As much as we hear about, talk about, and think about screens, they remain incredibly confusing. The volume of information coming at us — from traditional media, research, social media, Substack, our kids’ schools, our kids — is overwhelming. We’re often looking for a way to organize what we see and read, which is part of why there was so much excitement about this paper from JAMA Pediatrics.
The paper is a review of existing studies of the relationship between screen time and child outcomes (cognitive and behavioral) that tries to separate out effects of different types of screen time. The authors argue that some kinds of screen time (watching together with your kids) are good, but other kinds (having background TV on, watching age-inappropriate content) are bad. Much of the coverage of the review seemed to celebrate the idea that — finally! — we could understand how to use screen time in the “right” way.
When I read the paper, I did not have this typical reaction. Instead, it made me more convinced that all of the studies we see of screen time are deeply flawed. Let me break down what I think is going on, and what we really learn from this.
The basic problem with studying screen time
There are many (many) studies of screen time and child outcomes, mostly focusing on cognitive outcomes (measured by test scores) and psychosocial outcomes (e.g. behavioral issues). The conclusion of these studies takes the form of “Screen time is associated with an increase/decrease/no change in outcomes.”
Nearly every study you see of screen time is based on an observational design. This means that the way the data is collected is by surveying people, and when we are exploring the effect of variation in the amount of screen time, it’s based on the natural variation that occurs in the world. This is distinct from a randomized study, where researchers would choose some children at random and assign them to be exposed to different amounts of screen time.
The overall concern with observational data is that many family characteristics move together. If we observe that test scores are lower for children who engage with more screens, is it really the screens, or is it the fact that there are other differences across families that may have an impact on test scores? This problem becomes very large when there are features of the family that the researcher cannot observe or measure.
To dig into this a little more, think about the visual below. There are three things represented here: a measured test score, a measured amount of screen time, and parental ability. Parental ability — think of this as something like IQ — is almost never observable to researchers. Test scores and screen time are things we can measure.
When researchers study screen time, they are interested in isolating the red arrow — the direct impact of screen time on ability. Isolating this is key to any actual decision-making. When we talk about using data to make decisions about screen time for our kids, we want to be making decisions based on the actual causal impact of screen time (meaning we want to make decisions based on the red arrow only). It would be easier if we could get a simple answer, like: If my child watches over X amount of TV per day, their test scores will go down. Great! Now I know not to exceed X amount of TV per day. No problem.
But in reality, it’s not that simple.
The problem is that the relationship represented by both of the blue arrows in this picture also exists. Parental ability clearly translates to child ability; this is well known. In addition, we have countless pieces of evidence suggesting that parents with a higher IQ have kids who have less screen time. With this insight, we can learn a bit more about the problem. The graph below is the same as above, but I’ve labeled the arrows with their expected signs.
- Higher parental ability → higher child test scores
- Higher parental ability → lower screen time
Imagine that the actual impact in the red arrow was zero — that, in fact, screen time had no direct impact on test scores. If you were able to estimate the causal impact of screen time on test scores in that case, you’d get a zero. However: if you were to use observational data to compare kids with more screen time to those with less, you’d conclude that screen time lowers test scores because of the blue arrows. Because parental ability is — this is a technical term — a confounding variable, you’ll observe more screen time associated with lower test scores even if there is no actual causal impact.
Essentially, it’s not just screen time that impacts test scores. Parental ability plays a huge role. And, more often than not, lower parental ability is associated with higher amounts of screen time, which in turn can lower test scores. To talk about the direct impact of screen time on test scores without including parental ability is leaving out a big part of the puzzle.
So far, so good. This is a version of an argument I have made many times in this newsletter. I talk much more about it in the post about why I look at data differently, including why in general approaches to controlling for these confounding variables are incomplete.
Evidence on types of screen time
Now let’s move on to the review I mentioned earlier, which separates out kinds of screen time. The review is very comprehensive. The authors pulled together basically all of the observational studies of screen time, classified details of the screen time use, and then combined studies that focus on different types of usage. They are therefore able to show impacts of different types of exposure to screens.
There are a number of results, but here are three highlights.
- Having television on in the background is associated with lower test scores.
- Exposure to age-inappropriate content is associated with worse behavioral outcomes.
- Co-watching television with children is associated with better cognitive outcomes.
Coverage of this study focused on the message “There are good and bad ways to expose your child to screens.” But we can also read this result through the lens of the discussion above. In particular, let’s think about drawing the same diagrams, but now instead of amount of screen time, using type of screen time (i.e. “watching age-inappropriate content” and “co-watching screens”).
