select search filters
briefings
roundups & rapid reactions
Fiona fox's blog
A study published in Nature Human Behaviour looks at early social media use and standardised test scores in schoolchildren.
Dr Simon R. White, Assistant Research Professor, University of Cambridge (Department of Psychiatry, and MRC Biostatistics Department), said:
“The key finding of the paper is an association between age/grade social media use (specifically registering for your first personal account rather than shared/parental access) and academic outcomes, with young people who take up social media earlier having worse scores compared to those who start later.
The paper uses the difference-in-difference methodology to attempt to inference that rises above association into a more causal statement. There are limitations with the method, which the authors consider in the appendix/supplementary material. There remains a gap from stating this is a direct causal link, however the authors present a strong and robust analysis which does add substantive evidence for the claim. Interestingly, as stated in the paper, many young people are reporting making personal social media accounts below the permitted age — explicitly violating the rules of use of the platform(s); so the implications for parents, schools, policy makers, governments, and digital platforms are potentially complex. Within the UK in particular, the government have announced proposals for a social media ban (although the specifics are not yet published).
The standardised scores are obtained through linking to a database, while the social media use is through a retrospective question. Recall about social media is stated to be very reliable, based on pre-testing and previous research. However, personal access to social media is only a single “landmark event” and there are some limitations within the methodology with regard how social media use might change over time — perhaps coupled with academic outcomes. With regard confounders, they use several methods of matching to account for bias and the findings are similar across these approaches. The authors have indicated they considered the issue of statistical multiple comparisons in Nature reporting summary, however the paper itself does not mention how — or if — the presented findings have been corrected; specifically there are four outcomes (Italian, Maths, English Listening, and English Reading) and three matching groups (6th–9th/later, 7th–9th/later, 8th–9th/later) — so there are twelve chances to find a statistical result. Conducting a crude Bonferroni correction to Table 2 would negate some of the statistically significant findings, so that is something to consider when interpreting the findings.”
Prof Kevin McConway, Emeritus Professor of Applied Statistics, The Open University, said:
“I’m very pleased that the authors are quoted in the press release as saying that these findings should be interpreted cautiously. I certainly agree with that. The direct quote from the research paper on this point says, “Given the sensitivity of the topic and the intense, polarized debate that has emerged in recent years, these findings must be interpreted with the utmost care and responsibility.” Indeed.
However, I’m somewhat concerned that the tone of some of the interpretations of their results, that the authors give in the paper, is rather too positive about whether they have established that starting to use social media at earlier ages is actually one of the causes of worse performance in school examinations. It certainly might be, but there are other possible explanations
The overall finding is that the earlier the age at which children open one or more social media accounts, the worse their performance in standardised school testing at grade 8 (the school year when they are aged 13 to 14) and grade 10 (age 15 to 16) will be.
The methods used to establish this result are quite complicated, and I think it’s right that the press release says only that the students who started earlier with social media “appear to perform worse”. (My emphasis.)
But the important question is whether this difference in test performance is caused, in whole or in part, by starting with social media at younger ages.
That’s certainly one possible interpretation. But, despite using some ingenious (and in my view generally appropriate) statistical methods, other interpretations can’t be ruled out.
Basically this is because the study is not a randomised experiment, where children would be chosen at random to open a social media account at different ages. It’s an observational study, and observational studies always raise issues of what is causing what.
The researchers used a method called, in the jargon, difference-in differences models. They also used ways of matching students who started social media use relatively late (at grade 9, age 14 to 15) with students who started earlier, in terms of a long list of factors that might also be associated with later differences in school performance.
Together, these approaches do go some way towards dealing with the possibility that the observed correlation between age of starting social media use and school performance is entirely due to some other factor or factors, that just happen to be associated separately with the age of starting to use social media, and with school performance.
But they can’t entirely rule out these other possible explanations of the correlation between starting social media earlier and doing worse at school. Other possibilities patterns of cause and effect still exist, particularly if they involve factors, about the child, about the family or about the wider context, that were not recorded and thus couldn’t be used at all in the research.
Actually, the doubts about cause are reflected by another quote from the research paper: “Therefore, we interpret the estimates as causal, conditional on the plausibility of the parallel trends assumption and the absence of time-varying confounders.”
This looks like statistical jargon, because that’s what it is – but very broadly, the researchers are saying that their correlations have an explanation of cause (early social media use) and effect (worse school performance in grades 8 and 10) only if two statistical modelling assumptions, the parallel trends assumption and the absence of time-varying confounders, are reasonable. They give some justifications, particularly of the parallel trends assumption, but I don’t find them entirely convincing.
So it certainly is possible that early social media use could be a cause, alongside others, of worse school performance later, but that’s not the only possibility.”
Prof Dennis Ougrin, Professor of Child and Adolescent Psychiatry and Global Mental Health, Queen Mary UoL, said:
“The study adds to growing evidence that the age at which young people first engage with social media may matter for their educational development. While the findings do not prove that social media alone causes poorer academic performance, they suggest that earlier, largely unsupervised use is associated with differences in later attainment, particularly in mathematics and language. The study also points towards plausible mechanisms, such as frequent phone checking and reduced attention, rather than simply screen time itself. The message for parents, educators and policymakers is not necessarily to ban social media, but to support young people in developing healthy habits when interacting with social media.
