They missed the sampling effect across ages. I've seen it in action quite a few times already. Prior: Conditional on Having a fitbit, the 70 yr old is a larger

They missed the sampling effect across ages. I've seen it in action quite a few times already. Prior: Conditional on Having a fitbit, the 70 yr old is a larger deviant from average health, than the 30 yr old. That effect could easily explain away the falling RHR with older age. The population in that age group is getting more healthy by self-sampling. The proposed solution in the article (the use of beta-blockers) seems inadequate. What proportion of the population and the Fitbit-users is on those medicines? (Article does mention some stats in the multicountry comparison.) As always the bayesian/multilevel approach could help: buying a Fitbit is a decision with information on your health-status in itself, model it likewise. Add information on weigth, BMI, activities per week etc. as indicators of in which class of users you fall (the uberfit, healthy, the aspiring, ...). Then use the predicted RHR per age instead of the data itself.