Demonstration of Structural Topic Modeling on Charter School Closures in 2019: Quantitative Text Analysis With Machine Learning

While evidence often exists in local newspapers, Facebook pages, and on other platforms, a lack of centralization means that researchers looking to determine the causes of school closure suffer from the unenviable task of manually hunting for data. Worse still, once they collect the texts, researchers need to sift through them to determine the underlining causes for closure. This sifting leads to a variety of issues related to human error. This paper demonstrates the efficacy of using a Structural Topic Model (STM) to automate this last step, reduce human bias, and save time. Topic Modeling is a machine learning technique that builds on a base of artificial intelligence research that seeks to automate complex meta-cognitive tasks. This method is new to the education space, but the paper aims at demonstrating the potential uses by leveraging closure data for the 2018-19 school year. After testing this method on the 2018-19 school year, the researcher determined that the top two reasons for charter school closure at the end of 2019 were financial fraud and low academic performance.