Gendered Signals in Adolescents’ Writing: Embedding-Based Index and Later-Life Outcomes for a 1958 UK Cohort

• In Event: Aging and the Life Course Section Roundtables
  In Refereed Roundtable (90 minute): 3: Life Course Experiences and Exposures and Long-Term Health Trajectories

Abstract

Using the 1958 British National Child Development Study, I construct a language-based index of gender atypicality from 10,507 age-11 essays imagining life at 25 and link it to adult trajectories. After masking explicit gender cues with a constrained large language model, essays are embedded with a transformer and used to train a sex classifier; the calibrated probability of the non-true sex defines a continuous atypicality score. Adjusted regression models show that higher atypicality predicts greater likelihood of managerial/technical class at ages 33 and 50, slightly poorer midlife health and wellbeing, lower odds of ever marrying, and higher odds of ever having a same-sex partner, with modest, sex-differentiated effect sizes.

Author

• Zerui Tian, University of Oxford