Behind the Scenes of the Consensus Statement on Potential Negative Impacts of Smartphone and Social Media Use

A sneak peak at how we tried to obtain consensus from over 100 experts about the impact of social media

Is social media ruining a generation?

This is a question that has sparked countless debates over the past few years. From Congressional testimonies and popular TV shows like Adolescence to best-selling books and documentaries, it feels like everyone has an opinion. Yet these debates have been polarized—even among academics. Some experts have called this concern a moral panic and others, are, well, panicked.

This debate has made it difficult to know exactly what to believe. To help cut through the noise and get a handle on what most experts actually believe, we decided to survey a large panel of scholars who do research on social media. This started as a fun exercise because we were eager to put people from multiple fields and every corner of the debate to find out if they could reach some kind of consensus. We were genuinely unsure what the results would yield.

After bringing together the experts and painstakingly sorting through the comments and critiques, we were excited to share our preprint of the project, entitled A Consensus Statement on Potential Negative Impacts of Smartphone and Social Media Use on Adolescent Mental Health. Our paper described how we pulled a broad panel of experts to evaluate the state of the scientific evidence regarding the impact of this technology.

To our surprise, our paper sparked a ton of discussion on social media. While many discussions have been constructive with people debating the results and the methods of our study, there has also been a great deal confusion and even misinformation about the paper. Therefore, we decided to explain the findings and give people a rare look behind the scenes about how we launched and conducted this project, including who we invited to our panel and the initial conversation that launched the project.

The findings from the study

As we wrote in our abstract: The impact of smartphones and social media use on adolescent mental health remains widely debated. To clarify expert opinion, we convened over 120 international researchers from 11 disciplines, representing a broad range of views. Using a Delphi method, the panel evaluated 26 claims covering international trends in adolescent mental health, causal links to smartphones and social media, and policy recommendations. The experts suggested 1,400 references and produced a consensus statement for each claim.

The Delphi method is a structured communication technique used to gather and synthesize expert opinions on complex issues, typically through multiple rounds of questionnaires. It aims to reach a consensus or identify areas of agreement and disagreement among experts without face-to-face interactions, fostering independent thought and reducing social pressures. It was originally used in the 1960s to assess trends in science and technology development.

After several rounds of discussion with our large panel of experts, we slowly revised a series of statements about smartphones and social media. We then put these statements to the test by polling our experts. Our paper reveals the initial beliefs of a large panel of experts and then, through the process of discussion, how these statements became more nuanced and achieved higher consensus. Overall, the results of this deliberative process and the set of concrete recommendations provided can help guide future research and evidence-informed policy on adolescent technology use.

If you want to see the discussion among the experts, we included over 170 pages of supplemental material in our preprint plus another 120 pages of partial consensus statements in the OSF page, describing all the steps of the discussion and listing the citations the experts shared as we slowly worked towards a consensus. Alas, the word limit for our article submission did not allow us to include all these details in the body of the paper but we decided to share them in the supplement and on OSF to maximize transparency.

Constructive Criticisms

As we noted above, we received many thoughtful comments and criticisms. Some critics on social media have questioned whether our Expert Panel represented the various viewpoints on the topic. We agree that these are legitimate questions. To help answer these questions, we thought it might be helpful to take you behind the scenes to explain how the project began, the steps we took to balance voices, what worked, what didn’t, and how we attempted to reach a consensus. We tried to convey this to journalists, but it never made it into their columns. Therefore, we decided to share it ourselves.

Before we begin, we also want to note that our paper was a preprint and has not undergone peer review. This makes it an excellent opportunity to make improvements, and we will continue to revise the paper and share an updated version once we receive (and address) comments from reviewers. We appreciate the people who have read our paper, engaged with the content, and offered constructive criticism. Thank you. We are saving the best suggestions and plan to incorporate many of them when we revise the paper. This will no doubt improve the paper, just like any review process. And please email us if you have any further suggestions.

Finally, if you are an expert on the topic who would like to complete our consensus survey please complete this form. We are planning to expand our pool of experts and include folks who missed out in the first round.

