Demographic Inference in the Digital Age: Using Neural Networks to Assess Gender and Ethnicity at Scale

Abstract

Gender and ethnicity are increasingly studied topics within I-O psychology, helpful for understanding the composition of collectives, experiences of marginalized group members, and differences in outcomes between demographics and capturing diversity at higher levels. However, the absence of explicit, structured, demographic information online makes applying these research questions to Big Data sources challenging. We highlight how deep neural networks can be used to infer demographics based on people's names, which are commonly found online (e.g., social media profiles, employee pages, and membership rosters), using broad international data to train and evaluate the effectiveness of these models and find that validity coefficients meet minimum reliability thresholds at the individual level (r_gender  =  .91, r_ethnicity  =  .80) highlighting their ability to contextualize and facilitate Big Data research. Using empirical data extracted from databases, websites, and mobile apps, we highlight how these models can be applied to large organizational data sets by presenting illustrative demonstrations of research questions that incorporate the information provided by the model. To promote broader usage, we offer an online application to infer demographics from names without requiring advanced programming knowledge.