Study on the most appropriate scale for use in questionnaires: an experiment with Kano's model