An alternative story could be that the women’s presented appearance online may have changed more than men’s and that real appearance changes could weaken the correlation between the paper’s stored photo-based beauty score and what instructors actually saw live. Maybe woman changed grooming effort more than men, or the effects of fashion trends that explicitly drove the woman towards less attractive styles etc.

if that mismatch increased more for women than men, the estimated “beauty premium” for women could fall even without any change in teachers’ discriminatory behavior. The paper just assumes the attractiveness stayed constant during the period, but seems to have had no data to verify this.