Now back to the very beginning…

The origins of the project

Our project began on August 13, 2024, after we each received invitations to contribute to a special issue of the Annals of the New York Academy of Sciences on public health and digital technologies since we both conduct research broadly in this area. Over email threads with Laura Globig and Steve Rathje, the idea first emerged: why not convene a broad panel of experts to map expert opinion on the potential negative impacts of digital media use on adolescent mental health, and try to reach a consensus about what we know and what we don’t know? This is what we call Stage 1 of the process in the paper: the conception of the idea.

At this stage, we began with the goal to include a diversity of perspectives—especially critics. Valerio came up with the initial idea in an email to Jay where he wrote “since the journal also accepts reviews and perspectives, I'm thinking about putting together a small team to write a short review or perspective piece on the impact of social media on mental health.” He suggested adding other authors currently writing about the topic, including Jonathan Haidt.

Jay responded a few hours later, writing “This sounds interesting to me. I was invited too but the focus on health is something we only do peripherally so joining a team for this makes sense. I've looped in my postdocs Laura Globig (who is interested) and Steve Rathje (who may also be interested as he is doing work in this area). Obviously Jon Haidt would be a good fit, and he also has a handful of critics who might be worth inviting to try and build a consensus statement. Of course, that could also be stressful. So I'll defer to you.”

Steve jumped in an hour later, expressing his interest in the topic and noting that he had recently published a paper that surveyed experts' beliefs on misinformation. He said “It could be interesting to get perspectives from Jon Haidt believers and his critics (e.g., Amy Orben's team) on some of these issues. From there, we started brainstorming names of experts from competing perspectives we wanted to invite, starting with Candice Odgers, Andrew Przybylski, Stuart Ritchie, and Jean Twenge. Valerio took the initiative and started inviting experts (more details later).

Meanwhile, we started brainstorming the claims to be evaluated, based on Jonathan Haidt’s The Anxious Generation, since it contained many ideas that were at the center of the academic and public debate. Valerio organized all the ideas into a first list of central claims, which is still available at this link. This methodological process is what we call Stage 2 of the Delphi process.

At no point in this chain did any of us express our beliefs about any of these claims. In fact, as funny as it sounds, the two of us have never discussed our personal views and we don’t know exactly what each of us believe about these claims. From our initial emails through the final survey, we tried to ensure that we obtained a broad set of experts, consistently and explicitly noting our desire to include critics. For complete transparency, we are happy to show our email exchange to anyone who wants to read it.

At this point, we moved to Stage 3 aimed at developing the precise claims to be tested.

The Claim Development Team

On August 19, Valerio reached out to five leading voices who represented a broad spectrum of positions on the public debate about smartphones/social media and health:

• Jonathan Haidt and his chief researcher Zach Rausch, so the claims would accurately reflect Haidt’s core arguments.

• Two of his strongest public critics, Candice Odgers and Amy Orben, who have published prominent papers on the topic.

• Generational trends scholar Jean Twenge, whose views are broadly aligned with Haidt’s and regularly writes about this issue.

Our aim was to include a fair representation of the range of views from the most prominent voices. All five experts were informed that our source of claims was Haidt’s book, but we did not disclose the full invitee list, to make sure their feedback was focused purely on the scholarly content. Moreover, at this stage we did not mention any specific claim to be evaluated. We simply invited them to join “a team to write a consensus statement on the causal impact of social media on mental health”, specifying that as “part of our methodology, we plan to select the central claims from The Anxious Generation and conduct a survey among experts to evaluate the extent of experimental evidence supporting each claim and identify key areas for future research.”

By the end of the day, Haidt and Rausch agreed to join; Orben declined, and Odgers had not replied. The next morning (August 20), to rebalance the viewpoints, we invited another well-known critic of Haidt, Andrew Przybylski, disclosing that Haidt and Rausch were already on board for transparency.

Two days later (August 22), while we were still waiting from responses from Odgers, Przybylski, and Twenge, we decided to further expand the group, including scholars with varying perspectives and career stages, and we extended invitations to Sakshi Ghai, Patricia Greenfield, Jeff Hancock, Ethan Kross, Sonia Livingstone, and Jay Olson. All but Ghai replied, and they all accepted.

In early September, a round of reminders was sent to Ghai, Odgers, Przybylski, and Twenge. Only Twenge confirmed; Odgers and Przybylski declined, and Ghai remained silent.

