Poster #80 - Text Analysis Approaches to Predicting Depression in Early Adolescent Girls

• In Event: 2-002 - Special Poster Session 05 with Continental Breakfast Reception
  In Poster Session: PS 05 - Interdisciplinary Section

Integrative Statement

It is well-documented that girls who mature earlier than their peers are at particular risk for myriad psychosocial problems including depression. Consequently, there is considerable stake in identifying the factors that contribute to this increased vulnerability to depression. Linguistic Inquiry and Word Count (LIWC) has been used to characterize writing styles associated with depression (e.g. Tackman et al., 2018). While lexicon-driven methods like LIWC are functional, topic modeling using contemporary machine learning techniques (e.g. Latent Dirichlet Allocation; LDA) can be used to improve prediction of depression in early adolescent girls.

The present research employed a quantitative text analysis approach to examine girls’ reported experiences during puberty. This research was guided by the following questions: 1) What central themes emerged across documents? 2) Do early-maturing girls meaningfully differ from peers in terms of writing content and affect? 3) Which themes and/or categories are predictive of depression?

Participants (N= 125 girls, M age= 11.61) engaged in four consecutive days of journal-style expressive writing focused on puberty. The prompts cued participants for changes associated with the pubertal transitions (i.e. social, family, and physical). Participants completed a questionnaire battery before the writing program and again approximately four months later.

Two methods of text analysis were utilized: LIWC (a closed-vocabulary analysis method) and LDA (an open-vocabulary method). OLS regressions were fit between LIWC categories, LDA topics, and depression.

K-fold cross-validation was used to select the optimal number of topics to derive from the text in LDA analysis. Ten topics were derived: 1) School – Friends; 2) School – Peers; 3) Family – General; 4) Relationship quality; 5) Puberty – General; 6) Comparative change; 7) Perceptions of change; 8) Physical changes (height and skin-focused); 9) Physical changes (breast-focused); 10) Physical changes (menstruation-focused). Early maturing girls had a positive relationship with writing about menstruation-focused physical changes (r = 0.21).

LIWC analysis indicated that early-maturing girls did not differ from their peers in terms of affect. There were no meaningful content category differences between early-maturing girls and peers.

Writing about menstruation-focused physical changes was positively correlated with depression at both time 1 and 2. Regression results indicated that writing about menstruation-focused physical changes predicted time 2 depression even when controlling for time 1 depression, pubertal development, and age. While several LIWC categories were correlated with depression (e.g. sad words, filler words), no LIWC category significantly predicted time 2 depression after controlling for time 1 depression.

These results suggest that girls who focus specifically on the menstrual changes associated with puberty are more at-risk for subsequent depression. Early pubertal timing was associated with focusing on these menstrual changes. Present findings may be used to inform education and intervention programs for pubertal girls. Normalizing menstrual changes may reduce the psychological distress that seems to come with reaching menarche before peers. These results also demonstrate the added predictive value of topic modeling over using only lexicon-driven methods.

Authors

• Mary Kate Koch, Cornell University
  Presenting Author

• Gabrielle Kubi, Cornell University
  Non-Presenting Author

• Jane Mendle, Cornell University
  Non-Presenting Author