Is agent-based modeling the future of behavioral science?

How has behavioral science shifted?

Nudging is as hot as ever. The application of behavioral science principles to real-world interventions has gained a huge popularity in the last 10 to 15 years. Admittedly, “nudging” and the wider family of interventions related to it have generated enormous amounts of impact at a relatively low cost (UK Pension auto-enrollment increased participation by close to 60%; FDA mandated calorie labeling in the states decreased daily calories consumed by about 100 on average, etc.). Given results like these, it’s not too surprising that people are excited about the approach (there are now over 500 “nudge units” across the world and counting). However, as with any scientific application, these first interventions have borrowed relatively simple principles from the behavioral sciences, focusing on easily modified frameworks such as biases or general rules about how people behave. This simplification has led to some failures and forced the field to advance.

As behavioral science research has come out on what works and what doesn’t, practitioners have learned that most findings don’t generalize as well as one might be tempted to suggest—increasing the importance and prevalence of running highly customized interventions. So, instead of just assuming that a signature at the top of a form is more likely to produce honest behavior because a (now proven dubious) experiment said so, a good practitioner might perhaps ask deeper questions like: what is the equivalent of a signature in the context of this app I’m working on? How can I more broadly define “honest” behavior to best measure impact? What kind of experimental design would allow me to see if this is an effective intervention?