Therefore, our “proto–Claim Development Team,” included: Capraro, Van Bavel, Globig, Rathje, Haidt, Rausch, Greenfield, Hancock, Kross, Livingstone, Olson, Twenge, Wormley (involved by Kross), and Yuson Lee (involved by Hancock).

In the weeks that followed, Valerio, working with the Core Group (Valerio, Laura, Steve, Jay) and with direct feedback from Haidt and Rausch, translated the book’s arguments into a first list of 24 potential claims for our list of experts to evaluate. For the evaluation of the claims, Valerio proposed to the Core Group to use the THEARI method that he and Jay had already used in a previous paper to evaluate a number of scientific claims about behavioral science related to the COVID-19 pandemic. This method was developed by Kai Ruggeri, a Public Health Professor, to classify and evaluate scientific evidence relevant to public policy. Valerio’s proposal was accepted.

Valerio prepared a first draft of the survey and circulated it to the wider proto-Claim Development Team, clarifying that the survey was not active: “At this stage, we are gathering feedback for additional improvements. Feel free to provide any comments on any part of the survey.”

Ethan Kross, Jay Olson, Alexandra Wormley, and Jean Twenge provided feedback on several parts of the survey, including the claims themselves, the questions to be asked, and the response options, while Jeff Hancock, Patricia Greenfield, and Sonia Livingstone did not respond. After receiving a reminder, Livingstone formally withdrew, simply saying “On reflection, I will not be participating”, while Hancock and Greenfield never replied. These shifts left us with ten members in the Claim Development Team: Capraro, Van Bavel, Globig, Rathje, Haidt, Rausch, Kross, Olson, Twenge, and Wormley.

Reflecting on this phase, we recognize that critics of The Anxious Generation became underrepresented, despite our deliberate efforts to include them. Attrition is a common feature of large collaborations: participants leave for many reasons, often unrelated to the work itself. Every invitee knew from the outset that our claims would be drawn from Haidt’s book, so we anticipated discussions about which claims to include and how to phrase them, not whether to base the work on this source. Indeed, there were many such discussions, and the final 26 claims were the point of convergence following an extensive deliberation.

Given that the central focus of our project was not the initial claims per se, but the discussion on each claim with an aim to reach a consensus (on at least some claims) we sought to have critics discuss and evaluate these claims. Specifically, we planned to include a large and diverse Expert Panel to engage in the critical step of evaluating the claims.

The core of the Delphi process lies in its iterative surveys and in the diversity of the Expert Panel. Through a series of surveys, the Expert Panel is involved in structured discussions of the scientific evidence regarding the various claims, aimed at delineating the largest common ground where most people can agree. With this in mind, we dedicated vastly more effort to assemble a diverse panel for the four surveys of the Delphi process and integrating the input from these experts until we could identify where people had a consensus (or where they did not).

The Expert Panel

To build a diverse panel, we combined several strategies: we re‑invited anyone from the previous stage who might yet wish to weigh in, we invited prominent critics such as Chris Ferguson, we created a list of experts in several fields, with various perspectives on the topic, and in different career stages, and we advertised our survey on targeted academic forums. Early respondents to the first survey were also invited to recommend peers, leading to further invitations.

There was misinformation shared on social media that the experts were selected by Jon Haidt. This is false: the overwhelming majority of experts had no contact or relationship with Haidt. They either responded to our surveys or were invited by other experts.

In total, 288 experts received invitations to participate in at least one of our four Delphi surveys. For complete transparency, the full list of experts included in this list is available using this link. Experts represent 11 different disciplines, including psychology (30%), communication and media (11%), health sciences (11%), business and management (11%), economics (8%), social sciences (7%), and psychiatry (4%).

Does the set of experts who responded to our surveys represent the various viewpoints about the debate?

It is now important to make a somewhat technical observation. From a statistical perspective, a sample is representative of a target population with respect to some variables, if those variables are proportionally represented in the sample as they are in the broader target population. In our case, the key variable is “viewpoint on the debate” and the target population are the “experts on the topic”.

To statistically test the representativeness of the Expert Panel we needed to know whether the various viewpoints of the Expert Panel proportionally corresponded to the various viewpoints of all the experts. However, this is impossible to check because ours is the first attempt to map expert opinions on this topic. So, nobody knows how these viewpoints are distributed and therefore, there is no benchmark distribution against which to compare representativeness.

