Abstract:

Nearly 100 million Americans attend church weekly to hear sermons, which convey cultural and religious ideas in the context of daily life. Research has shown that regional factors, like wealth or natural disasters, shape the content of religious belief. However, these findings largely rely on survey data and imperfect proxies of religious content.

In this presentation, the content of 30,000 sermons from 5 Christian denominations across hundreds of US counties (2019-2020) is analyzed using multiple methods. First, natural language processing techniques describe sermon content. Then, variance decomposition identifies sources of variation in content (church, denomination, county, state, census region). Finally, elastic net models are used to explore 1) how well sermon content predicts local cultural indicators and 2) which sermon aspects are most indicative of these indicators.

Sermons primarily focus on biblical themes but prominently feature mundane topics, like personal anecdotes and God's role in one’s relationships. Church-level localization is significant, with geography playing a secondary role. Finally, sermon content predicts certain cultural markers like political orientation and income per capita, but not others.

Religious sermons represent an untapped data source for scholars to investigate how religious institutions convey culture and interact with the socio-ecological environment.

Speaker's social media:

Twitter: @MatthewBillet1