Search

Deep learning techniques can use public data such as facial photographs to predict sensitive personal information, but little is known about what information contributes to the predictive success of these techniques. This lack of knowledge limits both the public's ability to protect against revealing unintended information as well as the scientific utility of deep learning results. We combine convolutional neural networks, heat maps, facial expression coding, and classification of identifiable features such as masculinity and attractiveness in our study of political ideology in 3323 Danes. Predictive accuracy from the neural network was 61% in each gender. Model-predicted ideology correlated with aspects of both facial expressions (happiness vs neutrality) and morphology (specifically, attractiveness in females). Heat maps highlighted the informativeness of areas both on and off the face, pointing to methodological refinements and the need for future research to better understand the significance of certain facial areas. [ABSTRACT FROM AUTHOR]

In this study we provide new evidence on the much-discussed effect of education on political participation by utilizing the quasi-experiment of twinning. By looking at the relationship between education and participation within monozygotic (MZ) twin pairs we are able to circumvent traditional sources of confounding of the relationship rooted in genes and early life family environment because MZ twins share both. The results of within-twin pair analyses based on surveys from the United States, Denmark and Sweden show that while the relationship between education and political participation is highly confounded by genes and/or familial environment in all three countries, a positive impact remains of years of education in the US and of high school completion in Denmark. No effect is found in Sweden. Robustness checks suggest that the observed effect is not confounded by within-twin pair differences in prenatal environment nor differential treatment during childhood, and, if anything, that it most likely constitutes a lower bound estimate. [ABSTRACT FROM AUTHOR]