For years parents have been bombarded with dire warnings about the need to ensure their children get enough sleep, starting in infancy. If they don’t reach a magical number, awful things will happen. Your child will be stupid, act aggressively, struggle to make friends, end up an entitled brat, and will surely never sleep well for the rest of his/her life. It’s horrible. Because of these warnings, parents are told a variety of things about how to ensure their children get enough sleep, like never nursing your child to sleep or co-sleeping, despite being biologically and anthropologically normal. Even more than what parents shouldn’t do, there is a push for things, like extinction sleep training (i.e., cry-it-out and controlled crying methods), that are supposed to help your child develop “healthy sleep habits”. If you do the wrong things, your child’s sleep will forever suffer, but follow these experts and your child will have healthy sleep habits for life. Of course, all this is premised upon the fact that “healthy sleep” = “reaching the magical number of hours of sleep that we are told to get” and of course we know how much sleep that is because we have guidelines, right?
Turns out, not really.
New research from – of all places – the Murdoch Institute in Australia (yes, the people that brought us the bad science attempting to show no long-term effects of controlled crying) set out to show that the amount of sleep children got on a regular basis was intricately tied to their behaviour and social well-being. These researchers – who seem quite keen on making sure their recommendations of sleep training and lots and lots of sleep for kids is heard – really seemed to have a preconceived notion that they were going to find a strong link between amount of sleep and behaviour. As said in their page about the research project (before publication of results):
“Current guidelines recommend that 3-5-year-olds should sleep 11-13 hours a night and 6-9-year-olds should sleep 10-11 hours a night. But while those are the official guidelines, they’re based on average sleep patterns, not how much sleep children actually need… Every age range had big differences. These big differences in children’s sleep make a big difference to their families’ lives.”
Note the assumption that the differences in age ranges must have a big difference in family lives. After all, this assumption is the basis for the myriad recommendations given to parents on a regular basis.
Much to their surprise, these researchers found that the amount of sleep children got actually had no bearing on any of their outcomes. There were quite a few outcomes assessed, including child mental health, child health-related quality of life, child BMI, child waist circumference, child learning (including language, literacy, and mathematics), child approach to learning, and parent mental health (both mother and father). As each of their measures included clinical cut-offs, they were able to analyze the sleep duration for each age range (4-5, 6-7, and 8-9) to see if sleep duration was different for the two groups (clinical versus non-clinical) as well as analyzing the data linearly to determine if there was a threshold for sleep duration that predicted problems. Yet none of it mattered, even when taking into account confounding variables.
So what can we say about this research and what it means? Let’s dig a little deeper…
Is this “good” or “bad” science?
Based on the recent piece I did on “Bad Science”, it’s worth examining the main questions that determine such a conclusion, especially as the research comes from a group that has had some bad science in the past.
- Did the research assess what it claimed to assess in terms of how they measured their variables? In this case, the answer seems to be yes, though arguably they could have looked at more outcome variables, but given they expected to see strong effects, what they were looking for was covered. The measures used were good enough for a large-scale study, especially the child measures as they were well-validated and commonly used measures. The researchers also included known confounding variables like socio-economic status, child sex, two-parent household, and English spoken at home. Even better, the sleep variable was assessed prospectively which removed the risk for any bias in reporting retrospectively.
- Did the researchers use appropriate statistics to analyze the data? Here is where I would have possibly included different analyses, but what they didn’t wasn’t inappropriate given their particular bias in their research question. They used mean comparisons between the two groups (clinical versus non-clinical) and controlled for confounding variables, all good. They also ran analyses to see if the effects of sleep duration influenced any of the outcomes irrespective of clinical status (i.e., to see if there was a threshold for sleep duration that predicts an increase in problems, even if not a clinical cut-off). Personally, I would have included a cluster analysis as well. This would have allowed the researchers to determine if there was a subgroup of families for whom there were problems associated with sleep duration and what the characteristics of that subgroup are. However, as the researchers were more interested in global effects (or rather, expected global effects), I don’t think finding small subgroups was on their radar.
