Three times in as many days, I have encountered articles that reference this 2012 study of sleep training and infant-maternal cortisol levels. The study got a lot of positive press when it was published, and is oft cited by sleep training opponents as “proof” of the damaging effects of sleep training on babies. The trouble is, the study doesn’t prove anything, least of all what the people citing it would like their audience to think it proves.

If you are unfamiliar with the study, here is a quick summary:

Researchers followed 25 infants (four to 10 months old) over a five-day sleep training program, during which the extinction method (putting the baby to sleep and not responding to any crying) was used. Cortisol levels of both mother and baby were measured on the first and third days of the study, before the beginning of the bedtime routine and again 20 minutes after the baby had fallen asleep.

On the first night of the sleep training program, babies exhibited behavioural distress (crying). Both mothers and babies had elevated cortisol levels. On the third night of the program, babies no longer exhibited behavioural distress, yet their cortisol levels remained elevated. Mothers’ cortisol levels decreased in the absence of babies’ crying.

The conclusions outlined in the study’s abstract suggest the following:

1. That sleep training causes a stress response in babies even if behavioural distress (e.g. crying) is not exhibited;
2. That sleep training causes a disassociation between infants’ behavioural and physiological responses; and
3. That when an infant fails to exhibit a behavioural response to distress, the infant’s and mother’s cortisol responses become asynchronized (the implication being that the mother is no longer attuned to the baby’s distress).

Those are some pretty heavy conclusions with pretty heavy implications, no?

Now I don’t have a science background, but even I can find a bunch of huge, gaping flaws in this research, to the point where I am baffled that such a study ever made it past the peer review process to be published in a medical journal.

Here is why the Middlemiss study does not prove anything:

The sample size is too small. We simply cannot draw large-scale conclusions from a sample size of 25 mothers and infants. Imagine if I were to go out and conduct a poll of 25 random mothers to find out whether or not their babies were sleeping through the night at one year of age. Suppose that 12 of these mothers reported that their babies slept through the night, and I subsequently published an article stating that according to my research, 48 percent of babies sleep through the night by one year of age. Can you imagine the ridicule I would face?

There is no control group. The study has a group of mothers and babies being observed and tested while undergoing sleep training in a laboratory environment. It does not have another group of mothers and babies simultaneously being studied while undergoing sleep training in the home environment, nor a group of mothers and babies being studied in the sleep lab whilst not undergoing sleep training. I could fill an entire blog post with reasons why this is problematic, but suffice it to say that without controls, the data obtained from this study is as good as useless, because we have no way of knowing which variables other than sleep training may have contributed to the results obtained, and to what degree.

Cortisol levels are a clinical finding only, and do not necessarily demonstrate that harm is occurring. Cortisol is produced in the adrenal gland, not only in response to stress, but also in response to circadian rhythms. There is a natural ebb and flow of cortisol throughout the day, with peak levels occurring around 8:00 AM, and lowest levels occurring between midnight and 4:00 AM. Acutely elevated cortisol occurs in response to temporary stress (which includes not only negative stress, but also “positive” stress, such as exercise). Chronically elevated cortisol occurs in response to ongoing stressors, such as post-traumatic stress and cumulative sleep deprival.

The study makes no distinction between acutely and chronically elevated cortisol levels. It is chronically elevated cortisol, not acutely elevated cortisol, that is responsible for deleterious effects on child development. Harvard University’s Center on the Developing Child states, in a fact sheet about toxic stress, “The prolonged activation of the body’s stress response systems can be damaging, but some stress is a normal part of life. Learning how to cope with stress is an important part of development. We do not need to worry about positive stress, which is short-lived, or tolerable stress, which is more serious but is buffered by supportive relationships.”

Cortisol levels were elevated on the first day of the study, prior to the beginning of the bedtime routine. This suggests that stress (either acute or chronic) was present before beginning the sleep training process. Elevated cortisol levels may have been caused by a variety of factors, including mom and baby being in an unfamiliar environment (the sleep lab), handling of the baby by strangers (nurses), the mother’s anxiety, or even chronic sleep deprivation; but one thing is certain — the babies could not have been exhibiting elevated cortisol levels in response to a “cry it out” process that had yet to be initiated.

Cortisol levels did not show any significant increase pre to post-bedtime. These mothers and babies already had elevated cortisol levels prior to beginning the sleep training process, and the levels did not increase as a result of “crying it out” on the first day, nor on the third.

Cortisol levels were only measured at two time points — immediately before the bedtime routine, and 20 minutes post falling asleep. There were no measurements done throughout the night to determine whether levels remained elevated (in relation to circadian rhythm norms) once sleep was well established.

Cortisol levels were only measured on the first and third days of a five-day sleep training program. Yes, they simply stopped measuring and recording data 60 percent of the way through the study. Perhaps on day four, the babies’ cortisol levels returned to normal and maternal-infant cortisol synchrony was re-established? Or perhaps not. Either way, to draw any reasonably accurate inferences, at the very least there ought to have been measurements taken on the final day of a five-day study. This was a huge (perhaps deliberate?) oversight, as one cannot reasonably imply that there exists a disassociation between babies’ behaviour and hormone levels, or between mothers’ and babies’ hormone levels, based on data taken from a single point over a five-day period. It seems irresponsible to suggest this, without having at the very least followed those measurements through to the end of the study.

