Why the EEG studies disagree

Search for evidence on brainwave audio and you will find studies reporting large effects, studies reporting nothing, and meta-analyses that hedge. That is not a sign the field is being suppressed. It is what a young literature with methodological problems looks like from outside.

Five problems account for most of the disagreement.

Sample sizes are small

A great many studies in this area run twenty to forty participants. At that size, a real but modest effect is easy to miss, and a spurious large effect is easy to find. Both happen, and both get published.

When you see a striking result from a small sample, the appropriate response is interest, not belief. The question is whether it replicates at scale, and for most claims in this category the answer is that nobody has tried.

Blinding is genuinely hard

You cannot give someone a placebo binaural beat that sounds like a binaural beat and is not one. Control conditions tend to be silence, unmodulated tone, or a different frequency — and participants can often tell which arm they are in.

This matters more here than in most fields, because the outcomes are frequently self-reported mood, focus or relaxation. Expectancy has a clear route to the result. Placebo and expectancy covers how large that route can be.

The protocols are not comparable

Two studies both described as "10 Hz alpha entrainment" may differ in carrier frequency, modulation depth, delivery method, session length, whether the rate was fixed or swept, what the participant was doing, and how long after the session the measurement was taken.

Pooling those into a single meta-analytic estimate produces a number that is precise and close to meaningless. When a review reports a small average effect with high heterogeneity, the heterogeneity is usually the more informative half.

This is one reason we publish the complete specification for every session rather than a band label. A session described only as "alpha" is not a reproducible thing.

Outcome measures drift

If a study measures six outcomes and reports the two that moved, the reported effect is not what it appears to be. Pre-registration addresses this and is not yet standard in this literature.

Watch for papers where the headline finding is a subscale of a questionnaire rather than the primary measure the study was built around.

Publication bias runs one way

Studies finding an effect are more likely to be written up, submitted and accepted than studies finding nothing. In a field where a fair number of well-run experiments genuinely find nothing, the published record will overstate the effect regardless of anyone's intentions.

How to read a claim from here

A workable checklist:

How many participants, and was it pre-registered? Under thirty and unregistered is a hypothesis, not a finding.

What was the control condition, and could participants tell? Silence is a weak control.

Was the outcome objective or self-reported? Both are legitimate. They warrant different confidence.

Did the effect outlast the stimulus? Effects measured during stimulation are much easier to obtain and much less useful.

Who funded it, and does anyone sell the thing? Including us. This site sells sessions, which is a reason to check what we say rather than take it.

What we do with all this

We describe the signal precisely, we describe the association between a band and a state as an association, and we do not claim outcomes the literature does not support. The science page sets out that standard and links the sources; the research page sets out what we would need to see before making stronger claims.

Sessions like Slow Gamma Maintenance are named for what they are rather than for a result, because we cannot promise the result.