Automated transcoloration? Skin tone classification algorithms and the prominence of non-epidermic features in the Mexican racial schema

The scale of anti-racist research employing independent measures for skin tone (i.e., classification algorithms and colorimeters) is increasingly growing. However, recent studies reveal that non-epidermic traits (i.e., hair, lips, gender) contribute to the transcoloration –whitening or darkening– of perceived skin tone, signaling that their importance in the racialization process has been underestimated. The magnitude of this effect of non-epidermic features is critical, and the appropriateness of technology-based anti-racist research hinges on it. To what extent are independent measures of skin tone a good approximation to the study of racial discrimination in Mexico? Examining a novel dataset (n=3,000), I compare two different skin color variables (one coded by humans and other by an automated algorithm), finding little agreement about exact estimates but substantial —and statistically significant— agreement on the general direction of the variables. These findings corroborate that non-epidermic features slightly alter people’s perception of skin tone, without causing a major digression from independent skin color assessments. The results validate and encourage research employing these technological methods, which tend to be more accessible and feasible than their alternatives.