Selecting significant respondents from large audience datasets: The case of the World Hobbit Project

International projects, online questionnaires, or data mining techniques now allow audience researchers to gather very large and complex datasets. But whilst data collection capacity is hugely growing, qualitative analysis, conversely, becomes increasingly difficult to conduct. In this paper, I suggest a strategy that might allow the researcher to manage this complexity. The World Hobbit Project dataset (36,109 cases), including answers to both closed and open-ended questions, was used for this purpose. The strategy proposed here is based on between-methods sequential triangulation, and tries to combine statistical techniques (k-means clustering) with textual analysis. K-means clustering permitted to reduce data to a small number of ideal-typical respondents: the ‘average spectator’, the ‘die-hard fan’, the ‘cultured spectator’, the ‘alternative spectator’. These clusters are the outcome of a crossvalidation process. Textual responses corresponding to each cluster, in fact, were also analyzed using a quali-qualitative approach, in order to both refine the clusters and identify meaningful discourse patterns. The methodological mix proposed can be used with confidence, since it proved to yield reliable results.