Which design best supports causal inference between social media use and later depression?

Longitudinal study with measured confounders

Cross-sectional survey with self-report measures

Single-time case series of depressed users

Ecological correlation across countries