Instead of testing for statistical representativeness, what we could do was test for a diversity of viewpoints. This corresponds to asking this question: are there any relevant viewpoints that are missing?

In our paper’s Expert Panel section, we laid out three converging lines of evidence suggesting that the answer is “no”:

1) First, at the outset of Survey 1, we asked participants to rate the overall impact of smartphones and social media on adolescent mental health. Responses spanned the entire continuum, from strongly negative through moderately negative, neutral, and moderately positive to strongly positive, with two‑thirds of experts emphasizing that effects depend on context and moderating factors. In short, we had a highly diverse list of experts with very nuanced views.

2) Second, we measured pre-consensus veracity judgments for each of the 26 claims. On 25 of those claims, experts’ responses included both “probably true” and “probably false”; only the claim linking chronic sleep deprivation to mental health decline was unanimous. This reveals diversity in our panel of experts, even at the granular level of the claims.

3) Third, during Surveys 2–4, panelists rated proposed consensus statements on a one-to-five accuracy scale. The lowest scores averaged around 1.4 while the highest reached a perfect 5.0, proof that both critiques and endorsements informed our Delphi process. In total, 250 critical comments from 60 named experts and 5 anonymous experts were addressed. In the OSF page of the article, we transparently reported all the stages of the process, including the critical comments, our responses, and proposed changes or rebuttals. We invite readers to see for themselves that no critical comment was ignored.

By the end of Survey 4, these indicators painted a clear picture: alarmist and skeptical voices alike shaped every iteration of our Delphi process, no substantive viewpoint was excluded. We created the broadest tent possible, incorporating or responding to all critiques from our team of experts.

Of course, to have a viewpoint represented is different from having that viewpoint proportionally represented. This is why, in the paper, we highlight that the quantitative results should be taken with caution. However, the consensus statements are far more robust to this potential issue, because of their qualitative nature: it is enough that a viewpoint was voiced by one person to be represented, and the Delphi process is designed to respond to minority perspectives when building consensus.

Other criticisms and revisions.

Objections have largely centered on who was at the table rather than what emerged from it. Even if our invitees had skewed one way or the other, that imbalance would only matter if it generated flawed judgments. Ultimately, the central question is whether any substantive errors exist in our 26 consensus statements. If any expert uncovers a genuine error or oversight, we stand ready to make any necessary corrections during the review process.

Our goal has always been to anchor the consensus firmly in empirical evidence and to create a process where divergent perspectives can converge without sacrificing accuracy. For this reason, we remain open to improving our paper if people observe any factual mistake, conceptual oversight, or misinterpretation of evidence in any of our 26 statements. Please email us if you spot any specific issues.

Challenges about Communicating Consensus

Reading the debate on social media, it seems that another source of confusion stemmed from the fact that the only figures in the paper referred to experts’ pre-consensus opinions about the claims, rather than the finalized consensus statements. As noted above, this was useful to reveal the range of perspectives among our experts. This ranged from initial consensus on some claims (e.g., sleep deprivation can reduce mental health), to a lack of consensus on other claims (e.g., decline in Nordic mental health). It also revealed a great deal of nuance with many experts noting they “don’t know” about several claims. These figures report “snap expert opinions” about stronger claims.

The finalized statements—summarized in Table 2—reflect the results from the Delphi process. These reveal a nuanced view among the experts that incorporated 250 critical comments. This table shows the finalized claim that emerged through discussion, the final consensus rating, and key evidence relevant to the claim. It also includes potential future research directions since our consensus statements do not reflect the final word on any of these claims and should inspire more research.

One inherent challenge here is that the output of a Delphi process is primarily verbal and qualitative, making it difficult to represent accurately through a figure. This is largely why we relied on a table to summarize the consensus results. The figure simply presents numerical indicators collected before the consensus process. We explicitly made this clear in the paper—the section is titled “Experts’ pre-consensus beliefs and awareness of the evidence” [emphasis added].

Many readers on social media appear confused as these details were in the paper. This distinction may not have been salient enough to people who did not read the paper and thought that the conclusions of the consensus statements were bolder. Given this confusion, we will try to communicate this more clearly during the revision phase, such as including more detailed figure captions, and including those captions in social media posts.

