2 - (SYM 87) Differential Language Analysis of Participant Narratives and Depression and Anxiety Symptoms During the COVID-19 Pandemic

Background: Studies examining factors associated with depression and anxiety during the COVID-19 pandemic have pointed to financial struggles, social isolation, childcare responsibilities, and prior mental illness. These studies, however, have relied on self-report questionnaires. Few studies have examined linguistic markers of depression or anxiety during the pandemic, although pre-pandemic research on this question has found higher use of negative emotion words and first person singular pronouns in depressed groups. To date, this research has primarily relied on closed vocabulary approaches using predefined word lists. New, data-driven approaches, such as Differential Language Analysis (DLA), may reveal more nuanced connections between language and mental health. In this study, we used DLA to examine whether language features of autobiographical memory narratives predict depression and anxiety symptoms during the pandemic.

Methods: Participants (n = 607), recruited in 2020 and 2021 via Prolific to increase accessibility and sample diversity, completed the Depression Anxiety Stress Scales-21 (DASS) and wrote a narrative response to the prompt, “Tell us how COVID-19 has impacted your life and what this situation means to you. What are you feeling and experiencing in this situation?” We extracted language features using DLA with ten-fold cross validation, in which the predictive model was trained on 90% of the data and tested on the remaining 10%. Resulting Pearson r values represent the correlation between the model’s predicted DASS scores based on participants’ narratives and their actual scores.

Results: Preliminary results indicate that the model demonstrated significant predictive power (R = .40, p < .001). Words and phrases associated with higher DASS symptoms included “anxious”, “anxiety”, “stressed”, “mental”, “health”, “feels”, “feeling”, and “I feel like”. Words and phrases associated with lower symptoms included “home”, “family”, “working”, “from home”, and “we”. Pearson r values for individual words and phrases ranged from -.17 to .19 (all p’s < .05).

Discussion: Using novel, data-driven techniques, this study provides evidence of a connection between language and mental health during the pandemic. Largely consistent with past research, words related to anxiety and feelings appear to be markers of depression and anxiety, whereas words related to social support may reflect better mental health. Demographic variables, including race and sex, will be examined in the final model and implications will be discussed.