{"id":"f97ccc00-4327-42ca-b017-0027977af1cc","arxiv_id":"2606.00900","paper_version":1,"verdict":"ACCEPT","confidence":"LOW","novelty_score":5.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"RCT evidence on social media harms is limited because it captures local individual effects rather than global changes in social environments from mass adoption.","lead":"The paper highlights that RCTs on quitting social media only test individual changes while peers continue using it, unlike the widespread adoption that changes social norms for everyone. Generalists should read it to better understand why individual studies may not settle debates on social media's population-level harms to mental health.","discovery_kind":"extension","skeptic_critique":{"model":"grok-4.3","headline":"No significant objection identified","rationale":"The reader's weakest_assumption correctly isolates the premise that collective-environment effects are the policy-relevant target. Because the paper is framed as notes on limitations rather than a claim that global effects exist and are large, that premise functions as a scope condition rather than an untested empirical assertion. The argument therefore stands on its own terms.","tokens_in":1730,"tokens_out":268,"duration_ms":13942,"concrete_test":"Extract every RCT cited in §2–3 and confirm that none randomizes at a scale large enough to alter the local peer environment for treated units; if any does, re-read the local–global contrast against that example.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is a methodological observation: standard person-level RCTs identify local intervention effects (individual de-adoption while peers continue), whereas policy-relevant questions concern global interventions that reshape peer environments and culture. This distinction follows directly from the definition of SUTVA and from the scale at which social media adoption occurred; the manuscript does not assert that global effects dominate or provide new empirical estimates, only that existing designs cannot isolate them. No internal contradiction, hidden parametric assumption, or over-extrapolation appears in the argument as stated.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper claims that published RCTs on social media harms typically estimate effects of local (small-scale) interventions in which an individual is assigned to quit while peers continue using the platform in large numbers. In contrast, critics focus on global (large-scale) interventions involving mass adoption that reshape both proximal social environments and broader culture, potentially affecting even abstainers. The manuscript discusses this local-global distinction at length, notes other limits of person-level RCTs and observational studies for learning about aggregate harms, and suggests that triangulating imperfect evidence sources may yield the deepest insights into social media's effects on teen mental health.","tokens_in":1806,"tokens_out":361,"duration_ms":18121,"significance":"If the central distinction holds, the note usefully highlights how SUTVA violations and interference at scale limit the policy relevance of existing person-level RCT evidence. It draws on well-established causal inference principles without new empirical claims or derivations, providing a clear conceptual framing that could inform interpretation of the literature and design of future studies. The emphasis on triangulation is a constructive suggestion for a field where perfect identification of global effects is likely infeasible.","major_comments":[],"minor_comments":[{"comment":"Abstract: the phrase 'offers other notes on the limits' is vague; the manuscript appears to concentrate almost exclusively on the local-global distinction, so a more precise characterization of scope would help readers.","section":"Abstract"},{"comment":"The manuscript invokes SUTVA and interference without citing a standard reference (e.g., Imbens & Rubin or Hernán & Robins); adding one or two citations would make the grounding explicit for a broader audience.","section":null}],"recommendation":"accept","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for their careful reading and for recommending acceptance. The referee's summary accurately captures the manuscript's central argument concerning the local-global distinction in interventions and the potential value of triangulating multiple imperfect evidence sources.","responses":[],"tokens_in":1200,"tokens_out":62,"duration_ms":12929,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main point here is that RCTs on social media harms usually randomize one person to quit while peers keep using it, but the policy question involves mass adoption that changes norms and environments for everyone, including non-users. This local-global split follows directly from interference and SUTVA, and the paper spells it out clearly in the context of teen mental health debates.\n\nIt does well by connecting this to existing causal ideas and noting that global shifts could affect abstainers through culture in ways local trials miss. The closing suggestion to triangulate imperfect evidence sources is straightforward and useful for anyone trying to interpret the current RCT literature.\n\nThe soft spots are that it stays at the level of conceptual notes with no new data, simulations, or formal extensions. It draws on well-known limitations rather than quantifying how much the local-global gap matters in practice or adding specific citations to the social media debates. The premise that global effects are the ones that count is asserted but not argued at length.