The following algorithm was used to estimate the respective proportions of high internalizingand high externalizing symptoms in the U.S. adolescent population that would not have occurredin each of the four counterfactual reductions in social media proposed.1. Using our adjusted multinomial logistic regression model, we estimated the predictedprobability of experiencing each outcome – high internalizing problems only, highexternalizing problems only, both problems, or neither – for each respondent in our dataset.2. For each participant, we added the probability of high internalizing only to the probability ofboth internalizing and externalizing to estimate to the total predicted probability of thatparticipant experiencing high internalizing symptoms. Similarly, we computed for eachparticipant the predicted total probability of externalizing problems.3. We multiplied these predicted probabilities by participants' respective survey weights, andsummed over all participants. This produced an estimate of the expected total number ofinternalizing and externalizing cases in the real U.S. adolescent population.4. We then built a new dataset where participants’ social media use was reassigned to match theappropriate counterfactual scenario. For example, for scenario 1, participants who reportedmore than 6 hours of social media use per day were reassigned to a value of at least 3 but nogreater than 6 hours per day of social media use.5. We then re-estimated the predicted probabilities and case counts using the new,counterfactual dataset by repeating steps 1-3. This produced an estimate of the expected totalnumber of internalizing and externalizing cases in a counterfactual U.S. adolescentpopulation.6. We took the difference between estimated case counts in the counterfactual and realpopulations. This estimated how many cases would be eliminated in the counterfactualscenario.7. Finally, we then divided this difference by the estimated case count in the real population.This estimated the proportion of cases that would be eliminated in the counterfactualscenario.