Model misspecification in stepped wedge trials: Random effects for time or treatment