What Drives Algorithm Use? An Empirical Analysis of Algorithm Use in Type 1 Diabetes Self-Management
49 Pages Posted: 25 Aug 2021 Last revised: 24 Oct 2022
Date Written: October 20, 2022
Advancements in algorithms hold promise to enhance operations by improving users' decision-making. However, people sometimes fail to use algorithms, which could be a barrier to achieving such improvements. Using the bolus calculator (algorithm) use behavior in over 221,000 bolus insulin decisions from a field experiment on type 1 diabetes self-management (Aleppo et al. 2017), we contribute field analysis to identify drivers of algorithm use. We find that 1) previous algorithm use increases future algorithm use, 2) being out-of-target with self-driven decisions increases subsequent algorithm use, whereas being out-of-target with algorithm-driven decisions does not impact algorithm use, 3) exposure to multiple, potentially-conflicting measurements for a single algorithm input decreases algorithm use, 4) increasing one's need for precision increases algorithm use and 5) previous deviations from algorithm recommendations decrease future algorithm use. These field results complement laboratory experiments in the human-algorithm interactions literature by supporting/rejecting natural hypotheses from existing laboratory results, identifying relevant drivers that have yet to receive attention, and providing effect magnitude estimates in setting with highly-invested decision-makers over several months. They also provide insight into how to increase algorithm use: encourage habit formation, strengthen performance feedback, eliminate sources of algorithm input uncertainty, highlight a need for greater precision, and intervene when observing people deviating from algorithmic recommendations.
Funding: None to declare
Declaration of Interest: None to declare
Ethical Approval: We obtained approval from the University of Southern California Institutional Review Board that our research is considered non-human subjects research as it meets the criteria for coded private information (study ID: UP-20-00856). That is, our data uses de-identified (publicly available) data from a published field study (REPLACE-BG).
Keywords: healthcare operations, behavioral operations, algorithm aversion, Type 1 Diabetes
Suggested Citation: Suggested Citation