Should We Ban Phones In The Classroom?

Yes, and laptops too

Electronics have had an enormous impact on the way in which we interact with education. They can both aid us in learning the material, and distract us from actually working. Despite their promise, their classroom usage is a negative for educational outcomes. Allowing phones and laptops into the classroom will reduce grades, with effect sizes comparable to an additional year of education.

In this article, I both want to summarize the state of our knowledge, and discuss why the studies which I favor are better than the studies cited by other people. As I have complained about before, the citation practices of many advocates for phones being a problem are bizarre. Rather than cite the studies which directly answer the question of interest, they rely upon studies which are either at best measuring a proxy for what we are after, or else are using methods which cannot reasonably be considered able to actually answer the question they are trying to ask.

We have a habit of treating a study, once peer-reviewed, as a one unit perfectly substitutable for any other study. This is not the case. Studies vary widely in their quality, and one must critically evaluate them. Because you will not be able to find the time to verify every study, it is my hope to gain a reputation as an honest, thorough, and competent evaluator of studies, so that you can trust my conclusions uncritically. I have been burnt too many times by the sloppy study evaluation of people banging out rapid reviews because they are too incompetent for creating real, novel ideas.

So how can we measure the effect of electronics on academic performance? The first thing you might think of is to measure how much different people use their electronics in the class, and then record what their grades in that class were. If the grades are sufficiently different, then we would say that the difference is statistically significant. We can even say how much of an effect it has – we can plot out phone usage and performance as a scatter plot, and draw the line which best fits the data. The slope of that line would then tell you how much increasing phone usage would be expected to reduce grades.

This is tempting, but wrong. The main reason why is that we have no way of saying that it was the phones that caused the difference in grades. We are not able to exclude the possibility that there is a third factor which affects both grades and phone use. What if, for example, less attentive students were both worse at paying attention to the lecture, and more likely to use their phone? When some factor is said to be causally responsible, then it means that changing that factor and nothing else would cause a difference in the outcome. Here, since it is the inattention simultaneously causing both phone use and lower grades, removing the phones would do anything at all.

For this, we need something “exogenous”, or outside of the system. For example, we might randomize whether or not students are allowed to access their phone during class. Then we may compare the two, and see if there is a significant difference. The randomization accounts for everything which might have a causal influence,

Randomization is better than papers which attempt to “control” for the things we can observe about someone. Not to get too into the weeds of how, but we are essentially trying to adjust for the normal effect of some variable, like socioeconomic status or intelligence. The trouble with this is that we have to make two big assumptions which do not often hold up: that we are able to observe and control for everything which is biasing our findings, and that the thing we are controlling for is not the pathway through which our variable of interest acts.

Let’s take an example of a study which is frequently cited, but does not answer the question we are interested in. Researchers at Michigan State University (Ravissa, Uitvlugt, and Fenn, 2017) had students in (presumably their) intro psychology class install an app which monitored what students were looking at on their laptop. They then found that spending lots of time on non-school material was associated with lower grades in the class. This is like the faulty study we outlined earlier. When you do this, there might be other variables causing both.

To control for this, they adjust for the student’s ACT score, as a proxy for ability. This can go wrong in two ways, though, and they point in opposite directions. First, it might be that both inattention and intelligence are responsible both for the grades and for the laptop usage. Controlling for one does not make the effect of the other go away entirely. Second, it could be that intelligence, as proxied by the ACT score, is causing the low grade through the mechanism of laptop usage. Then controlling for the ACT score would cause us to underestimate a true effect. Such a variable is a mediating variable. Much better, instead, is to use randomization or an exogenous shock, and to be clear about what the direction of causality is.

The evidence for phone use in the classroom unanimously concludes that it is bad for academic performance. We have several papers using quasi-experimental variation in policy to evaluate the effects. The first is Beland and Murphy (2016), who compare the difference in the differences between schools that banned and did not ban cell phones in the classroom in England. “Differences-in-differences” is an econometric technique which accounts for the possibility that schools that enacted phone bans might have differed in their characteristics, or the trends in those characteristics. They find that test scores improved by 6.7% of a standard deviation, driven by lower ability students doing relatively better.

Sara Abrahamsson (2024) takes the same approach, but with Norwegian data from a later time period. Beland and Murphy’s study was before the ubiquity of smartphones, and it is plausible that the effects are different now. She found that the GPA for girls who come from worse socioeconomic backgrounds rose, although she is primarily concerned with mental health outcomes. Girls increase their GPA by .08 standard deviations, and their math grades increased by 22% of a standard deviation. She explains the difference across genders by referring to time use studies – in Norway, girls use their phones much more, so there is more scope for the ban to have an effect.

