What Neurodivergence Research Says Versus What We Hear

One of the things that consistently irks me as a therapist, and as someone at least reasonably competent at reading research, is how much a research finding can get contorted by the time it reaches the popular press or social media. I’ve spent years cruising research abstracts as a bit of a hobby, and more recently I have been doing deeper dives so I can write about the research itself. I keep seeing again and again where a study that says one thing gets quoted or misinterpreted as saying something else.

Sometimes the impact of these misrepresentations is significant. A study might look at a specific aspect of social media use and ADHD symptoms in a particular group of adolescents, and before long the public version becomes something like, “See, social media causes ADHD symptoms.” From there it’s not a very long jump to the old, false claims that ADHD or autism diagnoses are exaggerated, invented, or simply the result of modern technology.

The thing is, none of this requires anybody to be deliberately misleading. There are a lot of steps between the publication of a research paper and whatever version of it eventually settles into public conversation, and the meaning can shift even when each person along the way thinks they are summarizing it fairly.

What Gets Lost When Research Is Simplified

Part of this is a problem of scale. In a research paper, the authors have time and room to make careful distinctions between diagnosis and symptoms, between one age group and another, between statistical significance and clinical importance, and between what they measured and what they think might explain the result. A news story has much less room to communicate its point, and social media has almost none.

There’s plenty of evidence that some of this changing of a study’s conclusion happens well before it becomes a news article or social media post.

You may have heard the old adage, “correlation doesn’t equal causation.” In a landmark 2014 paper, Petroc Sumner and colleagues looked at 462 university press releases about health research, and compared the original journal articles with 668 related news stories. They were specifically looking for several kinds of exaggeration, including cases where correlational research was described using stronger and possibly unearned causal language.

Among the correlational studies they examined, almost a third of the press releases contained causal claims that went further than what the journal article concluded. When that happened, 81 percent of the related news stories also contained that stronger causal language. Even when the press releases stayed in line with the causal language of the research paper, 18 percent of news stories still exaggerated causation.

This means the situation is not as simple as careful scientists producing nuanced findings, and journalists distorting them. The descriptions of a study can begin shifting well before a reporter begins writing, and sometimes the research papers themselves invite several reasonable interpretations. Usually, everyone involved is trying to answer the same practical question: What does this research finding actually mean?

Part of the difficulty involves which level of the research we’re talking about. There are the measured results, there are the authors’ own interpretations of those results, and usually there are possible mechanisms discussed. Then there are broader claims that may seem reasonable, but only if several additional assumptions also happen to be true. And all of these layers can get folded together, or compressed, because writing about all of this can be quite cumbersome and complicated.

None of this requires malice on the part of the human chain going from researcher to final news article or social media post. But the conclusions in a news article or post can sound a lot more certain, more causal, and more impactful than what the research study itself may have said.

When an Association Gets Described as a Cause

A recent study of problematic social media use and ADHD symptoms is a great example. I followed several inflammatory social media posts back to the actual study, and found this:

Nagata and colleagues followed 11,286 adolescents across five annual waves of the Adolescent Brain Cognitive Development Study. Instead of just asking how much time the participants spent on social media, they measured “problematic social media use,” which has a specific definition including preoccupation, difficulty reducing social media use, and perceived negative consequences.

The researchers were interested in seeing if there were changes in the same person over time. If adolescent-reported social media use became more problematic at one wave of the study, did parent-reported ADHD symptoms change in the next wave of the study? Already you can see this is a very specific construct, measured at very specific times, in a very particular way. It would be reasonable to be wary of any sweeping conclusions derived as a result.

The study showed that during later parts of adolescence, increases in problematic social media use were followed by small increases in parent-reported ADHD symptoms the following year. The standardized effects were around 0.08, with stronger associations among male participants. The reverse pathway, in which ADHD symptoms themselves predicted later increases in problematic social media use, appeared much less consistently.

There is useful information in this study, and I am in no way trying to criticize the study itself. Following the same adolescents across several years gives researchers information that a one-time snapshot can’t provide. Using the longitudinal approach, the order of events can become clearer.

But causation is a different question entirely.

The study was observational. The researchers reported they could account for some but not all of the factors that might be influencing both social media use and ADHD symptoms. Researchers call factors like these confounding variables. We’ll come back to that idea shortly, because it matters far beyond this one study. But you can see that there are a lot of changes going on with any teenager over the span of their adolescent years. Sleep, stress, family circumstances, school demands, other forms of media use, emerging mental health problems, peer relationships, and a lot of things that are harder to measure could also be influencing both social media behavior and attention. If, for example, worsening stress led a teenager to use social media more problematically and to have more trouble paying attention, then some of the apparent relationship between social media use and ADHD symptoms could actually be due to the stress.

