Step 1: To call something a cause, you must be sure the exposure came before the disease and that you can measure how often disease develops in exposed versus unexposed groups.
Step 2: Only a cohort design does this naturally. You begin with a disease-free population, split it by exposure status, and watch over time. The forward (prospective) direction guarantees temporality and lets you calculate incidence and relative risk straight away.
Step 3: Compare the rivals. The case-control approach looks at past exposure in already-ill patients, so timing is reconstructed and bias creeps in. A cross-sectional survey is a one-time snapshot, mixing old and new cases, so cause and effect blur. An ecological study compares whole populations, not individuals, and can mislead through the ecological fallacy.
Step 4: If experimental designs were on the list (randomised trials, or pooled evidence from meta-analysis) they would rank higher, but among the purely observational choices given, cohort wins for proving cause.
\[\boxed{\text{Cohort}}\]