A study finds that students who eat breakfast score higher on tests. Does breakfast cause better scores? The students chose for themselves whether to eat breakfast, so this is an observational study. Breakfast eaters might also sleep more or have parents who help with homework. Any of these lurking variables, called confounders, could produce the higher scores, with breakfast simply along for the ride.
Now imagine a school randomly assigns 100 volunteers: 50 get a school breakfast and 50 don't. Random assignment spreads sleep habits, family support and everything else roughly evenly across the two groups. If the breakfast group then scores higher, the only systematic difference left is breakfast. That is why an experiment can say "causes" while an observational study can only say "is associated with."
Sometimes an experiment is impossible or wrong. Nobody can randomly assign people to smoke for 30 years, so the link between smoking and lung cancer came from observational studies: many of them, in different countries, showing that heavier smokers got sick more often, backed by lab evidence of how smoke damages cells. Strong observational evidence can be convincing, but each single study still shows only association.
Two groups do not make an experiment. Comparing students who chose band with students who didn't is observational, even with equal group sizes and a huge sample; a bigger sample shrinks random error but does not remove confounders.
Think of it like this. Counting umbrellas on a street and noticing that wet pavement comes with them might suggest umbrellas wet the ground. Rain causes both, a hidden common cause, just like a confounder. An experiment would hand out umbrellas at random on a dry day and check the pavement, which breaks the link to rain and shows umbrellas do nothing to the ground.
Core idea. Observational studies record what already happens (association only); experiments assign treatments and can establish causation.