There is increasing evidence for age-related heterogeneities in how people use the internet and social media for politics. Convenience samples that do not ensure sufficient variation in this crucial covariate cannot be used to generalize results to any relevant population. In particular, Amazon’s Mechanical Turk is the most prominent source of subjects for these experiments, but a vanishingly small number of these subjects are over 65 years old. The problem is worse for “digital literacy” (strongly correlated with age in practice but which is a theoretically distinct treatment moderator): 100% of subjects recruited via Mechanical Turk are above a threshold of digital literacy below which there are many internet users. We argue for the use of Facebook advertisements to recruit subjects which vary along this dimension, but caution that research with older adults can pose novel ethical concerns.