Over the past decade, social media platforms have been scrutinized for their potential impact on mental health. Among the general public, claims about social media harms have been widely publicized in both television documentaries [1] and whistleblower accounts [2]. Within the academic literature, multiple studies have also linked social media usage to symptoms of depression [3,4], psychological distress [5], poorer well-being [5,6], and lower self-esteem [7].

Two theories have been proposed to explain why social media platforms may compromise mental health [1,2]. First, such platforms allow users to compare themselves with celebrities or peers whose web-based posts portray more ideal lives than those of typical users [8,9]. This form of upward social comparison may result in users feeling worse about themselves, placing them at risk for poorer mental health [8,9]. Second, social media platforms are designed to draw users’ attention for as long as possible [10,11]. In turn, this allure may result in excessive social media consumption, again impairing well-being [10,11].

To address public concerns about these social media harms, app developers have introduced digital well-being features to help users manage their engagement [12,13]. Nonetheless, it remains unclear (1) whether users know or use these features and (2) whether the use of these features predicts better psychological well-being. Consequently, this study examines these questions by focusing on Facebook as a case study.

With 2.9 billion users worldwide, Facebook is the most widely used social networking platform in the world [14]. Given its popularity, it has also been the focus of most research studies that document the link between social media usage and poorer mental health [7,9,15]. As a result, Facebook developers consulted mental health experts and launched a series of digital well-being features, with the high-level goal of making subsequent Facebook usage “intentional, positive and inspiring” [16].

Facebook’s digital well-being features broadly address the two proposed theories for social media harms. First, to minimize the amount of social comparisons, several features allow users to curate the content that they see. For example, the Unfollow option allows users to hide posts from selected friends, pages, or groups, while the Snooze option hides these posts for a 30-day duration [17]. Further, the Off-Facebook Activity feature allows users to customize how the platform integrates information from external apps to customize their feeds [17].

Based on prior surveys, content curation features seem to be adopted when users want to avoid friends’ boastings, inappropriate posts (eg, racist content), content that they disagree with (eg, on account of political ideology), or excessive and irrelevant posts [18-21]. In turn, deploying these features can cause users to feel unburdened [22]. Consequently, we sought to examine whether the adoption of these features predicts better mental health.

In the second category, a separate set of digital well-being features enables users to monitor their usage patterns and curb excessive use. For example, the Your Time on Facebook feature displays the amount of time that a user has spent on Facebook over the past week, while the Set Daily Reminders feature notifies users when a predetermined cutoff has been reached (eg, 45 minutes of Facebook use) [17]. Additionally, the Notification Settings feature allows users to manage the in-app notifications that they receive, minimizing the amount of content that draws the users’ attention.

As we are not aware of any study linking Facebook’s digital well-being tools to mental health, we conducted a cross-sectional survey to address our two primary aims. First, we sought to document the extent to which Facebook users know and use the six outlined features—the (1) Unfollow, (2) Snooze, (3) Off-Facebook Activity, (4) Your Time on Facebook, (5) Set Daily Reminders, and (6) Notification Settings features. Second, we sought to replicate previous findings that linked Facebook usage with poorer mental health and examine whether participants’ use of the well-being features was associated with better outcomes.

The participants were 608 Facebook users who were recruited from Amazon’s web-based panel (Mechanical Turk) in June 2021. All participants met the following eligibility criteria: (1) individuals aged 21 years or older, (2) individuals who were proficient in English, (3) individuals based in the United States, and (4) individuals with a positive track record on the platform (human intelligence task approval rate: >95%; number approved: >500).

Participants gave their written consent in accordance with the Declaration of Helsinki, and were given a nominal sum of US $0.50 upon study completion. This study was approved by the Yale-NUS College Ethics Review Committee (approval number: 2021-CERC-001) and was preregistered on ClinicalTrials.gov (trial number: NCT04967846).

Predictor and outcome variables were measured through a 10-minute survey that was hosted on the Qualtrics website (Qualtrics International Inc) [23]. The questions were written for a seventh-grade reading level and were pilot-tested before this study.

