What drives preference for the compromise option? Disentangling position‐based versus attribute‐based effects