Digital Qualitative Content Analysis

The core analytic logic — building a codebook, systematically coding text, interpreting meaning — is the same. What differs is the data context: digital texts are produced on platforms with specific affordances (likes, retweets, hashtags), may be anonymous or pseudonymous, are often asynchronous, and can be very large. Digital QCA therefore adds steps for ethical data collection, platform-aware interpretation, and corpus sampling that standard QCA guidance does not fully address.

Computational methods (topic modelling, sentiment analysis, NLP pipelines) prioritise scale and automation but treat meaning statistically. Digital QCA prioritises interpretive depth: a human researcher reads segments, applies theoretically grounded categories, and explains what the content means — not just how often words appear. Mixed-methods designs can combine both: computational methods to sample or pre-code large corpora, followed by digital QCA for interpretive depth.

Reporting intercoder reliability (e.g., Cohen's kappa or percentage agreement) strengthens credibility, especially when multiple coders apply the same codebook. However, in purely interpretive or hermeneutic variants of qualitative content analysis, some scholars argue that consensus through discussion and codebook refinement matters more than a reliability coefficient. Either way, the decision must be made explicit and justified.

Public visibility does not automatically imply consent for research use. Ethical practice requires assessing whether participants could reasonably expect their posts to be studied, whether verbatim quotes are identifiable, and whether platform terms of service permit data collection. Pseudonymisation of usernames, IRB or ethics committee review, and adherence to the Association of Internet Researchers (AoIR) ethical guidelines are standard expectations.

There is no single rule; the appropriate corpus size depends on the research question and depth of analysis. For deep interpretive work, purposively sampled corpora of 50–500 posts or threads are common. For broader descriptive studies, several thousand coded units may be used. What matters is that sampling is principled, documented, and sufficient for the categories in the codebook to be meaningfully populated.