How Respondents Use Verbal and Numeric Rating Scales: A Case for Rescaling

Abstract

The dominant practice among researchers is to treat verbal rating scales as interval in nature because of the vast array of analytical techniques that this opens up when it comes to analysis. This practice prevails despite warnings to the contrary that go back over half a century. A similar assumption seems safer when it comes to numeric rating scales. This paper revisits the issue to caution researchers to use only methods appropriate to the level of the data unless the proper rescaling is employed. The change in chi-square technique is developed to supplement rescaling using correspondence analysis, to uncover how scales are used by respondents. These techniques are applied to a sample that uses a verbal scale and three samples that use numeric rating scales. In all cases, the assumption of interval behaviour of the data proves to be a poor one. Rescaling is found to preserve the association among the variables. Strong evidence that rescaling changes the distribution of the variables leading to changes in the meaning of basic descriptive statistics is provided. Further research in this area and in the field of cross-cultural research is suggested.