Topic models: a novel method for modeling couple and family text data

J Fam Psychol. 2012 Oct;26(5):816-27. doi: 10.1037/a0029607. Epub 2012 Aug 13.


Couple and family researchers often collect open-ended linguistic data-either through free-response questionnaire items, or transcripts of interviews or therapy sessions. Because participants' responses are not forced into a set number of categories, text-based data can be very rich and revealing of psychological processes. At the same time, it is highly unstructured and challenging to analyze. Within family psychology, analyzing text data typically means applying a coding system, which can quantify text data but also has several limitations, including the time needed for coding, difficulties with interrater reliability, and defining a priori what should be coded. The current article presents an alternative method for analyzing text data called topic models (Steyvers & Griffiths, 2006), which has not yet been applied within couple and family psychology. Topic models have similarities to factor analysis and cluster analysis in that they identify underlying clusters of words with semantic similarities (i.e., the "topics"). In the present article, a nontechnical introduction to topic models is provided, highlighting how these models can be used for text exploration and indexing (e.g., quickly locating text passages that share semantic meaning) and how output from topic models can be used to predict behavioral codes or other types of outcomes. Throughout the article, a collection of transcripts from a large couple-therapy trial (Christensen et al., 2004) is used as example data to highlight potential applications. Practical resources for learning more about topic models and how to apply them are discussed.

Publication types

  • Research Support, N.I.H., Extramural

MeSH terms

  • Couples Therapy / instrumentation
  • Couples Therapy / methods*
  • Humans
  • Models, Psychological*
  • Psycholinguistics / methods*
  • Qualitative Research*