|
Abstract:
|
In the past 35 years, statisticians have seen substantial methodology developments for ordinal data, yet many researchers in other fields still analyse such data using ordinary normal theory methods by naively assigning scores to ordinal levels. This talk presents a description of the dangers associated with such a naive approach. Along with some well-known disadvantages, such as floor/ceiling effects and the violation of normality assumptions, we show other possible dangers. For example, it is common to believe that when the number of ordered response categories is large, it is acceptable to treat them as continuous. However, it is not necessarily true for survey data. We discuss this issue using an example survey where respondents answered questions using both a 7-point Likert scale and a 0-100 scale. Also, we propose a possible way to evaluate average scores for a group of respondents using stereotype models.
|
