Population descriptors (PDs) in biomedicine, like race and ethnicity, suffer from two problems. They introduce explanations with low proportionality, which fail to capture relevant causal relations; and their usage is motivated by a focus on global instead of local health issues. Both problems amount to, what we call, the ‘missing locality challenge’ of PDs leading to epistemic biases and stereotypes against local communities. We introduce a new framework to meet this challenge. It ranks PDs from ‘global’ to ‘local’ contexts of application, which helps selecting appropriate descriptors for different kinds of disease causalities and groups without stereotyping them.