Periods of temporary disruption in mother-infant endocrine synchrony happen over the course of normal infant development. They are not known to be harmful to the development or maintenance of secure attachment, and are part of the adaptive process.

Nurses were responsible for putting the babies to sleep. Does anyone else see how being exposed to a brand new sleep routine, in an unfamiliar environment, with a stranger — rather than a loving, attached parent — responsible for putting the baby to bed might cause him to have increased cortisol levels at bedtime? And how removing the mother from the bedtime routine might contribute to asynchrony in maternal-infant cortisol levels?

This was a poorly designed and even more poorly executed study that seems completely engineered to support a specific point of view, without any regard for the validity of the scientific process.

I have seen writers cite this study as evidence that controlled crying is harmful (remember, the study was actually conducted using the extinction, or CIO method, not controlled crying), and as evidence that babies who stop crying during sleep training are  still experiencing distress (for reasons I outlined above, this study proves nothing of the sort, especially when you consider that cortisol levels didn’t even increase during the sleep training process!).  It has also been used to argue that sleep training causes mothers to become detached and unresponsive to their babies’ needs. Unfortunately, the study doesn’t prove any of those arguments. I’m not even sure that this study provides enough valid data to form the basis for further and more large-scale research.

The takeaway from all this?

1. People misappropriate research studies all the time, in order to give the illusion of authority and lend credence to their point of view. Just because somebody cites a study and provides an official-looking footnote at the bottom of their article does not mean that the study referenced actually reflects the opinion they are trying to argue. Don’t rely on the title of the study — check PubMed and read the summary.

2. Research studies must always be viewed with a critical eye, and their methodologies questioned. Just because a study has been published, or has been the recipient of media attention, does not make it legitimate. Indeed, the vast majority of studies do not prove anything at all; they merely provide an argument for the basis of further research.

3. Always consider opposing points of view. There can be much to learn from them, and they can help with critical analysis.

Finally, while many parents are not proponents of the extinction method of sleep training, I have to admit that one of the thoughts that briefly crossed my mind after dissecting this study was, “Wow, they sleep trained 25 infants using ‘cry it out,’ and by the third night all of them went to sleep on their own, without crying?” But of course, by my own admission, a sample size of 25 infants can hardly be extrapolated to the population at large🙂.

Image courtesy of www.failblog.org

Image courtesy of http://www.failblog.org


  1. Nice review. I’m curious, though – if you’re not a proponent of the extinction method, what do you recommend? I’m about 28 weeks pregnant and gathering info about sleep training. Thanks!

    • Hi Shawna,

      For Oliver, we used the “check and console” method, and it really worked well. It takes longer than extinction, and it is more work in the middle of the night — at least for a few nights — but for us, it struck the right balance between firmness and responsiveness. The method outlined in Dr. Ferber’s book (Solve Your Child’s Sleep Problems) was essentially what we used. If you’re doing lots of reading, I highly recommend that book. He gets a bad rap in some circles, but it’s an excellent book and very, very humane.

      Another book I liked was The Baby Sleep Solution, by Suzy Giordano. Unfortunately, we never really followed her advice (which is why we ended up paying to hire a sleep consultant), but from what I know now, it would have worked well, if only we’d followed the plan🙂

    • I’m afraid a number of your analysis is not supportable simply with a cursory reading of the actual Abstract of the study (and not your incomplete synopsis). For one thing, the measurements were not taken only day 1 and day 3. The abstract clearly states that days 1 and 2 showed the expected/normal responses and beginning on the 3rd day, the asynchronous cortisol levels began to be observed. You also claim that the 1st day of the study *prior to commencement of the CIO training*, the cortisol levels were increased. From the abstract, they did not say this, they said that on the first day, the synchronous levels of cortisol were seen (nothing about BEFORE the CIO training began). Additionally, the cortisol levels would be higher for both infant and mother in the case that this was performed in a laboratory (as you say causing discomfort) but the levels would be *synchronous*.

      The other points you present (lack of a control group, environment of the study in a lab, size of population) can be of concern. The paper passed peer review likely because experimentation is expected to be conducted in a controlled environment (IE not an individual’s home) and when you have unanimous results from a small population, a larger population is not required to verify the results especially if non-numeric evaluations are concluded (ie “cortisol levels become asynchronous as a result of this study” and NOT “100% of infant/mother pairs that experience CIO will have asynchronous cortisol levels”). Finally, you make a lot of scientific claims in your analysis for someone who has no real technical background on the subject so I have to conclude that you did some cursory research when responding to this study. HOWEVER, did you do this objectively, or did you search for references that support your predetermined conclusion that this study is flawed? Is it likely that I could dig up some scientific support for the infallibility of this study? It is likely.

      I think we’ll continue to respond to our baby when he cries at night since your analysis is not really convincing… certainly not enough to chance it IMO.

  2. Pingback: Biological, not logical | April Moments·

  3. Pingback: Is Sleep Training Child Abuse? - Troublesome Tots·

  4. Pingback: 9 Things Mothers Who Choose CIO Don’t Want to Hear | The Littlest Dodd·

  5. Great post! It’s worth noting that her statistics are also a mess. She uses the wrong analyses, and is missing over half her post-sleep sample for the third night. So her claims in fact rest on a sample of 10 mother infant pairs. Just 10.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s