The key strengths are:
– The large sample of more than 5,000 students over multiple years
– The link between self-reported social media use and objectively measured standardised educational outcomes
– The use of statistical methods designed to account for a range of confounders.
– Also, the dose-response pattern uncovered strengthens confidence that the association is important.”
Dr Poppy Gibson, Lecturer in Education studies (Primary), The Open University, said:
“In this longitudinal study of 5,227 Italian students, these authors found that opening social media accounts earlier in adolescence was associated with lower Italian and mathematics test scores later in school, equivalent to roughly six months of learning. While observational, the findings suggest excessive social media use, distraction and reduced parental supervision may contribute to poorer academic outcomes. With a reasonable sample size (5227 participants), this paper offers a reflection on how social media use may affect academic success in standardised tests. As an educator, however, I argue that standardised tests in their very design are a flawed way to measure ‘success’, and perhaps what sits behind the data of poorer scores is, as the authors mention, the lack of parental supervision. Less parental supervision of a child or young person’s social media use may in theory suggest less parental support with wider academic study and learning in the home environment. The implication for the real world is that young people may find it harder to engage in offline activities, but we must not assume that learning does not happen in the online spaces on social media platforms. In fact, participation in social media networks may lead to learning as well as confidence building, relational communities in finding others with whom the young person relates, and inspiration for their own behaviours and interests. Standardised tests, in a change landscape of human living, may become less important over the valuable skill development and human qualities that individuals may need to problem solve, create and dream when Artificial Intelligence can absorb many of the global tasks that previously needed specific tick box ‘knowledge’.
So perhaps this study is bigger than test scores; this paper offers a timely provocation of how phones are impacting young people; it is an interesting and thoughtful study that encourages both educators and parents and carers to think about the future impact phone use may have on behaviours and wider success.”
From our friends at SMC Ireland:
Dr Dean McDonnell, Lecturer of Psychology, South East Technological University, said:
“This is one of those studies that when you stumble upon it, you adjust your seat. It uses longitudinal standardised data and matching and difference-in-differences models that are far stronger than our usual cross-sectional methodologies. Although the design makes it difficult to eliminate all unmeasured differences between groups, it highlights that earlier social media adoption is followed by lower scores in two of the three subjects: Italian and mathematics.
“A broader question is what these tests capture. I am reminded of research on the Flynn effect, showing how changing environments can alter the ways people interpret and categorise information. An earlier generation might connect two animals behaviourally (one chases the other), whereas a contemporary response may classify both abstractly (as mammals). Both answers are correct and demonstrate understanding, but assessments will generally reward particular forms of reasoning.
“The same caution applies here. Findings that indicate lower standardised scores are eye-catching and meaningful, but should not automatically be translated into a general loss of intelligence or learning. It can be possible for contemporary digital environments to cultivate some competencies while displacing practice in others. For example, the absence of an effect for English is especially resonant here, as exposure to English-language material online may have offset some disadvantages of pervasive use. For me, I would think that this suggests how outcomes may depend on the interaction between platform design, content, attention, and displaced activities, rather than simply whether a child has a social media account.”
Dr Eoin Whelan, Professor in Business Analytics and Society at the University of Galway, said:
“While the press release for this paper suggests that earlier social media use may be harmful, that should be interpreted with an element of caution. Firstly, the study does find that earlier access to social media is associated with lower test scores, so what is important to understand is why this negative association exists.The researchers did investigate this and found that the child’s’ inability to control their use of their smartphone, and lack of parental involvement in managing online activity, were the main reasons why this pattern exists. This finding would support calls for particular features of social media platforms, such as auto-play and infinite scroll, to be modified and subject to appropriate safeguards. Second, the study found no link between exposure to social media and well-being for the average child. But of course there is no such thing as an average child. What this means in reality is that some benefit from social media, for others it is largely benign, but some children do have negative experiences online. So we should focus on who, why, and when children experience detrimental consequences when using digital technology.”
Dr Jason Power, Associate Professor, School of Education, University of Limerick, said:
“Researching the impact of specific technologies on people can be very challenging. There are so many factors that impact each of us every day that it can be exceedingly difficult to isolate the signal from the noise. The authors of this study have done an admirable job of navigating the many challenges and have provided a valuable insight into the impact of social media use on specific learning outcomes.
The study found that children who created social media accounts in 6th or 7th grade (around ages 11–12) experienced significant drops in math and language scores compared to peers who waited until 9th grade (age 14). By 10th grade, early adopters lost the equivalent of six months of math learning, faced higher risks of repeating a grade, and reported lower academic confidence. Distraction during homework, bedtime, and school hours was the primary driver of this decline.
However, important limitations must be considered. Account creation dates relied on student memory years after the event. Additionally, the study excluded high school dropouts and vocational students, groups with higher digital media use and lower test scores, meaning the findings likely understate overall harm. Finally, results reflect students in Northern Italy during a specific period dominated by smartphone video apps, so findings may vary across different cultures or technologies.”