More recently, the state of Florida banned phones in the classroom in 2023. Because all of the schools were affected at the same time, we may not want to simply look at the change in trend for the whole state afterward. There could be statewide changes which happened at the same time – perhaps a change in curriculum. What Figlio and Ozek (2025) do is construct a measure of how exposed a school is to the change in policy from preexisting phone usage. The idea is that schools in which people used their phones a lot see a bigger change in behavior than schools where people used their phones only a little. They don’t see much in the first year, besides a rise in behavioral incidents involving Black males, consistent with them conflicting with the teacher over having to turn in their phone. This increase is notably large, almost 30%. In the next year, test scores rise – people go up by about 2-3 percentiles – while the suspension spike goes away.

How large are these effects? We should put it in the context of other educational interventions. Decreasing the size of a class by one student increases test scores increases test scores by 2.5% of a standard deviation. Having students complete an additional year of education increases test scores by 25% of a standard deviation.

But what about laptops? Many people use laptops or notepads to take notes, consult the lecture slides, or to answer particular questions they might have about the material. While promising in theory, the best studies have found that laptops have a negative impact on learning.

The very best study comes from the Military Academy at West Point, by Carter, Greenberg, and Walker (2017). Being the Army, they have considerable latitude to just tell people to do things, which makes conducting an experiment easy. The Military Academy has an emphasis on small class sizes, so the introductory economics course (which is required) has numerous sessions and instructors. Some of these classes were assigned to allow laptops, while others banned them; a third arm allowed tablets but placed flat upon the table. They found that having computers of any kind in the classroom reduced exam scores by fully 18% of a standard deviation. Remember that this is an enormous difference, comparable to an additional year of education.

I wish to emphasize that one does not need for the researcher to randomize the assignment of treatment themselves. While randomization is ideal, we are able to infer by indirect means the causal effects of laptops. The best study to do this is Patterson and Patterson (2017), who have data on a liberal arts college of about 5,000 students. While everyone is required to possess a laptop, the policy on whether it is allowed, mandatory, or banned is left up to the discretion of the instructors. What they are able to exploit is that whether you have a class for which the laptop is mandatory or banned in your day affects how likely you are to bring your laptop to the classes for which it is merely allowed. If we presume that students are not scheduling their classes around laptop access (and a survey indicates that the vast majority have no idea what the laptop policy would be in advance) then we have got plausibly exogenous variation in laptop usage.

By contrast, as discussed earlier, simply measuring the association between laptop usage and eventual course grades is insufficient. What I am surprised by is that these studies have so few citations compared to studies which don’t actually answer the question. As I write this, Carter, Greenberg, and Walker (2017) has 301 citations, and Patterson and Patterson (2017) has 145. Ravizza et al has 265, and gets cited in the NYTimes by one of the foremost advocates for regulating screens.

The evidence for phone use outside of the classroom impacting performance has historically been poor, due to the lack of plausibly exogenous variation. We don’t have control over people’s lives all of the time – we can hardly ban some kids from accessing phones for their entire childhood. Lately, however, we have gotten a blockbuster study out from students at a Chinese university. (I believe this university, unnamed in the text, to be Jinan University, on account of it being their IRB which is in charge of the study).

The method on which the study rests is called a “shift-share” instrument. Here’s the idea – when a massive hit of a mobile game comes out, in this case Genshin Impact, the people who use it more are those who were already gaming more before. This allows us to isolate just the variation due to Genshin Impact, and find that variation’s effect on

Since roommates are randomly assigned, we can also use this as a source of variation – if you are assigned a roommate who plays video games more often, you too will use video games more. Finally, China imposed a sharp limit on how much minors could use video games. Some of the sample was below the age cutoff, but we can also use who you played video games with before as an instrument. If people are not choosing their friends before the ban in order to avoid it in the future, then we have an exogenous shift in usage.

Playing video games was quite bad for academic performance, and, in a rarity for this type of study, for job outcomes as well. Increasing one’s app usage by one standard deviation decreases GPA by .716 points out of a hundred, or by 36% of a standard deviation within a given cohort-major. A one standard deviation increase in roommate app usage decreases GPA by .450 points, or 22.7% of a cohort-major standard deviation. Gaming apps are particularly bad, with a one s.d. increase in gaming time causing a 1.119 point reduction, or 56.6% of a standard deviation.

The mechanism appears to be through time use. Students who play more due that at the expense of studying – they are tardier, and do not stay in the study halls as long. They do less searching for jobs, and are less likely to obtain a professional certification.

That the effects of screens appears to be mediated through time use is born out from studies of television when it rolled out. Whether it had negative (Hernaes, Markussen, Roed (2019), Durante, Pinotti, Tesei (2019)) or positive (Gentzkow and Shapiro (2016), Nieto Castro (2025)) effects was caused by what activities it crowded out doing. If it crowds out hanging with your delinquent friends or sitting in silence, it’s beneficial. If it crowds out reading, it isn’t.

So what do we know? When we ban phones and laptops from the classroom, we get better performance. Even if you truly believe that you are using it just to stay productive, it would probably be better to stow it away and make yourself listen to everything that is presented.

The mechanism seems to be through time use – studies of phone and television usage outside of school come to similar conclusions. It is not too early to say that banning cell phones is a good idea – and so be a ban on laptops.