To go back to the bullets:
The difference in these two diagrams is the sign of the impact of parental ability on the screen outcome. I am arguing here — based on other things we know about impacts of education and income — that parental ability is negatively associated with watching age-inappropriate content and positively associated with watching screens together.
- Higher parental ability → higher child test scores
- Higher parental ability → less likely to have child watch inappropriate content
- Higher parental ability → more likely to co-watch with child
Why would the relationships look like this? It could be for many reasons. One is information. Parental ability is associated with education, income, and resources. This group is more likely to have guidance about age-appropriate content, and also more likely to have heard from various sources about the value of watching television together. For this argument, the “why” isn’t important. What’s important — and, I’d argue, almost certainly true — is that this is the direction of these relationships.
Consider again the world in which neither type of screen exposure drives child test scores — both red arrows are zero effect. Because of the different signs of the blue arrows, though, if you actually look at the relationship between these measures and test scores, what you will see in the data is a positive relationship with co-watching (because both blue arrows are positive) and a negative relationship with age-inappropriate content (because one blue arrow is negative and one is positive). But this happens only because of the other variable, not because of any difference in the causal effects (which we have assumed are both zero).
Some of this discussion gets a little technical, but, at the core, the problem here is not complicated. These different screen behaviors link up with different family types. And it’s those differences that make some screen time look good and some look bad. It’s not the type of screen time itself.
Closing thoughts
Does this mean that infinite screen time of all types is fine? No, of course not. I’ve written many times about this — the post about a different way to look at screen time is a good summary — and my repeated argument is that you want to think about opportunity cost. Six hours of television a day is too much, but not because the screens are themselves bad, but simply because it doesn’t leave enough time for school, being outside, reading, socializing, or sleep.
Similarly, one of the conclusions here is that kids shouldn’t watch age-inappropriate content. That is completely sensible and reasonable, but not because of these data.
The bottom line
- Studies that focus on screen time generally establish an association between screen time and outcomes but cannot make causal arguments.
- It’s not necessarily the type of screen time (whether you watch with your child or not) that matters, it’s that different screen time behaviors are often reflective of the child’s environment.
- Screen time is something you should fit into your family time thoughtfully. But you do not always need to watch with your kids.
Log in
Does anyone know if there is validity to the studies that have found that kids under a certain age can’t learn from TV? I don’t understand why we expect they can learn from flipping through books but not from TV. Clearly kids even younger than 12 months respond to slap stick and obvious emotions in cartoons. They’re clearly “understanding” in some level. Is it just that no one has tried to study whether kids pick up vocabulary when cowatching with a parent under 18 months?
I’m pretty sure this is in response to parents putting their 18 month old in front of an “educational” iPad game or show and saying…”but it’s educational!” They can definitely be entertained by TV, but they’re not likely to pick up new vocabulary, shapes, colors, etc
Pet peeve: when programs should be age appropriate and are watched together with family (i.e. professional sports, the watching of which is a gateway to in-person interactions among family and peers, analysis, ice-breaking conversations, bonding with a tribe, motivation to participate in healthy activity, etc.) but are punctuated by incredibly violent commercials for movies or crime shows.
I understand that there are probably studies correlating parental IQ with screen time (the article says “countless”), but I’m pretty disappointed in the choice to equate IQ with “parental ability” in your explanation of this paper. IQ scores are indicators of some things (i.e., they are associated with academic and professional success), but to use them synonymously with “parental ability” is giving them far too much weight. I know that it is just an example to illustrate that correlation can be generated by a third, often unmeasured, variable–and I like the explanation of the directed acyclic graph. I just think that using IQ as a proxy for competence at something as complex as parenting is a poor choice.
I think “parental ability” here doesn’t mean parenting ability—it looks kike it’s a technical term for the parent’s cognitive ability, and that’s why “IQ” is a reasonable synonym.
I’m not sure that it’s a technical term, I think it’s just a vague catch-all and IQ was the chosen example, but I might be wrong. I just think that given that IQ scores are context and culturally dependent, reflect competence in only a specific socially constructed dimension, and have been used in really harmful ways in the past they should not be casually substituted for parenting ability.
I hear what you’re saying, and I, too, had this initial response to ability = IQ. As she goes on, she talks about how “parent ability is related to education, income, and resources” so it is more complicated than just IQ. I think it could just as well be equated with more measurable characteristics such as parental education, income, or availability/work status (is the parent able to spend time at home vs. needing to be out of the house to work) along with more nebulous qualities like emotional availability/maturity, parenting experience or interest level. I can see why just calling it IQ seems more efficient, but I agree that it is unfortunate that she chose to go that route rather than a more nuanced description.