INSIDE STORY

The late, great George Dantzig once told me in an interview marking his 80th birthday that any mathematical model that portends to approximate “reality” yet doesn’t take uncertainty into account is doomed to failure.

Of course, modeling uncertainty is not easy, which explains its absence in many models, and the job becomes magnitudes of degrees more difficult when you’re modeling human behavior instead of making widgets. Now imagine the human subjects – millions of them – you’re modeling are in near-panic mode during, say, a deadly, fast-spreading pandemic and your job is to predict how they will behave in order to predict how the disease will spread, thus helping the medical community effectively respond to the crisis.

That, in a nutshell, is the problem researcher and University of Toronto professor Dionne Aleman outlines in “Predicting the spread of pandemic disease” (page 24). As Aleman points out, for many years, the most common disease-spread models were based on the premise that populations were homogenous, and that people tended to become infected – and infect others – at about the same rate. Unfortunately, reality paints a different picture. Some people travel more than others. Urban dwellers tend to come in more contact with others than do rural folks. Some people seek immediate treatment, others don’t. In other words, humans don’t always behave according to the “model.”

Aleman explores the concept of agent-based simulation models (where people are agents) as an alternative to the homogeneous mixing model, but she cautions that “disease-spread models are not crystal balls.”

Decision-making in the face of uncertainty goes to the sweet spot of decision analysis. Last fall, a “Who’s Who” of decision analysis, including Ronald Howard of Stanford and Ralph Keeney of Duke, met to discuss the state of decision analysis at an NSF-sponsored workshop. Howard said, “Decision analysis requires mastering uncertainty, the main source of difficulty in decision-making. … When other approaches to decision-making analyze decision problems, they attempt to minimize or ignore uncertainty. Decision analysts surf on the sea of uncertainty instead of drowning in it.”

Ali Abbas, a professor at the University of Illinois who helped organize the event, captured the thoughts of many of the workshop luminaries including Howard’s for this issue of OR/MS Today (“Decision analysis: past, present and future,” page 30).

Gary Cokins, an analyst with SAS and an experienced analytics consultant, addresses the art and science of “selling” analytics and overcoming social, behavioral and culture issues in his essay “Obstacle course for analytics” (page 18). Humans are naturally adverse to change, so analytical consultants need to understand that the barriers to analytics corporate buy-in are far more likely to be human issues rather than technical problems.

“Most organizations believe that applying analytics is 90 percent math and 10 percent organizational change management with employee behavior alteration,” he writes. “In reality it is the other way around.”

You don’t necessarily need a degree in psychology or sociology in order to succeed in O.R. and analytics, but if you’re working with people, it clearly helps to know what makes them tick.

— Peter Horner, editor
peter.horner@mail.informs.org