How we rate evidence

The principle

The Balance of Health exists so that someone who disagrees with us can inspect the same evidence we used and understand exactly how we reached a conclusion. That means showing sources, saying how strong the evidence is, presenting the strongest counterargument, and being explicit about what is uncertain.

If the evidence is inconvenient for a plant-based diet, we show it anyway.

How evidence is structured

Every important statement on the site passes through the same chain, and each link is a record you can open:

Claim: a single, quotable statement.
Short answer: the claim in plain English.
Key statistic: one number, with its unit, geography, year and source.
Explanation and chart: why the number is what it is.
Evidence strength: a rating and, always, the reason for it.
Source records: what each source found, its limitations, funding, and which claims it supports or challenges.
Original research: a direct link to the paper, report or dataset.
Methodology and limitations: how the numbers were produced and what they leave out.
Last reviewed: when a person last checked the record.

How we rate evidence strength

We use four levels. The rating describes the evidence for the claim as stated, not how much we like the conclusion. A claim can be rated strong and still have large uncertainties in its exact size.

Strong evidence: Supported by multiple systematic reviews or large, consistent studies using different methods, with few plausible alternative explanations.
Moderate evidence: Supported by good-quality studies that point the same way, but with gaps: fewer studies, mostly observational data, or some inconsistency.
Limited evidence: Some relevant studies exist but they are small, indirect, short-term or conflicting. The direction is plausible; the size of the effect is not settled.
Uncertain: The evidence is too thin, too mixed or too dependent on assumptions to support a confident conclusion either way.

Special care with health evidence

Most nutrition evidence comes from observing large groups over time. That can show associations but struggles to prove cause, because people who eat differently also live differently. We say when a finding is observational, when trials exist, and what the trials measured.

We report relative risks alongside a sense of absolute risk, because a 25% reduction of a small risk is a small number. We do not treat a vegan diet and a whole-food plant-based diet as the same thing.

Special care with environmental evidence

Environmental footprints come from life-cycle assessments, which depend on system boundaries and allocation choices. We report averages with ranges where the source gives them, compare foods on a nutritionally sensible basis, and note where methods (for example, how methane is counted) change the numbers.

We do not treat all plant foods as environmentally equivalent, nor all animal farming.

Countries, production systems and personalisation

Every statistic and food figure carries a geography. When you choose a country, we use its figures where a source provides them and fall back to global averages otherwise, always saying which. Country records also summarise local dietary guidance, fortification practice, welfare law and water stress, each with its own source.

Some foods have figures for several production systems (beef from dairy herds, grass-fed, feedlot; free-range and indoor chicken). The comparison tool shows the all-systems average by default and lets you choose a system; the source changes with it.

Telling us what you are mainly interested in reorders what comes first on a page. It never removes evidence. The preference and country are stored as cookies on your device; there are no accounts.

Counterarguments

For each major question we present the strongest opposing argument, as its proponents would state it. We then say what is correct about it, what the evidence supports, what may be misleading, and what remains uncertain. We do not construct straw men.

Source records

A citation on this site leads first to our record of the source, not straight off-site. The record summarises the study, its type, population and sample size, key findings, limitations, funding and conflicts, and an assessment of its quality. From there, one link takes you to the original.

Sources can support some claims and challenge others. We record both.

Units you have a feeling about

Square metres per 100 g of protein is precise and almost meaningless. So alongside the scientific unit we show the same figure against something familiar: tennis courts of farmland, kilometres of driving, eight-minute showers. The scientific unit is always there. The familiar one is never a substitute for it.

Each reference object is a record with a source, listed with its factor and its limits on the calculator methodology page. A comparison of size is not a claim of equivalence: driving and eating are different things, and most of the water behind food is rain falling on fields rather than water you could have used at home.

Why we show what other people are doing

Where we have verified figures for how consumption is changing in a country, we show them, because a direction of travel is a fact about the world and facts about the world belong on this site.

We are careful about what that is doing. Research on dynamic norms found that telling people a behaviour is becoming more common increased the share choosing a meatless lunch, but a direct UK replication failed and an in-store trial found nothing. We record those failures on the source record. The trend is shown because it is true and checkable, not because we are confident it changes anyone's behaviour, and it is never presented as an instruction.

A trend statistic appears only once it has been verified. Until then the section does not render at all, rather than showing a placeholder number about what other people are doing.

Sample, draft and reviewed

The site is being built ahead of its research database. Our claims, sources, statistics, articles and questions are the records we offer for citation, and each carries one of three states, shown wherever the record appears, so you always know how much weight a number will bear.

Sample: a real publication, but the figure and summary have not been through our verification process. Some figures are approximate and some rest on assumptions, which we state on the record. Sample records show how evidence will be presented. They should not be cited.
Draft: someone has opened the original source and checked the record against it. A second read is outstanding, so treat it as not yet citable.
Reviewed: checked and signed off. The marker disappears and the record can be cited.

The figures the tools run on

The comparison tool, the dishes, the impact calculator and the country notes run on a second kind of record: the per-kilogram footprints, nutrient values, portion weights and typical diets that make the arithmetic work. We do not offer these for citation, so they do not carry a review status. Each one instead names the source it came from and says how it was arrived at.

Sourced: taken from the named publication, in that publication's own terms.
Derived: our arithmetic on a sourced figure, such as converting a carcass weight to retail weight. The conversion is stated on the figure.
Proxy: the nearest entry the dataset has, standing in for a food it does not cover. Lentils and chickpeas use its pulses figure, oat milk its soy-milk figure, salmon its farmed-fish figure. The food page says so.

What verification involves

Moving a record from sample to reviewed is a fixed sequence, the same for every record type. Nothing is promoted because it looks plausible.

Open the original paper, report or dataset, not a summary of it.
Confirm the figure, its unit, its geography and its year against what the source actually reports.
Record what the source found, its study type, population and sample size.
Record its limitations, its funding and any conflicts of interest.
Record which claims it supports and which it challenges. Inconvenient findings are recorded, not dropped.
Set the new status only after these checks are complete.

Review, corrections and funding

Each record carries its verification status and original-source attribution. Our editorial policy, correction policy and funding disclosure are being drafted and will be published here. Until then: we will correct errors promptly and note the correction on the page.