at Wells Fargo, during a conversation: “People in the real world are looking for solutions to problems—straight answers, not ‘If …’ or ‘But …’ or ‘Maybe …’.”

And people will listen to anyone who provides these answers. At a House Committee hearing, titled “Raising the Bar: Progress and Future Needs in Forensic Science and Standards,” Rep. Jerry McNerney asked, “How do you see AI methods as advancing the practice of forensic science?” It was a good question for those who understand AI (artificial intelligence) and ML (machine learning) methods as promises to solve all problems.

As statisticians, however, most of us are accustomed to working carefully through problems to ensure safe and reliable solutions. In his lecture as winner of the 2019 International Prize in Statistics (Kuala Lumpur, August 19, 2019), Brad Efron talked about prediction, estimation, and attribution. He noted that statisticians focus on estimation—estimation of the model and its parameters and standard errors, and attribution of the effects (significance testing) —whereas data scientists focus on prediction. And often the target is prediction for a specific set of circumstances, not for a general set of conditions that would require estimation. He shared some insightful views on the two disciplines.

Many people frame the “invasion” of data science as “How can we convince data scientists that we are data science?” Or, to use Efron’s insight, “How can we do more prediction, as people want?” But I am not sure we want to go that route. Perhaps the question for us is rather, “How do we convince people they really want statisticians’ insight, estimation, and valid inference from data?” How do we do what economists did in the 1950s—convince people, especially high-level decision-makers, that they cannot live without us?

I do not think all the recent attention on p-values will persuade people to listen to statisticians. (It may have the opposite effect. People may decide they don’t need statistical methods at all.) Nor will waving a “credential” like “accreditation” (unless it’s the Nobel Prize), nor overselling the methods we develop. We have been properly cautious when caution is appropriate. So what is the answer?

Appelbaum attributed the revolution in the world’s view of economists to two important events: the failing US economy in the 1970s and the presence of an enormously influential economist named Milton Friedman. According to Appelbaum, Friedman’s message was that government should reduce its role in the economy and put its trust in the markets to allocate resources. “And for a generation that is confronting the failure of the economy, this has enormous appeal,” said Appelbaum.

I hope it does not take a failing economy for statisticians to be seen as indispensable and that we can find ways, during good times and bad, to prove our value and impact. Of course, some statisticians have risen to positions of indispensability: Janet Norwood (commissioner, Bureau of Labor Statistics, 1979–1991); Katherine Wallman (chief statistician, Office of Management and Budget, 1992–2017); Stella Cunliffe (director of statistics, UK Home Office, 1972–1977). How can the respect they earned as individuals be translated into recognition of indispensability for the entire profession?

People are drawn to experts when they see them providing solutions to their problems. As John Tukey wrote long ago, “Finding the right question is often harder than finding the answer.” Indeed, we are quite good at that—especially when the proposed solution may be subject to serious, unanticipated biases that had not been recognized by others. We all have seen that, by calling attention to the shortcomings in a proposed approach or running a (possibly flawed) experiment or analyzing a (possibly biased) data set, statisticians have saved their collaborators much time and many resources.

We face an uphill battle with the explosion of data science. Recently, the International Data Science in Schools Project, chaired by Nicholas Fisher with 12 committee members from statistics and computer science, issued a report titled, Curriculum Frameworks for Data Science. Nick’s committee offers frameworks that “will provide the basis for development of courses in introductory data science for students in their final two years of secondary school and of courses to teach teachers how to teach introductory data science.”

After the ASA’s successful efforts introducing statistics into grades K–12, we may soon discover “statistics” is nowhere to be found in those grades while students all know what “data science” is—even if the concepts and methods they are learning such as “select appropriate numerical and graphical summaries to answer questions posed about a single feature/variable in a data set,” “interpret numerical and graphical summaries in the context of the original problem to answer questions posed about the original problem and make discoveries” and “classify questions and hypotheses as to whether they apply to the sample at hand or to a larger population” (Curriculum Frameworks for Data Science, pp. 20–21) are those we learned as “statistics”! (It’s a very thoughtful report, by the way.)

So, we face challenges: ensuring “statistics” does not become an obsolete term; enlightening data science administrators who think any class with data should be taught as data science; and reminding our colleagues we solve problems. We can tackle these challenges together. Here are some ideas, and I encourage you to contribute others.

First, we can all remind our data science colleagues that, when they are teaching statistical concepts, those concepts have been around for some time. They’ve been taught in statistics courses for the past few decades. (My colleague Jordan Rodu proposed a slogan for our department: “Statistics at UVA: We Put the Science in Data Science.”)

Second, we can resurrect the many resources statisticians have developed for introducing data analysis and statistical concepts in the classroom. People might realize those resources, having been refined through years of use, may be more practical and suitable for their purposes.

And, most importantly, we can seek—and seize—opportunities to demonstrate our skills at solving problems—the bigger, the better.

Remember the economists; they took on the problems of the entire economy! Maybe we can convince the world that we statisticians can solve everyone’s data problems and save the world. And then, instead of being “lowly [statisticians]” in dark offices, we will see our talents become indispensable to scientists and advisers and counselors, because we have proved ourselves repeatedly in areas like survey development and analysis, clinical trials, risk assessment, signal extraction, experimental design, physical modeling—areas in which other disciplines may have difficulty claiming expertise.

My dad used to say, “Bad times are good times for good people.” The world’s challenges may give us opportunities that will lead to good times for statisticians. We will have to work together to make that happen, and I welcome your ideas!