When must scientist conduct controlled experiments?

Usually when scientist wants to test whether one compound or one phenotype is related to another phenotype. Then they need to control this variable when conducting experiments.