The use of topic modeling with latent Dirichlet analysis with open-ended survey items.

Open-ended questions are a common component in surveys and questionnaires that are used in the social sciences. Such items can provide researchers with insights into respondents’ attitudes and opinions that cannot be easily gleaned from closed-response items based on a traditional Likert-type response format. However, the use of open-ended items can also be associated with a set of analytic problems, particularly in terms of identifying coherent themes that are supported by the data. Topic models present the researcher with a statistically based tool for identifying underlying topics within text, based on how words group together, much in the way that factor analysis uses correlations among observed variables to identify potential latent variables. The current study demonstrates the use of topic models with open-ended response items in order to illustrate how the resulting topics can both provide insights into respondents’ attitudes and create variables that can be used in data analyses with closed-ended item responses. A full illustration of topic modeling in this context is given and discussion of the utility of these models is provided. (PsycINFO Database Record (c) 2018 APA, all rights reserved)