To not have selection bias so you can measure the effects