Identifying memory mechanisms in Bayesian models of behavioural change during epidemics

Accurate modelling of epidemic dynamics often requires accounting for how individuals modify their behaviour in response to perceived infection risk. While existing behavioural change epidemic models attempt to capture this feedback, they usually make simple, ad hoc assumptions about how memory affects responses. For example, they mostly focus on recent cases only, thereby overlooking how earlier experiences continue to shape current perceptions of risk. This study introduces a unified framework of Memory Mechanism Enhanced Behavioural Change (MEBC) models within a Bayesian SIR epidemic modelling setting. Five alternative memory mechanisms are examined - memoryless, sliding window, power-law, exponential, and reciprocal - each characterizing a different way in which past epidemic information influences current behaviour. A fully Bayesian data-augmented MCMC scheme is used to jointly estimate transmission and behavioural parameters, while accounting for uncertainty in infectious periods. Simulation experiments demonstrate that the MEBC models recover parameters accurately and remain robust under misspecified memory structures. Applications to early-stage COVID-19 outbreak in Miami-Dade County and to the 2023-2024 influenza season in Manitoba show that incorporating an easy-to-interpret memory mechanism substantially improves model fit, highlighting the critical role of collective memory in shaping behavioural adaptation and transmission dynamics.