Does Deactivating Facebook or Instagram Improve Users’ Emotional Well-Being?

Allcott et al. (2024) ask whether deactivating Facebook or Instagram improves users’ emotional well-being during the 2020 U.S. election period. They analyze survey data from more than 30,000 adult users who were randomly assigned to deactivate for six weeks (treatment) or one week (control). They find that Facebook deactivation increased an emotional state index by about 0.060 standard deviations, while Instagram increased it by about 0.041 standard deviations. They report that effects were driven mainly by adults over 35 for Facebook and young women for Instagram. Their results suggest that stepping away from these platforms modestly but meaningfully improves well-being.

Why This Article Was Selected for The Policy Scientist

This article addresses a timely question amid growing concern about the psychological effects of social media, particularly during high-salience political periods. Its policy relevance is broad: platform regulation, youth mental health, and election information environments all depend on credible estimates of causal effects. The authors deploy a very large, pre-registered randomized controlled trial with strong compliance measurement using platform data rather than self-reports. The dataset is unusually robust and permits credible causal inference. Although the findings come from the United States, they may generalize to other advanced democracies with similar social media ecosystems. The statistical methods are rigorous and well-aligned with best practice in causal research.

Full Citation and Link to Article

Allcott, H., Gentzkow, M., Wittenbrink, B., Cisneros, J. C., Crespo-Tenorio, A., Dimmery, D., Freelon, D., González-Bailón, S., Guess, A. M., Kim, Y. M., Lazer, D., Malhotra, N., Moehler, D., Nair-Desai, S., Nyhan, B., Pan, J., Settle, J., Thorson, E., Tromble, R., Velasco Rivera, C., Wilkins, A., Wojcieszak, M., Franco, A., Kiewiet de Jonge, C., Mason, W., Stroud, N. J., & Tucker, J. A. (2025). The effect of deactivating Facebook and Instagram on users’ emotional state. American Economic Journal: Economic Policy. Advance online publication. https://doi.org/10.1257/pol.20240806

Central Research Question

The authors ask whether deactivating Facebook or Instagram for six weeks causally improves users’ emotional well-being in the run-up to the 2020 U.S. presidential election. Their question is motivated by heightened public concern regarding social media’s psychological effects, the prevalence of political content during elections, and the limits of correlational evidence. They focus on whether randomized abstention from social media yields measurable improvements in self-reported emotional state (happiness, anxiety, and depression), how large those effects are, and whether they vary across demographic groups. They also seek to contextualize the magnitude of any observed effects and contrast experimental estimates with non-experimental methods that dominate the literature.

Previous Literature

A large literature attempts to link social media use with psychological well-being, but most evidence is correlational, based on cross-sectional survey associations or longitudinal correlations. Meta-analyses find small but heterogeneous correlations between social media use and anxiety, depression, and loneliness, but experts disagree about causal interpretation. Several scholars hypothesize that unfavorable social comparisons and exposure to political content reduce well-being, while others emphasize social connection benefits. Experimental work is sparse, often involving small samples, short abstention periods (one to four weeks), reliance on self-reported compliance, and limited power for subgroup analysis. Only one major experiment before this study (Allcott et al. 2020) tested Facebook deactivation at scale, but even that project had fewer participants, shorter abstention, weaker compliance measurement, and lacked an election context. Additional quasi-experimental evidence (e.g., staggered university rollouts) has suggested negative mental-health effects from Facebook, but those studies reflect earlier platform architectures and user demographics. In short, the dominant gap is causal identification at realistic scale, especially for Instagram, which has been widely scrutinized for its effects on younger women but had not previously been tested in a large randomized setting.

Data

Meta drew large national samples of U.S. Facebook and Instagram users aged 18+ who logged in at least once in the prior month and averaged at least 15 minutes of daily use. Roughly 10.6 million Facebook users and 2.6 million Instagram users were invited; of those, 19,857 Facebook users and 15,585 Instagram users met the pre-registered criteria and completed the baseline survey. Participants were randomly assigned to a deactivation treatment (no access for six weeks) or control (no access for one week). Surveys measured emotional state using three items (happy, depressed, anxious) coded on a five-point frequency scale from “never” to “all of the time,” then standardized. Participants also completed extensive political outcome batteries, but those are analyzed separately in Allcott et al. (2024). The study included passive smartphone app metering for a subset of participants, allowing direct measurement of substitution behavior—an unusually strong feature relative to prior studies. Compliance data were not self-reported; instead, Meta tracked whether accounts logged in and viewed content. This yields higher measurement fidelity than any previous experimental design in the field. Attrition was 10–13 percent, slightly higher in control groups, but below typical field experiment norms and analyzed through robustness checks.