As an assay of mental health, participants completed the 21-item Depression, Anxiety, and Stress Scale (DASS-21) [25]. The DASS-21 has been well validated and widely used, consisting of 7 items for each of the following subscales: depression (eg, “I couldn’t seem to experience any positive feelings at all” and “I found it difficult to work up the initiative to do things”; Cronbach α=.87), anxiety (eg, “I was aware of dryness of my mouth” and “I felt I was close to panic”; Cronbach α=.89), and stress (eg, “I found it hard to wind down” and “I tended to over-react to situations”; Cronbach α=.89). Each item was rated on a 4-point scale (ranging from “0: did not apply to me at all” to “3: applied to me very much or most of the time”), and scores were summed and multiplied by 2.

As part of data cleaning, we first verified that participants had read the questions through two verification items that asked participants to check boxes as instructed (modeled after the widely used CAPTCHA technique on the internet) [26]. Of the 608 participants, 10 (1.6%) failed the verification and were removed from the data set, resulting in a final sample of 598 participants. We then summarized participants’ baseline characteristics by using medians (with IQRs) and counts (with percentages). For count data, error margins for the 95% CIs of proportions were computed by using the prop.test function in R (R Foundation for Statistical Computing).

As the primary analyses, we ran a series of linear regression models, using each DASS-21 subscale score (depression, anxiety, and stress) as an outcome measure. In the first model, we sought to replicate the oft-reported link between one’s duration of Facebook use and poorer mental health [8]. To this end, we entered the number of hours that participants spent using Facebook as a predictor. As the visual inspection of the data revealed a right-skewed distribution, this variable was log-transformed to achieve linearity (model 1).

In the second model, we addressed this study’s primary aim—examining whether the adoption of Facebook’s well-being features predicted better mental health (having controlled for the duration of Facebook use). Correspondingly, model 1 was repeated with 6 additional predictors that coded for the use of each feature (Unfollow, Snooze, Off-Facebook Activity, Your Time on Facebook, Set Daily Reminders, and Notification Settings; model 2). For each predictor, nonusage was coded as “0” and usage was coded as “1.”

Finally, we assessed the robustness of our findings by repeating model 2 with the inclusion of demographic variables (age, gender, race, religion, marital status, education, employment, family income, household size, and living setting; model 3).

Across the models, the type 1 decision-wise error rate was controlled at an level of .05, with adequate statistical power (0.80) for detecting small effect sizes (f²=0.05). All statistical analyses were carried out on SPSS 27 (IBM Corporation) and R version 4.0.3 (R Foundation for Statistical Computing).

Of the 598 participants, 309 (51.6%) were aged <35 years, and slightly over half of the participants (360/598, 60.2%) self-identified as men (Table 1). In terms of baseline Facebook usage, participants reported using the platform for a median of 3 (IQR 1-7) hours each day in the preceding week. Further, 289 (48.3%, 95% CI 44.3%-52.3%) participants accessed Facebook multiple times a day, while 130 (21.7%, 95% CI 18.4%-25%) logged in once a day (Table 1). On Facebook, participants were most likely to view a friend’s photos or to read the news feed (Figure 1).

Of the 598 participants, 580 (97%, 95% CI 95.6%-98.4%) were aware of at least one of Facebook’s well-being features (Figure 2). However, awareness levels differed across features. For example, while 508 (85%, 95% CI 82.1%-87.9%) had heard of the Notification Settings feature, only 259 (43.3%, 95% CI 39.7%-47.3%) of participants knew about the Your Time on Facebook feature.

In terms of usage, of the 598 participants, 332 (55.5%, 95% CI 51.5%-59.5%) had used the Snooze feature, 316 (52.8%, 95% CI 48.7%-56.7%) had used the Off-Facebook Activity Tracker feature, 315 (52.7%, 95% CI 48.6%-56.6%) had used the Your Time on Facebook feature, and 309 (51.7%, 95% CI 47.7%-55.7%) had used the Unfollow feature. Less than half had adjusted Notification Settings (n=260, 43.5%, 95% CI 39.5%-47.5%), and fewer still had used the Set Time Reminder feature (n=104, 17.4%, 95% CI 14.4%-20.4%). Where the repeated use of features was possible, participants were most likely to report using them “sometimes” on an ad hoc basis rather than on a routine basis (based on the median ratings for Snooze, Off-Facebook Activity, Your Time on Facebook, and Unfollow; Figure 2).

For the primary research question, we sought to predict participants’ depression, anxiety, and stress scores as a function of whether they used Facebook’s well-being features.