The paper itself acknowledges these limitations. But the findings become easier to communicate, and more flashy, once a number of crucial words disappear. Problematic social media use can become social media use in general. Parent-reported ADHD symptoms can become ADHD itself. A small prospective association during particular adolescent developmental periods can be summarized as social media makes ADHD worse. Then a reasonable reader of an article or post might come away with the incorrect conclusion that social media causes ADHD.

I’ve seen a number of social media posts about this study, using it to claim exactly that. Further comments on those posts then reiterate the tired old misinformation that ADHD is overdiagnosed, that teenagers are just lazy, that all technology is evil, that parents aren’t parenting sternly enough, and on and on and on. It irks me every time, and enough so, this time, I decided I had to write about it.

When One Outcome Category Includes Very Different Conditions

Studies measuring risk can create a related problem, because broad clinical categories tend to survive into public discussion more easily than the nuanced details.

Zhang and colleagues used Swedish registry data to examine long-term ADHD medication exposure and cardiovascular disease. Their study population included 278,027 people with ADHD. Among them, the researchers identified 10,388 who developed cardiovascular disease and compared them with people who did not.

Looking at the data, they found that cumulative exposure to ADHD medication was associated with higher odds of cardiovascular disease. Those numbers are worth taking seriously, especially because decisions about long-term medication involve balancing benefits and risks over many years. But it’s important to note that the category called “cardiovascular disease” included several different outcomes, and this detail didn’t survive into some of the public discussion about the study.

The strongest associations were for hypertension and arterial disease. However, the study did not find statistically significant increases for several other important outcomes, such as arrhythmias, heart failure, ischemic heart disease, thromboembolic disease, or cerebrovascular disease. And once those distinctions disappear, a broad phrase such as increased cardiovascular risk can get interpreted as risk of heart attack or stroke. But the study didn’t establish that.

Odds ratios create another layer of confusion, because they sound clearer and more direct than they really are. An adjusted odds ratio of 1.27 doesn’t mean that someone who takes ADHD medication for several years has a 27 percent chance of developing cardiovascular disease. Instead, it describes a relative difference in odds between groups. If the comparison group had only a 1 percent chance of developing cardiovascular disease in the first place, for instance, an odds ratio of 1.27 would put the other group at about 1.27 percent, not 27 percent. That is why a number like 1.27 can sound much more dramatic than the actual difference in someone’s likelihood of developing the condition.

We mentioned confounding variables earlier, and this is probably one of the most important concepts for understanding how research can get misinterpreted. In plain language, a confounding variable is a third factor that is related to both of the things researchers are studying, and can make the relationship between them look stronger, weaker, or more causal than it really is. Think about it this way: there are many reasons why someone who takes ADHD medication over a course of years might also have somewhat higher risks for certain cardiovascular problems, that don’t have to involve the medication at all. ADHD severity, diet and other health factors, use of other medications, or differences in health care access and use could all matter. The researchers adjusted for some of these factors, but they couldn’t account for every relevant difference between the groups.

This doesn’t erase the association the researchers found, but it certainly disputes the way this study also got misinterpreted in social media as “proving” that ADHD medication caused cardiovascular problems. Again, it’s possible that future research may be able to establish some causal relationship. But this study did not, and that massively important distinction gets easily lost in public discussion.

Finding a Relationship Is Different From Explaining It

Researchers can find that two variables are related, without knowing what exactly is producing the relationship. In the discussion section of a paper, they may propose possible mechanisms, meaning processes that could explain how one variable might influence another. Those explanations themselves are often easier to remember, and often get described in popular media more than the statistical finding itself. The mechanisms give the result a story about what might actually be happening, and it’s the stories that some people latch onto more than the nuanced data beneath them.

In the adolescent social media study, for example, the researchers discussed several possible pathways that might connect problematic social media use with later attention difficulties. Frequent switching between streams of information might affect sustained attention, and reward-related processes might play a role. Problematic social media use might interfere with sleep, which could then affect attention and executive functioning. But crucially, the study did not directly test those pathways.

