Purpose
Student evaluations of faculty and courses have been used as a method of quality control in higher education for almost 100 years. Analysis of the data generated by these surveys has been the focus of considerable research for at least the past 20 years, with the bulk of this analysis using techniques which assume that the survey data, usually Likert item responses, are numerical variables. This paper uses an alternative form of analysis, correspondence analysis, which makes no such assumptions on the variables, to provide new insight about student evaluations.
Method
Correspondence analysis is used to compare data which comprise profiles of categories (in this case, the item responses from the surveys) and makes no assumptions of order between categories or of numerical values attached to categories. The method can, however, be used to produce a numerical value known as an optimal scaling; this reflects the inherent differences between ordered categories more faithfully than a simple linear scale such as “Strongly Agree” = 5, “Agree” = 4, etc. We have applied this process to several sets of survey data comprising student evaluations of faculty and unit for a tertiary institution.
Results
The surveys analysed included items to evaluate both unit and instructor in the same instrument. When applying correspondence analysis to the questions as a whole, the expected pattern of response transition from Strongly Agree through Neutral to Strongly Disagree was not observed; instead, the Strongly Disagree category appeared between the Agree and Neutral categories. Separating the items into two groups based on whether they evaluated unit or instructor resulted in a pattern with Agree having a higher score than Strongly Agree for items relating to unit evaluation whereas instructor evaluation items again placed the Strongly Disagree category between Agree and Neutral. Preliminary results for surveys from another institution show the expected response pattern suggesting that this may be a characteristic of the institution under investigation.
Conclusion
The use of an arbitrary scale to report results for individual Likert items on student surveys, whilst providing a simple way of reporting results, may not be giving an accurate picture of student evaluations. The reason for differences in response patterns between unit and instructor responses may be due to students completing surveys without giving sufficient thought to responses and providing random answers, however the results call for further investigation.