Using structural topic modeling to understand ethnicity-related narratives
Rachel Jacobson, Dulce Wilkinson Westberg, Edward Chou, Moin Syed, Sara J Weston
*Journal of personality*
Abstract
OBJECTIVE: Narrative identity is a promising approach for understanding the content of individuals' ethnic identities but can be limited by the time-intensive nature of human coding and the reliance on preestablished coding systems. BACKGROUND: The aim of our preregistered study is to elucidate the content of individuals' ethnicity-related experiences using a novel statistical approach. METHOD: We applied structural topic modeling (STM), a natural language processing tool, to narratives written by an ethnically diverse sample of 1149 young adults about a moment they felt aware of their ethnicity. RESULTS: We identified 14 topics within ethnicity narratives and analyzed how each topic related to both the participant's ethnicity and the human-coded themes of agency and communion. For example, the topic Gained perspective of structural inequality was associated with greater agency, whereas Peer dynamics was associated with greater communion. Ethnic/cultural celebration was associated with both. CONCLUSIONS: This study introduces STM as a useful tool for extracting topic content in narrative data and demonstrates how the multi-method assessment of ethnicity narratives provides greater insight into the content of ethnic experiences. These findings contribute to our understanding of contextualized aspects of personality, including the innovative ways we might examine them.
What we did
Narrative identity research asks people to tell stories about their lives. Reading and coding those stories by hand takes a long time, and it usually relies on coding systems chosen in advance. We wanted to see what structural topic modeling (STM), a text-analysis method that finds clusters of words that tend to appear together, could add. In our preregistered study, 1,149 ethnically diverse young adults wrote about a moment when they felt aware of their ethnicity.
What we found
STM identified 14 topics in the narratives. We then asked how each topic related to the writer’s ethnicity and to two themes that human coders had already rated, agency and communion. In other words, we compared what the algorithm found with what people found. Some topics lined up with those themes. For example, a topic we labeled Gained perspective of structural inequality was associated with greater agency, and Peer dynamics was associated with greater communion. Ethnic/cultural celebration was associated with both.
Pairing STM with human coding gave us greater insight into the content of ethnic experiences than either would alone.
This summary is based on the published abstract.