From our friends at SMC Spain:
María Paz Prendes Espinosa, Professor of Educational Technology, University of Murcia, said:
“I believe the press release summarises the main findings of the study well, but they must be interpreted within the context of this study and should not be generalised or taken as definitive. On this subject, there are many studies, and they do not all agree; therefore, as the authors themselves acknowledge, the results must be analysed with caution.
“It is a quantitative study with a very large sample (more than 5,000 students from Lombardy, Italy), which is its main strength. However, a retrospective questionnaire was used. This means that information about the past was gathered from the respondents – in this case, adolescents – with all the biases that this may entail, as the results are influenced by memory. It is striking that the questionnaire was developed specifically for this study (ad hoc), and no explanation is given as to how it was validated, nor are any data provided on this (validity and reliability statistics) – aspects that are fundamental to any rigorous social science research.
“Therefore, whilst the robustness of the data is supported by the size of the sample (which the authors claim is representative), the study design itself may introduce factors that introduce bias into the results and undermine their validity. Furthermore, it is curious that only three subjects were studied (it is unclear why these three and not others) and that the study shows social media use influences only two of them (Italian and mathematics), but not the third (English).
“Social media has attracted interest ever since it entered our lives and has been the subject of numerous studies, but it has suddenly become the focus of the political debate on teenagers and their use of digital technologies. This study, whilst undoubtedly interesting, overlooks many aspects that cannot be ignored when analysing the consequences of social media, as the exclusive causal link between social media use and its impact on academic performance is difficult to accept. Does the family context not play a role? What about the social context? Do peer groups not have an influence? What about the adolescents’ personality traits? Does socio-economic status not play a part? And there are many other factors that could be added. In such a complex issue, which must not be oversimplified, there is also research demonstrating the usefulness of social media in promoting communication, social relationships, or even as tools for academic tasks integrated into students’ learning environments.
“The main limitation may stem from the questionnaire, which collects data on the respondents’ self-perceptions of their past. Can a teenager accurately recall exactly how much time they spent on their mobile and for what purposes when they were at primary school? Will they tell us this in a precise and reliable manner? In this type of research, it is useful to combine data collection with other qualitative techniques within a mixed-methods research design, so that the information can be triangulated. Relying exclusively on a questionnaire to collect data is very limited, particularly when a teenager is required to recall and describe what they did when they were a child.
“The main implication is that social media form part of our reality from an early age, and this study confirms this. Given this situation, banning the use of technology will not solve the problems associated with it. On the contrary, we must promote an education tailored to the real-world context in which we live; we must train and educate people to know how to manage the use of technology. In a digitalised society and in the age of artificial intelligence, we must not only be concerned about social media, but about technology in general. And this concern must lead us to strengthen our education system so that it can meet this need for training in digital literacy, recognising that digital literacy goes far beyond technical knowledge, as it involves the responsible, safe, critical and ethical use of technology and, by extension, of social media. If we want to protect our teenagers from the potential risks of technology, we must prepare them to live in a society shaped by technology”.
Lucas Gortázar, a researcher specializing in educational policies and reforms and director of the Education Department of the EsadeEcPol Center for Economic Policies, said:
“The press release is clear about the study and its limitations.
“It is the first longitudinal, quasi-experimental study to address the relationship between social media use and learning outcomes on a global scale. I believe that is the major contribution of this paper. The conclusions are well-supported and, whilst further studies are needed, the evidence is consistent in showing that the earlier a social media account is opened, the greater the effect. That said, it is a pity that they have not reported results by gender or socio-economic status.
This is the first contribution of its kind to this body of literature; clearly more is needed, but it is a major finding that provides further evidence on a fundamental question, one of which public authorities should be more aware.
All the factors that could be taken into account have been considered. It is possible that they could carry out some additional robustness tests, but I think that, generally speaking, it is sound.
It is a solid finding that points to the need to delay the age at which teenagers start using social media to, at the earliest, 15 or 16 years of age”.
‘Longitudinal effect of early social media use on standardized learning outcomes during school career’ by Marco Gui et al. was published in Nature Human Behaviour at 16:00 UK time on Monday 3rd August.
DOI: 10.1038/s41562-026-02522-4
Declared interests
Dr Simon White: “No conflicts of interest”
Prof Kevin McConway: “I don’t have any conflicts of interest to declare.”
Prof Dennis Ougrin: “No DOI.”
Dr Poppy Gibson: “I declare that I have no known competing financial interests or personal relationships that could have appeared to influence my comments or thoughts on this paper. I have no affiliation or connection to the authors of the paper.”
Dr Dean McDonnell: “I have no financial or commercial conflicts of interest relevant to this report. I am a developmental psychologist, member of the Psychological Society of Ireland and lecturer at South East Technological University. The views expressed are my own.”
Dr Eoin Whelan: “Eoin declares no DOIs.”
Dr Jason Power: None received
María Paz Prendes Espinosa: “She has stated that she has no conflicts of interest”
Lucas Gortázar: “He has stated that he has no conflicts of interest”
For all other experts, no reply to our request for DOIs was received.