The absence of medicines to cure COVID-19 calls for preventive strategies, including mask-wearing. Despite its protection against exposure to coronavirus, not everyone chooses to wear a mask. Some studies addressed mask-wearing behaviour from the standpoint of behavioural economics, one being the effect of herding behaviour. This occurs when people base their decision on the decisions of others. A literature review shows that herding can play a role, yet it has only been empirically studied up to a correlational relationship and not a causal relationship. This study focuses on quantifying the effect of factors that influence mask-wearing, emphasising herding. This study can be used in a more realistic epidemic transmission study that involves preventive health behaviours and can also help policymakers simulate the impacts of their policy designs more accurately. The study comprises two phases: choice and agent-based modelling. A literature review is performed to identify possible factors. Four possible factors are identified: Health Belief Model (HBM), herding-related, situational cues, and demographics. After collecting 151 questionnaire respondents within the Netherlands, a Latent Class Cluster Analysis identifies two clusters representing health beliefs. Class 1 are people who are more risk-averse towards COVID-19 and believe more in mask-wearing efficacy. Class 2 is the opposite. The choice modelling confirms a statistically significant effect of herding, only within friends and/or family and the random people. To explore the macro-level population dynamics, the interdependent behaviour between people is considered. An agent-based model (ABM) has been formalised, implemented, verified, and validated. The main insight is that the herding effect is stronger when the majority is not wearing masks than when the majority is wearing masks. In other words, there is a tendency towards no-mask-wearing, instead of the opposite. These discoveries unveils a new insight on how herding affects mask-wearing using an under-explored way of combining static and dynamic research methods. This study has several limitations. First, the risk of low-quality responses from online survey is uncertain. Second, care should be taken in generalising the result to the Netherlands population. Third, the literature review may miss important factors that were undiscovered in the previous literature up to April 2021. Fourth, this study generalises the factors in a broad categorisation. Finally, the ABM employs simplified input values. Future studies may also incorporate interviews, a more representative sampling method, and a more fine-grained specification of factors. Finally, five recommendations are derived for the policy-making process. First, this study recommends policy-makers maintain clarity in communicating mask-wearing policy. Second, enforcement of mandatory policy is recommended, especially in outdoor spaces. Third, mask-wearing can be encouraged through social campaigns, if necessary. Such campaigns may contain figures that people can closely relate to. Another alternative is to increase the importance of policy by putting signs in more prominent places and informing people about the enforcement. Fourth, the modellers in the policy-making domain could look at incorporating herding to such research and enrich its realism. Lastly, this study can also be utilised for other policy-making processes outside the mask-wearing context and/or outside the COVID-19 context.