Research explainers

Clear explanations of anxiety science, including the uncertainty.

What anxiety studies measure, how strong their findings are, and what remains unanswered. Each explainer opens with what the evidence supports and ends with its sources and the date they were checked.

An open notebook of handwritten notes beside a printed graph, a pair of glasses and a brass balance scale on a desk

The ledger

From “Plausible Mechanism vs Proven Benefit”

Three popular claims, graded on the human evidence

Claim	What the human evidence shows	Our reading
GABA supplements calm anxiety	Unclear whether oral GABA reaches the brain; most supportive studies by researchers with potential conflicts.	Weak
Magnesium eases anxiety	18 studies in anxiety-prone groups; 4 of 8 in mildly anxious samples found benefit; overall quality poor.	Weak
Probiotics ease anxiety	34 trials: small effect (d = −0.10). A 2026 review of 49 anxiety trials: larger (SMD −0.43), but high heterogeneity and moderate quality.	Mixed

The mechanisms are plausible in every case. The question is whether they produce a change people notice.

Read how each was graded →

Latest explainers

All sources checked Oct 16, 2026

Claims examined · Mechanisms · Research literacy · Supplements

What Is the Difference Between a Plausible Mechanism and a Proven Benefit?

Why a convincing story about how something might ease anxiety is not evidence that it does, and what it takes to get from one to the other.

Why a plausible mechanism isn’t proof something eases anxiety: animal-to-human gaps, surrogate outcomes, and how GABA, magnesium and probiotics compare.

What the evidence supports

A plausible mechanism shows how something could ease anxiety; it doesn’t show that it does. In one review, only 37% of highly cited animal findings were later replicated in human randomised trials.

Published: Oct 6, 2026
Sources: 11

Research literacy · Study design · Causation

How Do Researchers Separate Correlation From Causation?

Why a link between two things rarely settles what causes what, which study designs get closer to an answer, and where each one falls short.

Correlation vs causation in anxiety research: why a link isn’t proof, and how follow-up studies, trials and genetic designs get closer to cause.

What the evidence supports

A correlation is real information but rarely settles what causes what. Randomised trials are the strongest single design for causal questions, yet participants’ expectations can still bias them.

Published: Sep 15, 2026
Sources: 9

Research literacy · Measurement · Questionnaires

What Does an Anxiety Study Actually Measure?

Questionnaires, interviews, body signals and lab tasks: what each one captures, how well it performs, and where it falls short.

What anxiety studies actually measure: how questionnaires like the GAD-7, clinical interviews, body signals and lab tasks perform, and what each misses.

What the evidence supports

In most studies, “anxiety” means a score on one specific measure. Before trusting a headline, check which measure was used, in whom, and over what period.

Published: Aug 25, 2026
Sources: 9

What each study design rules out, and what it leaves open

From the correlation explainer. Most causal conclusions about anxiety rest on several designs pointing the same way.

Design	Helps rule out	Leaves open
Cross-sectional survey	Very little on its own	Reverse causation, confounding, measurement artefacts
Longitudinal study	Some reverse causation	Unmeasured confounding
Statistical adjustment	Measured confounders, if measured well	Unmeasured or poorly measured confounders
Randomised trial	Confounding, on average	Expectation effects, poor adherence, limited generalisability
Genetic and family designs	Some confounding by genes or environment	Violations of the design’s assumptions

First

Every explainer opens with a short statement of what the evidence supports.

Last

Every one closes its main text with “What remains uncertain,” then lists numbered sources and the date they were checked.

Throughout

Hypothetical examples are clearly labelled. Real studies are cited by number, with their sizes where they matter.