- Did the researchers use the right sample for the question of interest? Yes, though there are limits that they had no control over. The researchers had access to a huge sample of Australian children through the Longitudinal Study of Australian Children which follows a large cohort of children year by year. Although all families in the cohort were eligible for the study, not all took part in the sleep diary portion. There were significant differences in those who did the sleep diary versus those who did not on certain confounding variables, such that those that participated were more likely to have a higher education status, be native to Australia or New Zealand, speak English at home, have a higher household income, and be married. This limits generalizability as the group is less “at-risk” for other well-being and health problems that are associated with some of these factors (like lower SES), but for the purpose of the question of how sleep influences certain outcomes, this may be a positive, something I will get into later.
All in all, I say we have “good” science here for the question at hand which assumes global effects. It does not, however, provide us with information about how sleep may influence certain specific children or individuals and it says nothing about how the researchers are taking this data to make any recommendations.
Were there any significant relationships?
Yes and although the researchers actually try to suggest the findings are meaningful, it is difficult to see them as practically significant. You see, the researchers had a very large sample of between 3434 and 3631 individuals for a given variable (the same children were used for analyses across the three age waves but not all had all outcome variables assessed). With such a large sample, getting statistical significance is actually quite easy (which speaks volumes about the lack of statistical significance found in most of the study).
The area the researchers highlight as being potentially important is mother’s mental health, where the sleep duration of the clinical and non-clinical groups differed at ages 4-5 and 8-9 (but not 6-7). However, the differences were miniscule with the sleep duration average for the non-clinical groups being 11.1 hours over a 24 hour period at ages 4-5 and 10.3 hours at ages 8-9 versus 10.9 hours at 4-5 and 10.2 hours at ages 8-9 for the clinical group. This is an average difference of 12 minutes over a 24 hour period. In fact, for all outcomes in the mean comparison analysis, the largest mean difference was 12 minutes and in some cases, like child mental health at 6-7, there was a significant difference between the two groups, but with the clinical group having an extra 12 minutes of sleep over the non-clinical group.
At the end of the day, these few significant differences are likely spurious and of no practical value. This is further confirmed by the scatterplots presented in the paper which demonstrate absolutely no relationship between sleep duration and the various outcomes at various ages.
How does this fit with other data that shows a link between sleep and behaviour or well-being?
In short, it doesn’t support previous findings. As mentioned above, one of the advantages of the current research over the vast majority of research linking sleep duration and well-being in children is that the measure of sleep is prospective and not retrospective. If a child is showing behaviour problems, it can be easy to underestimate how much sleep they are getting if we believe that may be a factor. The current research also used a large cohort which usually makes the findings more generalizable, though the higher SES and marriage rate may not be reflective of families in other countries.
This is where I wanted to mention the positive of the study being more comprised of a lower-risk sample. Specifically, this helps avoid the issue of third variable problem. In many of the previous studies looking at the effects of sleep, there are other factors that are at play, including poverty, family stress, health issues, and so on (see  for a discussion of some of these factors). It is possible that previous research has actually identified a canary in the coalmine with respect to sleep duration and not the problem itself.
That is, when there are other factors that may influence child and parent well-being (such as stress due to poverty, health problems, etc.), they are also likely to influence sleep. In these cases, sleep isn’t the cause of lower well-being, but simply one side effect of problems that affect well-being and sleep. Having a sample that has a decent mean income, is overwhelmingly married (89.3%), and so on means that some of these confounds are less likely to appear and we have a better chance of seeing the real effect of sleep per se. Which, on a global scale, turn out to be nothing.
So can we say that sleep duration really doesn’t matter?
Not really. When I said I would do a cluster analysis, it’s because I do believe there are groups of children for whom sleep is an issue, but like anything else, it is going to be widely variable. For example, there are children who will need 14 hours of sleep a night at age 4-5 and if they get less than that, they will suffer. Conversely, there will be kids the same age who only need 8 hours a night. This blurs the lines when we’re looking for large group effects. Like anything, we need to take a more individualized approach to sleep and children, with families being given guidelines on how to recognize signs of sleep deprivation instead of being given guidelines on total sleep that don’t consider the individual child at all.
