‘Making an Impact’ in 25 Years

Twenty-five years ago, Leo Breiman (whom the statistics world sadly lost in 2005) delivered the 1994 commencement address, “What Is the Statistics Department 25 Years from Now?” to his statistics department at Berkeley. It’s a wonderful address for many reasons, but especially for his predictions for our field. He was astoundingly on target. He began by reminding us about how pervasive statistics is—and yet how little many people know about what statisticians do!

He went on to describe the changes in statistics departments, from theory to applications:

Breiman predicted that today, in 2019, statisticians will be computing Fourier transforms of speech, analyzing text, working with MDs on a 10-year study of brain cancer treatments, and watching “some astronomers and Professor Stark arguing about how strongly the data [show] evidence of a Big Bang origin for the universe.” (Wouldn’t he be pleased to hear about Stark’s appointment to the advisory board of the US Election Assistance Commission and its Special Committee on Election Security?)

The Joint Statistical Meetings (JSM) last week in Denver showcased many opportunities in which statisticians, often in collaboration with other scientists, have made an impact and will continue to do so. Might some of them have been on Breiman’s list today if he were to give another commencement address?

Forensic science: The Center for Statistics and Applications in Forensic Evidence is coordinating research teams for projects related to pattern and digital media evidence (e.g., fingerprints and cell phones) on the effects of interventions in forensic laboratory procedures to reduce errors in process and interpretation and effects of presentation of evidence and analyses to jurors to minimize misunderstandings. Though common, pattern evidence is not the only type of forensic evidence. Other forms of forensic evidence equally in need of validation include trace evidence such as paint; glass; tape; anthropological and environmental evidence such as bones, animals, and explosions; and chemical evidence such as gun-shot residue, toxicology, and drugs. Collaborations with chemists, biologists, and engineers, as well as forensic scientists, are essential for progress in this area.

Gun violence: The National Institute of Statistical Sciences (NISS) held an inaugural Statistics Serving Society two-day forum (honoring the memory of Ingram Olkin) titled “Gun Violence—The Statistical Issues.” More than 50 invited attendees (criminologists and statisticians from government agencies, nonprofit organizations, and universities) identified issues in data acquisition and discussed analysis methodologies and outstanding statistical issues, including measures of effectiveness of interventions and analysis of police shootings. Roundtable discussions focused on future challenges that invited the statistical community to collaborate with other researchers, laying the groundwork for interdisciplinary collaborations.

Social networks: A high-impact area of statistics, particularly for understanding sociological behaviors, lies in the analysis of social networks and complex social systems. Network data are noisy, with massive observational errors, requiring statistical methods to extract “signal” (mechanisms of human interactions) from “noise.” For instance, to study drug abuse behavior of adolescents, statistical analysis of social networks can lead to better inferences and more reliable prediction and monitoring of health-related behaviors. Networks of citations and collaborations among scientists can point to the discovery of intrinsic connections between different scientific subjects and research topics and quantitative assessments of mechanisms regarding human knowledge in different fields driving scientific research collaborations. Statistical analysis of networks related to social security have demonstrated valuable insights into terrorist organizations that reveal potential risks. It also raises important questions about differential privacy and how to ensure an individual’s privacy. (For more about research in this area, see, for example, “Prediction Models for Network-Linked Data” by Tianxi Li, Elizaveta Levina, and Ji Zhu in the Annals of Applied Statistics.)

Medical Records: The federal statistical system has faced multiple challenges during the past three years, particularly in resources to ensure completeness of data collection and accuracy of reporting. An important area relates to the growing opioid crisis in this country, which has put severe strains on the system through which deaths are analyzed and reported in the United States. Opioid deaths are expensive to investigate; they require autopsies and sophisticated toxicology. The reporting systems, which were already stressed, have been unable to cope with the rapidly increasing number of deaths. Jay Kadane at Carnegie Mellon University is working with Karl Williams at the Allegheny County Office of Medical Examiners in Pittsburgh to address issues with incompleteness and inaccuracies in reporting deaths due to drug overdose. Some of these issues recall the under-reporting that arose with HIV and AIDS, and with sexually transmitted diseases before that, but today’s drug overdoses involve much larger numbers and hence more difficult challenges. (For more on this problem, see “Accurate Reporting of US Opioid Deaths: Level, Type, Temporal and Geographical Comparability” by Kadane and Williams in the Statistical Journal of the International Association for Official Statistics.)

In what other areas can we forecast big impacts of statistical design, algorithms, and methodologies? I could not end this column any better than Leo Breiman ended his 1994 commencement address, so I’ll simply quote him (for those who never saw any of George Lucas’ Star Wars movies, his last line comes from the character Obi-Wan Kenobi):

Editor’s Note: Donna LaLonde, Jay Kadane, Tianxi Li, and James Rosenberger contributed to this column.