There’s another research concept that is really important here: a mediator, sometimes called an intervening variable. A confounding variable offers an alternative explanation for why two things appear to be related. A mediator is different. It sits somewhere in the pathway between the things being studied. For example, if problematic social media use really does contribute to later attention problems, then maybe part of that effect occurs because late-night use disrupts sleep, and the sleep disruption then makes attention more difficult. In that hypothetical pathway, sleep disruption would be a mediator.

The researchers themselves raised sleep disruption, multitasking, and heightened reward craving as possible mediators. But they didn’t measure them directly. So the study gives us evidence about the relationship between problematic social media use and later ADHD symptoms, while leaving open the question of what processes, if any, produced that relationship.

That difference is easy to lose sight of, because the proposed mechanism or explanation can sound more psychologically satisfying than the original finding. Someone hears that greater problematic social media use may have predicted slight increases in later ADHD symptoms, then they hear that constant novelty might affect attention, and the two pieces feel like they fit perfectly into a neatly packaged conclusion.

But that conceptual fit is absolutely not the same thing as evidence that the study tested or proved that proposed mechanism.

This is especially relevant in neurodivergence research, because a lot of theories already circulate in clinical and popular culture. Dopamine, reward sensitivity, masking, executive functioning, nervous system activation, sensory overload, and trauma can all become explanatory shortcuts. In some cases, research genuinely does support those explanations, and sometimes they really are useful clinical formulations. But sometimes they are, at best, unproven hypotheses.

What I Find Myself Checking When I Read Research

Every piece of this has taken on great importance for me as I’ve started obsessively learning more about my own neurodivergence and that of my clients, so that I might better support them. The whole reason I started writing these Research Notes articles is because I want to find out what the research really says versus the way it’s sometimes portrayed in the media, and I hoped I might be able to shine a light on research that might interest some of my clients.

As I’m doing so, I have to decide what to leave out. I have to choose which numbers deserve space in the article, which methodological details could change an interpretation, and how much terminology a reader might need before my own article becomes harder to follow than the research paper itself. So yes, I’m vulnerable to the same sort of compression of nuance or misinterpretation of results as anyone else writing about research. The temptation can be insidious even if it’s not obvious: research becomes much easier to explain once it fits into a neat conceptual story.

There are a few basic questions I keep returning to when I’m reading a study, in an effort to keep my understanding accurate, and to hopefully communicate it more accurately.

What was actually measured? A clinical diagnosis, a symptom scale, a self-report, a parent report, medication records, or something else?

Who was included? Adolescents in one longitudinal cohort might tell us something important, without being misinterpreted as representing all people with ADHD.

What kind of inference does the research design actually support? Knowing that one variable came first can strengthen some conclusions, but confounding variables could still change the interpretation, and possible mediators may remain untested.

How large was the observed relationship between the things being studied? Statistical significance is important, but sometimes it can nonetheless represent a very small effect.

And when the authors explain why they think something happened in their research, did the study itself actually directly test that explanation?

None of these questions is particularly new or groundbreaking. But they are basic parts of reading research carefully, and I’m trying to make them more automatic in the way I read and write about studies.

What Pattern Shift Research Notes Is Trying to Do

The reality is, academic research wouldn’t be all that helpful if every finding ended with just a list of limitations and a shrug.

Observational studies can identify patterns, which later research needs to explain. Small effect sizes can matter when millions of people are exposed to something. Longitudinal studies can narrow plausible explanations, even when they may not be able to prove causation.

At the same time, there can be uncertainty in almost every part of a research claim. A single study could show fairly strong evidence that two things are associated, but might give only weaker evidence about the direction of that relationship, and maybe little evidence at all about why it occurs. Treating the whole package as either of two extremes, proven fact or useless speculation, risks losing a lot of information that is important to know.

This is the part of research interpretation I’m most interested in for these Research Notes articles. When I write about a study or group of studies, I want readers to hopefully come away with the same conclusion the evidence actually supports. Sometimes that means explaining why a seemingly exciting result is actually narrower than it first sounds, and sometimes it means defending a meaningful finding from being dismissed simply because it came from observational data or because the effect size was modest.

Is it possible I’ll make the same mistakes I am ranting about here? Sure it is. I still vividly remember the sensation of beating my head against a wall in a grad school statistics course, and wondering why I felt so smart about some things and dumb about others. But I’m taking this opportunity to grow my own understanding of research, and trying to point out ways that my clients and readers can see through the occasional incorrect media portrayal of a study and see what’s really going on underneath the hood. If I get something wrong, I hope I’ll have the opportunity to correct it.

Thanks for reading.