Is Rosie Really Riveting? Measuring occupation classification bias in the 1940 US Census
-
SeriesResearch Master Defense
-
SpeakerDiego Dabed Sitnisky
-
LocationOnline
-
Date and time
August 31, 2021
11:00 - 12:00
I measure gender bias in occupation classifications on the 1940 US Census. To do so I develop a deep learning classification algorithm that uses pretrained word embeddings and socio-demographic characteristics of the individuals. I find that the algorithm learns to classify woman more frequently into non specific occupations and together with nurses and housework. Moreover, the algorithm is less likely to classify woman into unpaid farm related occupations and men into paid farm related occupations. Further, I evaluate FastText as a job title to occupation classification algorithm.