My month-long sleep intervention experiment

Over the month of November 2025, I tested two different sleep interventions (mouth taping and nose strips) against a couple of variables on a single subject (myself):

• self-reported energy level in the morning

• self-reported mouth dryness in the morning

• Whoop tracker biometrics: deep sleep, REM sleep, sleep efficiency

Nose strips did nothing, but mouth tape produced two statistically significant results:

1. Mouth tape reduces self-reported mouth dryness.

2. Mouth tape reduces deep sleep by 30 minutes on average.

The first finding is quite intuitive: mouth tape stops dry air from entering the mouth.

The second finding is counterintuitive and suggests mouth tape might have negative effects on sleep quality. There is no effect on REM sleep.

I do not attempt to explain the effects further, and recommend a second, longer study in a regular sleep environment.

Experiment design

The core problem of study design is getting clean correlations between the proposed cause and the proposed effect, without any other variables confounding the relationship.

In observational studies such as the ones with surveys, where we do statistics on a population level, this is quite difficult because we cannot randomize variables. They will always be correlated with something else.

If we did a survey asking three questions: “do you use mouth tape”, “do you use nose strips”, “how do you rate your energy level in the morning”, this would be a mess. These variables will always be correlated with something else, like gender, age, or the answer to “do you sleep well in general?”. That means we have to ask additional questions beyond what we want to study, and control for any confounders.

Doing a controlled trial with a single subject actually resolves many of these issues, as we can randomize the treatment assignment and observe the effect. With enough data, and assuming no decisions made depending on the randomized treatment (e.g. the subject is not going to bed later due to using nose strips), the statistical analysis will come out perfectly clean.

The main downside of a single-subject trial is, of course, that scientists are interested in what is true for the general population, and the subject might be idiosyncratic.

To equidistribute the treatments across the nights, I used a “randomized complete block design” with 4 treatments: Mouth Tape, Nose Strip, Nose & Mouth, and Control. I repeated randomized blocks until the end of the study period, which was intended to be 28 nights.

The measurements I took were:

1. Self-reported energy level in the morning (1-10 scale)

2. Self-reported mouth dryness in the morning (none: 0, some dryness: 0.5, dry: 1)

The subject

The subject is a healthy male in his 20s with an inconsistent sleep schedule and daily caffeine use; no alcohol use. The subject breathes through the nose during waking hours. The subject had a persistent stuffy nose for more than a month before the experiment started, and had noticed dry throat in the morning and worse sleep overall in the same time period. The condition faded away after the experiment started.

Preregistration

The prediction before running the experiment was:

I expect mouth tape to produce statistically detectable positive effects on mouth dryness, sleep quality (as measured by Whoop), and energy levels; and nasal strips to not do anything.

Apart from this, I generated the schedule of interventions in advance.

Safety measures and other annoyances

Mouth tape seems like it could be unsafe for people who are not able to get enough oxygen through their nose.¹

Searching all over the Internet either does not give actual safety recommendations or expresses vague warnings on asphyxiation. One Reddit comment says “just tape part of your mouth, not like a hostage”. I believe the Reddit comment. I do the following safety measure: tape the mouth, not like one would do to a hostage, but with a part of the mouth free in case of emergency.

Nose strips pose no health risk except the occasional irritation on the skin. They are, however, very difficult to put on after applying skincare products such as a moisturizer.

Alterations to the preregistered intervention schedule

I skip the first 3 days of the preregistered experiment because of a combination of frequent naps and inconsistent sleep tracker use.

On two occasions, nose strips were not possible to put on because I did some facial skincare immediately before going to bed and was not willing to wait for it to dry. I switched the intervention for those nights and made it up with an additional nose strip intervention on the next night that was preregistered not to have a nose strip.

On two occasions, I tore the mouth tape off during the night (and was clearly sleeping without it). I label the night as having or not having mouth tape depending on whether I remember waking up without mouth tape during the night.

Controlling for confounders

With the intention to “control for” additional variables that could affect sleep, I tracked:

• time of last caffeine intake

• magnesium intake

• whether I went to the gym or not

Before writing this post, I noticed I did not actually know what it means to “control for” these variables. As my design was randomized to switch between treatments every night, and the above variables do not depend on the treatment assignment in any way (I don’t even know what the assignment for the night will be before going to bed), they are therefore by definition not confounding my measurements.

More plots!

Do I trust these results?

As there were multiple hypotheses tested, the significance levels need to be proportionally weakened. However, both main results have p-values of less than 0.01.

There are five interventions per setting, so the implicit significance level is under 0.05; I’m inclined to believe the findings would be directionally correct in the exact same setting.

It is a different question entirely whether the findings would reproduce in a different setting. What could change?

1. Location: I stayed at Lighthaven for most of it; the rooms there are quite different from my usual sleep environment.

2. Sleep schedule: The experiment took place during a 30-day writing residency. The period around the submission deadline (midnight every day) was often eventful, and thus the sleep schedule was necessarily offset.

3. Chemicals: Caffeine and magnesium act on sleep in important ways. This experiment was my first period of taking magnesium regularly. Different supplements and stimulants could change results.

4. Health state: I had a stuffy nose early on in the experiment.

5. Subject: Of course, as n=1, I do not have any data on whether another subject would experience any of the effects I observed. YMMV.

There is also the question of whether the two interventions (mouth tape and nose strip) can be treated as independent, as they are clearly acting on the same mechanisms (easier nose breathing through the night). To disambiguate this I would need to increase the sample size of the study and basically run the study with four treatments (Nose Strip, Mouth Tape, Nose & Mouth, and Control) instead of two.

Would I run a similar study again?

I find doing the mouth tape and nose strips interventions on myself to be particularly tiring, as I have to apply the intervention in the part of the day where I am tired the most. Measurement of effects was not particularly difficult, but ideally I would want to just use a sleep tracker.

What would I do differently? Definitely a different metric for “energy level” in the morning, as I find it unreliable. I would also print out the schedule of the interventions to avoid looking at my phone immediately before going to bed.

In the future I would like to do studies on other randomized interventions (caffeine ², melatonin, etc.), as long as I do not need to apply a randomized decision every night immediately before going to bed.

“You’re supposed to breathe in and out through your nose, but mouth taping is not safe at all,” said Kaninika Verma, MD, clinical sleep director at OSF HealthCare. “This is nothing a physician would ever recommend to a patient.”

1.

2. The effect of caffeine timing on sleep quality seems particularly interesting, and it could be studied without any changes to my routine, by using decaf coffee in the afternoon every other day.