How to read a small study without overreading it

We publish a research library that grades every study we cite. Grading is only useful if you understand what the grades mean, so this is the reasoning behind them.

It's also, deliberately, an article that will make you harder to sell to. That seems like a fair trade.

Why sample size matters more than people expect

Flip a fair coin ten times and get seven heads. Unremarkable — that happens roughly one time in six.

Flip it a thousand times and get seven hundred heads and something is genuinely wrong with the coin.

Small studies are the ten-flip version. They produce dramatic-looking numbers that mean very little on their own, not because the researchers did anything wrong, but because with few participants ordinary human variation is large compared to the effect being measured.

A finding in twelve people is a hypothesis wearing a lab coat.

This matters for grounding specifically, because most of the literature sits between eight and sixty participants. One widely-cited pilot on muscle soreness had four people in each group.

The multiple-comparisons trap

Here's the part that catches almost everyone.

If you measure one outcome in eight people, you probably won't find anything. If you measure twelve outcomes in eight people — white blood cells, neutrophils, lymphocytes, bilirubin, creatine kinase, a couple of ratios, subjective pain — you will almost certainly find that at least one of them looks impressive.

Not because grounding did something. Because you rolled twelve dice and reported the sixes.

This isn't fraud and it usually isn't even intentional. It's a well-documented statistical trap, and the defence against it is pre-registration: declaring in advance which outcome you're testing, so you can't pick the winner afterwards. Very little grounding research is pre-registered.

What to look for: how many things did they measure, and did they say in advance which one mattered?

What blinding does, and why its absence is a problem

If you know you're being treated, you report feeling better. This is one of the most robust findings in all of human research, and it's strongest for exactly the outcomes grounding studies tend to measure: sleep quality, pain, mood, energy.

A blinded study means participants don't know which group they're in. A sham-controlled study gives the control group something that looks and feels identical but doesn't do the thing.

Sham-grounding is genuinely hard. The sheet has to be indistinguishable, the cord has to look connected, and nobody in the room can know which is which. Most grounding studies didn't attempt it.

The important exception: a 2025 randomised, double-blind, placebo-controlled trial used an identical sham mat, ran 31 days with 60 participants, and measured sleep with actigraphy as well as questionnaires. That's why our library grades sleep Promising and nothing else.

One good trial is not a body of evidence. It's a good trial, and it's the reason to run more.

Markers versus outcomes

A lot of grounding research reports changes in things measured in blood: creatine kinase after exercise, red blood cell aggregation, cortisol in saliva.

A marker moving is not the same as feeling better, recovering faster, or living longer.

Sometimes markers track outcomes closely. Often they don't — the history of medicine contains a long list of interventions that improved a number and did nothing for the person, and a few that improved a number while making things worse.

What to look for: did they measure how people actually felt or functioned, or only what their blood looked like?

This distinction gets collapsed constantly in grounding marketing. A study of ten people over two hours showing changed blood viscosity becomes "supports a healthy heart." We don't do that, and the study in question is graded Emerging in our library with a limitation saying exactly this.

Who ran it, and who paid

Not a reason to dismiss a finding. A reason to want it replicated.

Researchers who believe in something are the ones who study it, and that's true across every field. But grounding has an unusually concentrated author pool, and several long-standing authors have had commercial relationships with the company that funded much of the work. A 2020 review found the studies most often cited by grounding proponents share overlapping authors, small samples, and rarely blind.

What to look for: has anyone unconnected replicated it?

For most grounding findings, the answer is not yet. That's the single biggest gap in the field.

A worked example

Take a study we cite: twelve participants slept grounded for eight weeks while cortisol was sampled. The authors reported night-time cortisol falling and the daily rhythm looking more typical, alongside better self-reported sleep.

Run it through the questions:

Question	Answer
Sample size	12. Very small.
Control group	None.
Blinded	No — participants knew.
Outcome or marker	Both, but sleep was self-reported.
Pre-registered	No.
Independently replicated	No.
Author independence	Overlaps the commercially connected group.

So: interesting, worth following up, and not something to base a purchase on. We grade it Preliminary, and the limitation on its card says you can't separate the effect from expectation or from simply paying attention to your sleep for eight weeks.

That's not us being modest. It's what the study supports.

The short version

Five questions, in order of usefulness:

• How many people? Under thirty, treat as a hypothesis.
• Was there a control group, and were people blinded? If not, expectation is doing unknown work.
• How many things did they measure? Many outcomes, one impressive result — be suspicious.
• Marker or outcome? A number moving isn't a person improving.
• Has anyone independent replicated it? This is the one that matters most and is answered least often.

Why we published this

Because you're going to encounter grounding claims elsewhere, and most of them are built on the studies above without any of the caveats.

If this article makes you more sceptical about what we sell, that's a reasonable outcome. We'd rather sell to someone who understood the evidence and bought anyway than to someone who was impressed by a number that didn't mean what they thought.

If you want the full picture, including everything that weakens the case: What we don't know yet.

Related: The research library · What we don't know yet · Editorial standards

These statements have not been evaluated by the Food and Drug Administration. This product is not intended to diagnose, treat, cure, or prevent any disease.