To anticipate relationships between future climate change and societal violence, we need theory to establish causal links and case studies to estimate interactions between driving forces. Here, we couple evolutionary ecology with a machine-learning statistical approach to investigate the long-term effects of climate change, population growth, and inequality on intergroup conflict among farmers in the North American Southwest. Through field investigations, we generate a new archaeological dataset of farming settlements in the Bears Ears National Monument spanning 1,300 years (0 to AD 1300) to evaluate the direct and interactive effects of precipitation, temperature, climate shocks, demography, and wealth inequality on habitation site defensibility—our proxy for intergroup conflict. We find that conflict peaked during dry, warm intervals when population density and inequality were highest. Results support our theoretical predictions and suggest cascading effects, whereby xeric conditions favored population aggregation into an increasingly small, heterogenous area, which increased resource stress and inequality and promoted intergroup conflict over limited productive patches. This dynamic likely initiated feedback loops, whereby conflict exacerbated shortfalls and fostered mistrust, which drove further aggregation and competition. Results reveal complex interactions among socioclimatological conditions, all of which may have contributed to regional depopulation during the thirteenth century AD.