Another challenge we faces is writing a paper with over 100 authors. Not everyone agrees with every editorial decision (e.g., which findings to include in the paper vs. the supplement or how to summarize the findings in the abstract). For instance, some coauthors wanted us to make stronger conclusions in our paper and others wanted weaker conclusions. This led to challenge as we sought to strike a balance between the various editorial suggestions we received, while retaining accuracy and communicating the key findings succinctly.

This is an impossible problem since we will never get complete agreement among every single author, especially when we intentionally invited people from such a diverse group of experts. One of our colleagues refers to this as the “Too many cooks in the kitchen” problem. Indeed, we lost potential authors from both sides of the debate for failing to fully comply with their desired framing. However, nearly all our collaborators were happy enough with the paper that they agreed to sign their name to the paper.

As such, our paper represents a compromise between many competing perspectives and a best *first* attempt at an honest representation of the methods and findings. But we will continue to edit the paper through the review process and hope the final published paper will be an even better approximation.

Despite these natural disagreements, our hope is that the consensus statements will better reflect the perspectives of our experts given the rigorous process we undertook to surface them. However, we are already brainstorming about how to include an even greater pool of experts and expand our project (perhaps as part of this paper or a future paper).

Avoiding a false equivalence

In most hot button debates, the people driving the public discourse—especially on social media—are often those with the strongest opinions. And people with more nuanced views often stay on the sidelines. This can lead to a funhouse mirror version of reality on social media, where people appear highly polarized despite a broad consensus on important issues.

Unfortunately, this can even produce a false equivalence—where it seems like an issue is evenly divided. This regularly happens in political coverage and science debates, when journalists select people from both sides for debates. Our hope is that a consensus statement can serve as a common starting point for public discussions. There appear to be many issues where people agree, and the consensus is quite nuanced.

Of course, it’s possible for the literature—and therefore consensus—to evolve. As more research emerges, we hope that areas of ambiguity will diminish, and the consensus will increase. In the near term, we think our paper can serve as a useful guide for topics where research is urgently needed. Rather than an end point for research, we believe this offers a roadmap. Specifically, far more research is needed on the causal impacts of social media and potential policy solutions that reduce social media access.

This post was written by Valerio Capraro & Jay Van Bavel, with edits from Laura Globig and Steve Rathje

---

In the News

Jay and former lab member Clara Pretus wrote an op-ed on their neuroscience research regarding what happens in the brain when people use different lenses to make decisions, including moral, hedonic, and pragmatic . You can read the op-ed here.

New Papers & Preprints

We have a new preprint out on the accuracy of expert predictions - specifically in the domain of climate change. Expert forecasting is increasingly used to anticipate the success of behavioral interventions. However, little is known about how accurate expert predictions are—particularly in the climate domain. In this pre-registered study, we asked academics, government officials, climate change communicators, and the general public to predict the effectiveness of 11 climate interventions on belief, policy support, and effortful sustainable behavior. We found that academics generally outperform the public, but how well they do varies. Unfortunatley, they do not outperform predictions based on simple heuristics. You can read more here. This paper was led by former lab member Kimberly C. Doell, along with Lukas Lengersdorff, Shawn A Rhoads, Boryana Todorova, Jonas P. Nitschke, James Druckman, Madalina Vlasceanu, Claus Lamm, and Jay.

Announcements

Our Master’s student Nadya Hanaveriesa won a Graduate Student Research Award to support her thesis research! Congratulations to Nadya!

We also want to extend our heartfelt congratulations to our former lab manager Yifei Pei who will be completing a Master’s of Education at Harvard this fall before entering Penn State’s cognitive psychology PhD program in Fall 2026.

---

If you have any photos, news, or research you’d like to have included in this newsletter, please reach out to our Lab Manager Sarah (nyu.vanbavel.lab@gmail.com) who puts together our monthly newsletter. We encourage former lab members and collaborators to share exciting career updates or job opportunities—we’d love to hear what you’re up to and help sustain a flourishing lab community. Please also drop comments below about anything you like about the newsletter or would like us to add.

And in case you missed it, here’s our last newsletter:

That’s it for this month, folks - thanks for reading, and we’ll see you next month!