In her Shoes: Gendered Labelling in Crowdsourced Safety Perceptions Data from India

In recent years, a proliferation of women’s safety mobile applications have emerged in India that crowdsource street safety perceptions to generate ‘safety maps’ used by policy makers for urban design and academics for studying mobility patterns. Men and women’s differential access to information and communication technologies (ICTs), however, and the distinctions between their social and cultural subjective experiences may mitigate the value of crowdsourced safety perceptions data and the predictive ability of machine learning (ML) models utilizing such data. We explore this by collecting and analyzing primary data on safety perceptions from New Delhi, India. Our curated dataset consists of streetviews covering a wide range of neighborhoods for which we obtain subjective safety ratings from both female and male respondents. Simulation experiments where varying the proportion of ratings from each gender are assumed missing demonstrate that the predictive ability of standard ML techniques relies crucially on the distribution of data producers. We find that obtaining large amounts of crowdsourced safety labels from male respondents for predicting female safety perceptions is inefficient in a number of scenarios and even undesirable in others. Detailed comparisons between female and male respondents’ data demonstrate significant gender differences in safety perceptions and associated vocabularies. Our results have important implications for the design of platforms relying on crowdsourced data and the insights generated from them.