Data cleaning is probably more important for text analytics. We didn’t do much (or anything) to clean our data before we feed them to Voyant. In real research, this step of data cleaning and preprocessing would be critical.
Sampling
Our sample size is very small, and not many variations have been revealed.
We only sampled one single year (2020), while social factors are deeply historical.
This is a biased sample, because we sampled from top 50 universities!
Analysis
We only read the documents, but didn’t examine the social, political and institutional contexts of vaccination expectations. These could be narratives on COVID as a political or public health issue, debates on state regulation vs freedom, and the eonomic costs on higher education institutions (e.g. decreased number of international students).
In our mixed methods analysis, we stopped at the summary statistics. What’s beyond the summary statistics?
Our unit of analysis is document, not university, and therefore metrics are not weighted when a university has more than one document.
Software for TM/A
Voyant is designed for a specific audience in mind. Therefore, its built-in algorithms serves that particular purpose, not necessarily ours.
Our analysis was limited by the available features in the software, and its built-in algorithms did not allow for much customization.
Approach and perspective
For QDA, we used a mixed method approach, which looked at our data through variables. This certainly is not the only way to study