In this project, I wrote functions for scraping data from SLU student evaluation PDFs, created an interactive Shiny app displaying this data, and performed ordinal logistic regression to examine the relationship between student evaluations and student class year and sex.
Student evaluation data is measured using a Likert Scale. This means that, as most students know, participants are provided with a set of statements to which they must rate their level of agreement. SLU evaluations employ a 7-point Likert scale, ranging from the lowest level of agreement (Disagree Strongly) to the highest (Agree Strongly). These responses are quantified, with the lowest level equalling a score of one and the highest a seven.
Likert scale data is ordinal, meaning the difference between each level is unclear. For example, what truly differentiates the responses Agree Somewhat and Agree? This impacts the analysis and visualization of the data. This project aimed to construct the appropriate visualizations for data of this type: heat maps and bar graphs.
Only counts and percentages of student responses, as opposed to averages, were used for this project. Averages of Likert scale data can be misleading, with average scores sometimes painting a very different picture from the actual distribution of responses. This is why Likert scale data should be visualized.
The final Shiny App provides a project description and links to my GitHub repository. Users are prompted to upload their SLU student evaluation PDFs. Users can upload multiple PDFs for an ordered comparison of responses across time or a single PDF. From there, users can choose count or percentage, question, color scale, and type of visualization, and can choose to download plots as PDFs. Lastly, the full data set and scraped student demographic data are available as well.