Byran Smucker, 2019 SPES Outgoing Chair

In one telling of the trajectory of the Section on Physical and Engineering Sciences, we are yesterday’s news. In this version, industrial and engineering statistics crested decades ago and SPES is just clinging to a small, niche corner of the statistics profession. We are more or less irrelevant in the era of big data analytics.

But there is a different lens through which to view SPES, one that suggests we are well-positioned to make a substantial mark on our field going forward. In engineering and the physical sciences, we observe increasing data abundance. While traditional approaches to data analysis, experimental design, quality control, and reliability are still critical for many members of SPES, a new graduate working as a statistician in industry or among engineers and physical scientists is more likely to build predictive models than to design a small experiment. Given the changing landscape, the new tools used in the analysis of big data are relevant and should be seized upon by our constituency.

But there’s more.

Not only are the tools of analytics just as relevant in SPES domains as in many others, the traditional areas we have cultivated over the years have an important role to play in advancing data science. Allison Jones-Farmer, in the 2018 Youden Address, argued that industrial statisticians have training in three areas that many data scientists do not: (1) up-front study design; (2) inferential validity; and (3) understanding the differences between confirmatory, explanatory, and predictive modeling.

To take the first—which is near and dear to my heart—the “old news” view sees the small data of traditional experimental design as increasingly irrelevant in a big data world. But look a little closer and connections between data science and experimental design are important and plentiful. Nathaniel Stevens pointed out that experimental design expertise is sought after by many technology companies, and working data scientists have noted this as well. In areas as diverse as online experiments, subsampling, active learning, causal inference, and algorithm tuning, design thinking and ideas are critical to data science.

The traditional areas in SPES are not going away. There are still increasingly complex and expensive processes and products to study, monitor, and improve. There are still difficult questions in the physical sciences that require classical statistical methods. But we shouldn’t limit ourselves to the methods that have always been used in our section. Let’s welcome new types of data and methodologies into our statistical repertoire; let’s use these new opportunities to make research contributions to areas in data science and machine learning and be alert for ways our training can be brought to bear in big data domains. If the future of data science is bright, so is the future of SPES.