In this paper, we present the first empirical study to evaluate Facebook’s digital well-being tools. First, echoing prior studies [8], we found that participants who spent more time on Facebook had more symptoms of depression, anxiety, and stress. Accounting for Facebook consumption alone explained one-tenth of the variance (range 8%-13%) in participants’ well-being. Consequently, we examined (1) whether participants used the platform’s digital well-being tools and (2) whether usage was associated with better mental health.

Although most participants (580/598, 97%) knew about Facebook’s well-being tools, each tool was used by only 17.4% (104/598) to 55.5% (332/598) of participants largely on an ad hoc basis. These adoption rates are lower than those of mainstream Facebook features that were introduced much earlier to the platform. For example, an estimated 4 in 5 Facebook users have deployed the Unfriend feature to remove contacts [27], while 3 in 4 have used the Untag feature to remove their name from a photograph [28].

Participants who used either the Notification Settings feature or the Unfollow tools reported fewer symptoms of depression, anxiety, and stress. Conversely, those who used either the Snooze feature or the Off-Facebook Activity feature had higher scores on each of these measures. Finally, there was no evidence that the Your Time on Facebook feature or the Set Daily Reminders feature was associated with well-being. This set of findings was robust and was observed regardless of whether we controlled for participants’ duration of Facebook use or their sociodemographic factors.

Taken together, our findings underscore the complexity of designing social media platforms to optimize user welfare. Of the 6 digital well-being tools we examined, only 2 were associated with a decreased risk for mental health symptoms—(1) a feature for toning down the amount of content that is brought to a user’s attention (Notification Shortcut Bar function) and (2) a feature that allows users to customize their news feeds (Unfollow feature), which, in theory, minimizes the amount of social comparisons made on the platform. Nonetheless, it remains unclear why two other features that supported the customization of news feeds (Snooze and Off-Facebook Activity) predicted a higher risk for mental health symptoms. Further research is thus needed to understand these patterns.

It is noteworthy that we found no significant associations between the use of time-monitoring features (Your Time on Facebook and Set Daily Reminders) and well-being. This finding is counterintuitive because the time spent on Facebook has been linked repeatedly to poorer mental health outcomes (including in this study) [8]. Consequently, most social media developers have incorporated time-monitoring features into their digital well-being programs, allowing users to track how much time they have spent on a platform or set limits on usage (eg, on YouTube, Instagram, Facebook, and TikTok) [13,16,29]. Nonetheless, we found no empirical support for this widely used strategy, consistent with “digital detox” studies reporting that interventions for curbing social media use have a limited impact on mood and well-being [30].

Moving forward, our study has several implications for research and practice. First, it appears that the current well-being measures taken by social media platforms may be insufficient. This begs the question of how digital well-being tools should be designed to maximize users’ benefits. Despite widespread calls for app developers to prioritize their users, there remains limited empirical data for guiding platforms in carrying out this mandate. We thus urge researchers to address this gap, thereby allowing for an evidence-based toolkit of in-app well-being features to be developed.

In reporting our findings, we noted several limitations of our study. First, we chose the design of an epidemiological survey [31,32]. In a new area of research, this allowed us to (1) document baseline adoption rates for digital well-being tools and (2) examine multiple tools at the same time. Nonetheless, correlation does not equate to causation, and our findings need to be followed up with randomized controlled trials. Second, we recruited participants within the general population of internet users; the participant demographics were comparable to that of US Facebook users [33]. Nonetheless, it is possible that stronger effects would be observed in vulnerable groups, such as among individuals with problematic forms of Facebook usage [34] or among adolescents. Further research should thus explore this possibility. Finally, we focused on Facebook because of its widespread popularity. It is currently unclear whether our findings would generalize to other social networking services (eg, Instagram).

In the 2022 State of the Union Address, President Joe Biden called for social media platforms to be held accountable and for companies to pursue users’ benefits over profits [35]. Amid these petitions, there is a need to understand how social media platforms can be designed to optimize users’ well-being. Accordingly, our study provides the first line of evidence that two digital well-being features may be linked to improved mental health. At the same time, we also caution app developers that (1) not all well-being features are alike and (2) certain features could backfire. Moving forward, we urge further research to develop and carefully investigate the impact of digital well-being tools on social media.