Thematic Analysis

One of the most widely used research methods in the social sciences is also one of the most misunderstood.

When a researcher sits down with a stack of interview transcripts and starts pulling themes out of them, what they are doing looks, from a distance, like sorting. They read the transcripts. They notice patterns. They group the patterns into themes. They write up the themes. The whole thing sounds straightforward, almost clerical, and for decades it was sometimes reported that way in journals. Two researchers named Virginia Braun and Victoria Clarke pointed out that this description obscures almost everything that matters about the method. Themes do not exist in the data waiting to be found. They are constructed by the researcher, using the data, in light of a question. Two careful researchers analysing the same transcripts can and will produce different themes, and this is not a failure of the method, it is how the method works.

This matters because thematic analysis is everywhere, including in AI research about education. Papers claim to identify themes from user interviews, or from student responses, or from teacher feedback, and present the themes as though they were discovered rather than made. When an AI assists with this analysis, the situation gets stranger, because the machine is now part of the construction process, bringing its own patterns of attention to the material, usually without anyone being explicit about what those patterns are. Good thematic analysis is reflexive, which is a word meaning that the researcher keeps noticing their own role in producing the findings and is honest about it. Most AI-assisted analysis is not reflexive at all, because the assumption is that the machine is just processing, and processing is thought to be neutral, which it is not. When a research finding was described to you as what the data showed, how confident are you that the data showed it, as opposed to being read in a particular way?

Last week we looked at decisions where some things must be true before others count. This week we look at one of the most common research methods in the social sciences and at what really happens inside it.