Methods

The study employed a pre-registered randomized controlled trial with two parallel experiments (one for Facebook, one for Instagram). Treatment participants were paid not to log in for six weeks, while control participants were paid not to log in for one week. Randomization was stratified, and analyses used instrumental variables to estimate complier average treatment effects because compliance, while high, was not perfect. Compliance was defined as the proportional reduction in platform use relative to control, constructed from internal usage data. The primary emotional state outcome was an index of the three well-being items, standardized within the control group. The authors used lasso-selected controls, stratification indicators, and survey weights to increase precision and adjust for observable imbalance. They also applied multiple hypothesis testing adjustments because emotional state was classified as a secondary outcome in a broader pre-analysis plan primarily focused on political outcomes. Subgroup analysis examined moderators such as age, gender, and baseline use, though only some moderators were pre-registered. Finally, the authors compared experimental estimates to non-experimental cross-sectional and longitudinal designs to illustrate direction and magnitude of observational bias.

Findings/Size Effects

Deactivation improved emotional well-being modestly but consistently. Facebook deactivation improved the emotional state index by 0.060 standard deviations (p<0.001). Instagram deactivation improved the emotional state index by 0.041 standard deviations (p=0.016), though significance did not survive the strictest pre-registered multiple-testing adjustments. Component measures for Facebook showed improvements of 0.064 SD (happiness), 0.039 SD (depression, signed positively), and 0.028 SD (anxiety, signed positively). Component measures for Instagram showed 0.044 SD, 0.026 SD, and 0.024 SD improvements respectively.

Substitution patterns were revealing: Instagram deactivation led to no significant reduction in total mobile app time, while Facebook deactivation reduced total mobile use by roughly 9 minutes per day. Both platforms’ abstention prompted substitution toward Twitter, TikTok, Snapchat, YouTube, and other apps. This suggests that improvements were not driven by offline time but by platform-specific content or comparison environments.

Heterogeneous effects revealed noteworthy demographic patterns: Facebook effects were concentrated among users aged 35+, whereas Instagram effects were driven by women aged 18–24 (0.111 SD, p=0.002). Effects did not vary significantly by baseline use, baseline emotional state, or political engagement. Benchmarks contextualize magnitudes: effects correspond to a move from the 50th to approximately the 52nd percentile of emotional state distribution; represent roughly 15–22 percent of the average psychological intervention effect size in a major meta-analysis; offset more than half of the well-being drop observed during the election period; and are smaller than earlier quasi-experimental rollout estimates. Comparisons to correlational estimates showed that observational methods yielded biased and sometimes incorrectly signed estimates, reinforcing the necessity of randomized designs.

Conclusion

This study makes a distinct contribution by delivering the largest randomized test of social media abstention to date, the first large-scale Instagram-specific experiment, and the first to operate in a U.S. presidential election context. Its methodological contributions are substantial: rigorous compliance tracking, pre-registration, large sample sizes, internal usage data, and smartphone metering together raise the evidentiary bar beyond most existing work. Substantively, the study provides credible causal evidence that abstention from Facebook and Instagram produces small but meaningful improvements in emotional state. The demographic heterogeneity findings are especially relevant given public concern about youth and gender differences in vulnerability, though the results neither confirm nor refute broader societal hypotheses about causation in long-term mental-health trends.

The study’s external validity is strongest for high-income democracies with similar media systems and social media penetration. Its limitations are appropriately noted by the authors: selection into participation, the short duration of abstention, and the election-period context. Nonetheless, the paper advances the literature by clarifying effect sizes, demonstrating the consequences of platform-specific content environments, and contrasting experimental estimates with flawed observational approaches that dominate the field. In doing so, it sharpens the empirical foundation for ongoing scientific and policy debates regarding social media, well-being, and regulation.