\n\nThis is for methodologists or policy analysts who already know the basics of interference but want the distinction applied to social media evidence. A reader looking for reminders on why individual randomization does not automatically scale to population interventions will find it helpful.\n\nIt deserves peer review as a short commentary because the reasoning is consistent and the topic is relevant, even if the contribution is modest.","headline":"The paper flags that person-level RCTs test local social media interventions while policy debates concern global ones, but mostly restates standard causal inference limits without new content.","tokens_in":2255,"tokens_out":348,"would_cite":false,"duration_ms":21491,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"RCTs on social media harms measure only what happens when one person quits while peers keep using it.","keywords":["social media harms","randomized controlled trials","local versus global interventions","teen mental health","aggregate effects","evidence triangulation","observational studies"],"falsifier":"A comparison showing that mental health among non-users is identical in high-adoption versus low-adoption communities, after accounting for individual usage, would undermine the claim that global interventions produce distinct harms.","tokens_in":2619,"feed_emoji":"","tokens_out":670,"duration_ms":17924,"temperature":0.7,"pith_summary":"The paper points out that most randomized trials assign individuals to stop using social media while their immediate social circle continues at normal rates. Critics instead describe harms from a society-wide shift where nearly all teenagers adopt the platforms, reshaping norms, relationships, and daily life for everyone including non-users. Because these two scenarios change different things, evidence from the first does not directly test the second. The distinction matters for deciding how much weight to give existing trials when assessing overall effects on teen mental health. The paper proposes combining several imperfect data sources to better approximate the aggregate picture.","feed_headline":"RCTs capture only local effects of quitting social media","feed_subtitle":"Mass adoption changes peer environments and culture in ways single-person experiments do not test, limiting what trials reveal about populat","key_machinery":"the local-global distinction between interventions, where local means one individual changes behavior while surroundings stay the same and global means collective change alters surroundings for all","core_discovery":"Published RCTs identify effects of a local intervention in which one person stops using social media while peers continue; critics focus on global interventions in which mass adoption alters both the proximal social environment and the broader culture, potentially harming even those who abstain. The paper examines this local-global distinction in detail and notes parallel limits in person-level observational studies, concluding that triangulation across imperfect evidence sources offers the best route to understanding social media's aggregate influence on teen mental health.","pith_inferences":["Researchers could examine mental health trends in communities that adopted social media at different speeds to isolate global effects.","Policy recommendations based solely on individual-level trials may understate risks if network-level changes are the main channel.","The same local-global gap likely applies to other technologies whose effects depend on widespread uptake, such as smartphones or online gaming."],"forward_implications":["RCTs that remove social media from one person while peers retain it will miss any harms that depend on changed peer behavior or cultural norms.","Person-level observational studies share the same limitation because they also compare individuals within the same broader environment.","Evidence on aggregate population effects must come from sources that capture collective changes, such as differences across regions or time periods with varying adoption levels.","Combining RCTs, observational studies, and aggregate data can narrow uncertainty about overall impacts even when no single source is decisive."],"fun_headline_variants":["RCTs test local quitting not global peer shifts","Mass adoption changes untested by single-user trials","Local experiments miss culture-wide social media effects","Person-level studies overlook environment alterations"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The harms of greatest interest are those produced by shifts in the overall social environment and culture rather than by direct effects on any single user.","fun_headline_variants_meta":{"raw":{"variants":["RCTs test local quitting not global peer shifts","Mass adoption changes untested by single-user trials","Local experiments miss culture-wide social media effects","Person-level studies overlook environment alterations"]},"model":"grok-4.3","cost_usd":0.003539,"raw_usage":{"total_tokens":1837,"prompt_tokens":628,"num_sources_used":0,"completion_tokens":53,"cost_in_usd_ticks":35387000,"prompt_tokens_details":{"text_tokens":628,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1156,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":628,"tokens_out":53,"duration_ms":9673,"temperature":1.0,"reasoning_tokens":1156,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-28T17:54:31.104878+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A comparison showing that mental health among non-users is identical in high-adoption versus low-adoption communities, after accounting for individual usage, would undermine the claim that global interventions produce distinct harms.","supporting_citations":[],"review_version":1}