Modeling habit formation in the real world: the case of hand-washing during COVID-19 times

Abstract

During the COVID-19 crisis, people were asked to wash their hands more often to decrease the spread of the virus. An efficient way to make people do this is to make it into a habit. The dynamics of habit formation over time have been modeled statistically, and more recently several computational models have been proposed to explain how habits form through behavior repetition. However, these computational models haven’t been validated by data from the real world nor compared. In this thesis, both approaches are used to examine habit formation in the real world. First, a simulation study was conducted, which showed that the computational models all described habit formation with an asymptotic curve, although they differed in how fast the habit strength increased and decreased, and what maximum could be reached. Second, a 28-days field study was conducted, in which 46 participants’ hand-washing behavior was measured by sensors, while their habit strength was measured with daily surveys. When modeling self-reported habit strength statistically, an asymptotic curve did not describe the habit strength trajectories better than a linear line. These empirical trajectories were then compared to the trajectories reproduced by the computational models using parameters estimated from the data. It was found that Klein’s model (2011) was best at describing how the habit strengths of the participants changed over time as a result of repeated behavior. To examine which characteristics of the models were useful in describing habit formation, the models were also compared with three naïve models. Klein’s model performed better than all naïve models. These results imply that even though people differ substantially in habit formation, the generic principles of nonlinear growth and decay are well captured by the Klein’s model.