Comparing Textual Data Analysis Methods in Science Education Research

Textual data, common in science education research, has traditionally been analysed using qualitative methods. However, the current datafication process of society and the emergence of Education Data Science, including Artificial Intelligence techniques, makes computational data-intensive methods an alternative approach. Through a case study on 223 essays in the context of a future-oriented science education project, we show the comparison of two methods: Thematic Analysis, which is a qualitative method, and Latent Semantic Analysis, which belongs to topic modeling methods. We compared the two analytical processes and results concerning the epistemic quality and method features, deepening the relationship between method, data, and analysers. This work contributes to the discussion of methods from an operational point of view by constructing a table of analysis for comparing methods on the epistemological and methodological levels. Moreover, we show that the nature of textual data collected in the Science Education context requires a balanced exchange between bottom-up and top-down analysis, possible in both qualitative and computational data-intensive methods. The complementary benefits and limits of method features suggest the possibility of integrating the two approaches, allowing a more extensive analysis of textual datasets while maintaining the necessary attention to contextual details.