Where does the research go next?
The researchers here are still looking for global effects, only now moving their efforts toward bedtimes and night wakings instead of total sleep duration. Again, they show their desire to find a one-size fits all approach to sleep, a desire that makes me frustrated.
Personally, I’d go in a very different direction. I would take this data and try to determine the range of normal sleep times and how individual fluctuations in children’s sleep influence these outcomes. That is, given a specific child’s optimum sleep level (whatever it may be), how far below do we need to go to see negative outcomes? Does that too vary widely or is the degree of sleep deprivation (either in proportion of sleep or total hours) that can cause negative outcomes consistent across children, regardless of optimum sleep time?
Also, I would want to look at interactions between variables to determine if certain factors can make a given degree of sleep deprivation result in worse outcomes. For example, does consistently getting 20% less sleep than a child’s given optimum level (for that child) mixed with malnourishment from poverty result in qualitatively worse outcomes than just the malnourishment or just the sleep deprivation?
Finally, I would want to get rid of the assumption that sleep is causal, instead testing models in which it is an effect. This speaks to the issue of the third variable problem and better understanding how sleep is influenced by a variety of life factors – including stress, poverty, mental illness, circadian rhythm vs. modern life rhythm, and so on. It seems logical to me that sleep may be one of the first areas that we can see disruptions in that can help us help children earlier and without focusing only on symptoms (like sleep), instead working towards addressing the underlying problems.
What does this mean for recommendations like extinction sleep training?
The focus of most sleep recommendations in infancy are under the guise of trying to make sure infants get enough sleep because of all the adverse outcomes associated with too little sleep. There is already research out there on normal infant sleep that counters this view, showing us that infant sleep is highly variable while still being completely normal, but this research adds to that which shows that even if we are concerned with later sleep habits and well-being, the link is possibly absent. Despite not having research that suggests sleep “problems” in infancy predict lower sleep duration and poor outcomes in childhood, people have made this claim in an effort to promote a variety of sleep training techniques. Now we have research that says these issues in childhood do not seem to be linked to sleep duration at all, and previous research may reflect the third variable problem or hindsight bias. In short, we now have even fewer reasons to believe anyone who tells us we need to use extinction sleep training on our infants.
What’s the take-home message?
Unfortunately, the media and researchers seem to feel the take-home message should be that the “amount of sleep doesn’t matter”. I disagree (although I agree with the researchers that we should scrap the sleep guidelines as they only tend to freak parents out and we now have evidence they aren’t based on anything but averages). I feel the take-home message has to be that each child is unique and as we all know from experience when we ourselves don’t get enough sleep, it does affect us. The issue is that what isn’t enough for me is different from what isn’t enough for you. In this attempt to treat all children the same, we are ignoring that they simply aren’t the same. And that’s okay.
I hope going forward we can help educate parents about the signs of sleep deprivation, but also how what is going on in our lives influences our sleep. When faced with a child who is suffering socially, emotionally, behaviourally, or health-wise, it’s essential that we look to everything that is going on with that child and not just at something like sleep. Only then can we actually help our children in the best way possible.
——————————– Price AMH, Quach J, Wake M, Bittman M, Hiscock H. Cross-sectional sleep thresholds for optimal health and well-being in Australian 4-9-year-olds. Sleep Medicine 2015; doi: 10.1016/j.sleep.2015.08:013  Cassels TG. ADHD, sleep problems, and bed sharing: future considerations. The American Journal of Family Therapy 2013; 41: 13-25.  Weinraub M, Bender RH, Friedman SL, Susman EJ, Knoke B, Bradley R, Houts R, Williams J. Patterns of developmental change in infants’ nighttime sleep awakenings from 6 through 36 months of age. Developmental Psychology 2012; 48: 1511-1528.  Hysing M, Harvey AG, Torgerson L, Ystrom E, Reichborn-Kjennerud T, Sivertsen B. Trajectories and predictors of nocturnal awakenings and sleep duration in infants. Journal of Developmental and Behavioral Pediatrics 2014; 35: 309-16.