Karen Kafadar

This year, the Joint Statistical Meetings (JSM) will be held in Denver, Colorado. They are “Joint” because they are sponsored jointly by the ASA, Institute of Mathematical Statistics, Statistical Society of Canada, International Biometric Society (ENAR and WNAR), and seven other statistical organizations; “Statistical” because all involved identify—to varying extents—with the field of statistics; and “Meetings” because the conference consists of meetings of sections and committees, as well as technical sessions, short courses, information tables, career services, book and software exhibits, and other activities. It can be overwhelming!

Has it always been this way? Has JSM become too large? Should we adopt a format from other societies to allow only refereed or invited sessions? More essentially, are in-person meetings an anachronistic concept whose time has come and can be replaced by virtual interactions, which are less costly and much less effort to organize?

These questions prompted me to return to one of my favorite presidential addresses by Dr. W. Michael O’Fallon, the ASA’s 95th president in 2000. (Who remembers ASA president’s addresses? I confess, I do not remember many, but I do remember this one.)

Dr. O’Fallon traced the history of the “annual ASA meeting” which, by its 70th in 1908, was judged to be a remarkable success because it attracted a total of 40 people (versus 6,000+ today)—record-breaking attendance because the ASA finally decided it was time to hold it outside Boston, where the previous 69 had been held!

A major change in the Annual Meeting occurred during the second and third decades of the 20th century (1909–1928), when ASA presidents were encouraged to pursue “loose collaboration with other associations, [especially] the American Economics Association. Indeed, the board minutes often contained instructions to the president to make every effort to determine when the economists were to meet and to schedule the ASA meeting accordingly.”

We no longer tie our meetings to suit the economists’ timetable, but we do coordinate our dates (seven years in advance) with our JSM partners. Starting in 1944, a program committee (versus solely the ASA president) organized the Annual Meeting; sessions of contributed papers were added in 1947 to the invited sessions; and ASA sections were invited to participate directly in planning the scientific program in 1950. In tracing the history, Dr. O’Fallon brought to light the developments leading to the current JSM and, most importantly, the value of flexibility and recognizing when change is needed. Our 40th ASA president, Helen Walker, also recognized the need to hold an annual meeting in 1944, following a two-year hiatus due to World War II.

Technology has changed many aspects of JSM. The majority of attendees today opt out of the hardcopy version of the program, choosing instead to view sessions on their phones or tablets via the JSM app, and presentations are far more elaborate than they were even 10 years ago. In fact, many of these presentations and/or the research behind them are available online before or after the meetings! Is JSM even necessary?

One interesting anecdote Dr. O’Fallon recalled from the 1930s was “a conflict that had a major impact on our profession, our association, and our meetings”: the balance between applications (on which the ASA had largely focused) and methodology (driven by “mathematical statisticians”). It is reminiscent of the debate we face today between “statisticians” and “data scientists.” That conflict led to the founding in 1935 of the Institute of Mathematical Statistics. What will today’s debate bring?

Last month, ASA Executive Director Ron Wasserstein and I attended the ACM-IMS Interdisciplinary Summit on the Foundations of Data Science in San Francisco, co-chaired by Jeannette Wing (computer science) and David Madigan (statistics). It was the first of many anticipated annual meetings with multiple partners in the vein of JSM. The program included presentations and panels by impressive researchers and developers, and it was streamed live. It prompts the question: Why attend the meeting in person at all? Or, any meeting, for that matter, when a paper or pre-print might be available on the internet?

Dr. O’Fallon asked the same question in 2000: “Why not a virtual meeting where we post our presentations on our personal websites and have the meeting program provide author, title, and URL?” (Actually, most of us do that now!) “Will this type of social and professional discourse be enough to hold us together as a professional society? I think not.”

Nor do I. Ron knew our in-person attendance at the data science summit would be useful—in some ways we could predict, but also in ways we could not. For me, it reinforced the value of in-person meetings. Several speakers raised my awareness about future areas of importance in which statisticians should be involved, as well as the value, risks, and consequences of “machine learning” algorithms. Some speakers were intentionally provocative, as is needed for a meeting in which statisticians, mathematicians, and computer scientists are still working on ways to ensure effective collaboration as “data science.”

Xiao-Li Meng raised the problem of ensuring accurate inferences arising from the privatization of the data forthcoming in the US 2020 Decennial Census and urged the audience to engage in problems raised by differential privacy. (JSM offers a short course on this topic Tuesday morning, as well as six sessions throughout the meeting.)

Several speakers talked about the value of “black box” algorithms, particularly their stunning successes in marketing and health care. Others talked about methods for validating, improving, and characterizing these algorithms, especially quantifying the uncertainties.

One highly notable comment came from panelist Andrew Gelman: “The ‘algorithmic bias’ that concerns me is not so much a bias in [the accuracy in] an algorithm so much as a social bias arising from the demand for, and expectation of, certainty.” (This sentence reminded me of comments from a 1984 JSM Panel of Distinguished Statisticians—including George Box, Sir David Cox, Morris Hansen, and C.R. Rao—when John Tukey said, “Our greatest failure as statisticians has been our failure to communicate to the average layperson the notion of uncertainty.” Andrew’s statement suggests we’ve not come very far!) Even those data science summit attendees who were reading and sending text messages, or working on their computers, during the talks were more engaged in the discussion than is possible when viewing a session remotely!

Like our 2000 ASA president, I have come to believe in the irreplaceable value of in-person contact. JSM remains one of the world’s largest meetings of statisticians. It offers a full range of topics in statistical theory, methods, and algorithms involved in all the applications in which statistics can “Make an Impact.” JSM Program Chair Richard Levine and his program committee, whose members were elected by ASA sections and JSM partners, have been working hard to ensure diversity in technical sessions, invited panels, introductory overview lectures, and workshops. We encourage you to search the online program for timely and important topics, including “big data” surveys, causal inference, Census 2020, data science education, deep learning, differential privacy, analyses of and inferences from microbiome data and electronic health records, reproducible research, statistical foundations and research, teaching data science, statistical communication and collaboration, and uncertainty quantification and visualization.

Rich’s committee has worked hard to ensure a broad range of methodologies, applications, and educational and career opportunities throughout the entire JSM, including Thursday! Wendy Meiring has facilitated your appreciation of the Sunday evening invited poster session by organizing the posters around themes: geophysical modeling, learning analytics, medical imaging/neuroscience, object-oriented data analysis, uncertainty quantification. It’s a terrific opportunity to interact, one-on-one, with the presenters.

Other notable sessions include the special free Public Lecture Sunday at 6 p.m., during which Mark Glickman will update the application of methods for author attribution used by David Wallace and Frederick Mosteller in 1964 for “The Federalist Papers: Hamilton or Madison?” to “The Beatles: Lennon or McCartney?”

We will have a broad array of topics in our plenary addresses and special lectures, IMS Presidential Address, the Wald and Medallion lectures, and the annual Fisher Lecture. I am thrilled that this honored list will include Teresa Sullivan, president emerita of the University of Virginia, who will speak as my invited speaker (Monday at 4 p.m.) about the 2020 Census from her perspective as a quantitative sociologist. These featured events will be complemented by a “late-breaking” session, numerous invited and contributed talks, and poster sessions.

The size of the meetings—43 parallel sessions in addition to the plenary sessions each afternoon—can be both a blessing and a curse. But mostly I believe it is the former, in part because the JSM Program Committee has tried to include your suggestions for improvements over the years. We hope this year’s JSM will offer you at least this much. Please come this year to learn something new, to develop and deepen your professional collaborations, and especially to tell us what JSM can do to make it even more relevant and valuable to you—our participants—next year!