Mediation analysis is a methodological approach that enables us the decomposition and understanding of the complex mechanisms by which an exposure variable can influence a dependent variable through one or more mediators. This technique is particularly relevant in the epidemiological context, where data are often structured hierarchically, with individuals grouped within clusters. This multilevel structuring adds a layer of complexity to the mediation analysis, requiring the adoption of specific statistical techniques and tools. Recent software make it possible to implement these analyses while adhering to the fundamental principles of causal mediation. Among these tools, the "mediation" package for R stands out for its ability to address challenges associated with multilevel data. We used this tool in a case study that assesses how different information sources influenced public adherence to restrictive measures during the COVID-19 pandemic. Using data from 17 countries, this study illustrates how multilevel mediation analysis can provide essential insights into the factors influencing public behavior during this unprecedented health crisis.