Three gaps in computational methods for social sciences: A research agenda