Comparative reliability of rank and choice data in stated preference models