Comparing grounded theory and topic modeling: Extreme divergence or unlikely convergence?

Abstract

Researchers in information science and related areas have developed various methods for analyzing textual data, such as survey responses. This article describes the application of analysis methods from two distinct fields, one method from interpretive social science and one method from statistical machine learning, to the same survey data. The results show that the two analyses produce some similar and some complementary insights about the phenomenon of interest, in this case, nonuse of social media. We compare both the processes of conducting these analyses and the results they produce to derive insights about each method's unique advantages and drawbacks, as well as the broader roles that these methods play in the respective fields where they are often used. These insights allow us to make more informed decisions about the tradeoffs in choosing different methods for analyzing textual data. Furthermore, this comparison suggests ways that such methods might be combined in novel and compelling ways.

Researching Culture through Big Data: Computational Engineering and the Human and Social Sciences

“…Baumer and others detected several points of convergence and divergence. Among other results they found that, even though statistical topic modeling, unlike grounded theory, ignores contextual information, it is able to "identify patterns that, at some level, align with those found by human researchers" (Baumer et al 2017(Baumer et al , p. 1406. These authors argue that the rigorous combination of these different approaches demands that researchers who use them keep in mind the different epistemological traditions from which they come: grounded theory draws on the interpretivist tradition, while most computational techniques derive from the positivist tradition, which assumes the possibility of an objective physical and social world.…”

“…[and] the most pressing questions in cultural sociology concern change over time" (Bail 2014, p. 474). Several others welcome the possibilities opened up by incorporating computational methods into the sociology of culture (DiMaggio et al 2013;Bail 2014;Muller et al 2016;Abramson et al 2017;Baumer et al 2017;Nelson 2017). They agree in arguing for the importance of incorporating the 'topic modeling' method in content analysis (e.g., newspaper and television transcript archives), for its potential usefulness in organizing, searching and understanding data sets on large-scale cultural artefacts which span an extended period of time.…”

Voices from the far right: a text analysis of Swedish parliamentary debates

“…To do this we use a probabilistic topic model approach, similar in spirit to that of Quinn et al (2010). Topic models, such as the popular Latent Dirichlet Allocation by Blei et al (2003), are a way of modeling latent semantic themes without any manual classification, but with similarities to qualitative text methods such as grounded theory Baumer et al (2017).…”

A nested computational social science approach for deep-narrative analysis in energy policy research

“…It centres around probabilistic topic modelling that distributes vocabulary over probability distribution. The high probability words in each distribution can be readily interpreted as recognisable themes, and are thus referred to as "topics" (Baumer et al, 2017;Grimmer & Stewart, 2013;Jockers, 2013). There are many methods of topic modelling, of which Latent Dirichlet Allocation (LDA) is the most popular in the field.…”

“…The novel nested approach presented here contributes to lessening directional biases in narrative-driven policy analysis. The innovation of the study lies in the nested treatment of two epistemologically parallel but methodologically distinct concepts (Baumer, Mimno, Guha, Quan, & Gay, 2017) of topic modelling (TM) and grounded theory (GT). These parallel methods are used to complement each other, to reduce directional bias in energy policy applications.…”