Experimental Design of the Social Media Deactivation Study

Meta selected a stratified random sample of U.S. users aged 18 or older who had logged in at least once during the past month. The invitations were placed at the top of users’ news feeds from August 31 to September 12. Users who clicked on the invitation were informed about the study and asked two key questions: what weekly payment they would accept to deactivate their account for either one week or six weeks. Those agreeing to deactivate for $25 per week and providing consent were then directed to complete a short enrollment survey administered by the National Opinion Research Corporation (NORC)^[1].

After completing the baseline survey, participants were randomly assigned into one of two groups. The randomization was stratified across 36 strata, defined by indicators such as residence in an election swing state, average daily platform use over the previous 30 days, self-reported political party identification, and race. Approximately 27% of participants were placed in the treatment group (Deactivation Group) that was offered a payment of $150, contingent on not logging into the focal platform (either Facebook or Instagram) for six weeks. The remaining 73% formed the Control Group, which received $25 and were asked to deactivate for only one week. This design ensured that all participants perceived themselves as part of a study involving deactivation, thereby reducing the risks of experimenter demand effects and differential attrition^[1].

The experiment was carefully scheduled. Once randomized, Meta deactivated the participants’ focal platform accounts starting on September 23. The Control Group’s accounts were reactivated on September 30, while the accounts of those in the Deactivation Group were reactivated on November 4. This deliberate timing created a clear comparison between a longer deactivation period (six weeks) and a shorter one-week deactivation. In addition to the deactivation incentive, participants were paid $5 for completing the baseline survey and at least $20 for the endline survey. Some participants also received additional compensation for allowing passive tracking of their smartphone app and web browser use, thereby providing detailed behavioral data to complement the self-reported outcomes^[1].

The primary outcome measure of the study was the users’ emotional state. Both baseline and endline surveys included three specific questions addressing how often respondents felt happy, depressed, or anxious during the previous four weeks. The available response options were 'All of the time,' 'Often,' 'Sometimes,' 'Rarely,' and 'Never', which were coded numerically and then standardized to produce the variables ‘happy’, ‘depressed’ and ‘anxious’. For the latter two, responses were re-signed (multiplied by -1) so that higher values indicated a more positive emotional state. An emotional state index was then created by averaging these three variables and re-standardizing the index. This careful measurement allowed the researchers to precisely quantify changes in emotional well-being as a result of the deactivation intervention^[1].

The study design was constructed to minimize potential biases and confounding factors. For example, by offering a short deactivation period for the control group as well, the design isolated monetary and duration differences as the key distinguishing features between the two groups. Pre-specified procedures, including controls based on baseline measures and demographic variables, were in place to adjust for any slight imbalances in the sample (e.g., a minor imbalance in baseline emotional state observed among Facebook users). Moreover, the researchers used an instrumental variables approach to account for imperfect compliance with the deactivation protocol. Passive tracking data were also collected for a subset of participants to assess how deactivation affected overall mobile application use and whether time freed from using Facebook or Instagram was reallocated to other apps^[1].