Agroecology Evidence Hub

Agroecology Evidence Hub

How well do agroecological practices perform, and where?

We merge the data behind published field studies and show what each practice actually changes.

Explore data How the data is built
See every comparison on a map, filter by region, crop, outcome and practice, and download what you pick. Explore the data

Each record is a single control against treatment comparison for one outcome: a place, a crop, a practice, and the measured difference. Filter the records, read them on the map, and download the selection.

The Hub is global in scope. Which countries appear today depends on which syntheses have been harmonised so far, not on any geographic limit.

Homepage text
To be filled Framing paragraph, headline findings and featured figures are still to be decided.
Draft 0.4Data CC BY 4.0Not yet citable
Effect size (log response ratio)
Point size is the treatment sample size

Comparisons

Click a row to find it on the map, a heading to sort
About

A living evidence hub for agroecological meta-analyses.

The Hub answers one question, and keeps answering it as new evidence arrives: how effective are agroecological farming practices at improving agronomic, economic, environmental and social outcomes, compared with non-agroecological practices or with natural and semi-natural habitats?

It is developed by the Alliance of Bioversity International and CIAT within the CGIAR Multifunctional Landscapes Science Program, and is built as evidence infrastructure rather than a single study. Scope is global, with no restriction on year, language or region.

Why it exists

The evidence base has grown quickly, but it is fragmented. Many systematic syntheses have already extracted primary study data, and those datasets then sit unused in supplementary files. The Hub brings them together so that results can be compared across sources.

Evidence already extracted once should not have to be extracted again.

Who it is for

Researchers who need comparable results across sites and sources, and funders and policy teams who need to see where evidence is strong, thin or missing. Current downstream users include work on nutrition and environmental targeting in nine African countries, food system modelling, and European policy analysis.

The four outcome domains

  • Agronomic. Yield, biomass, land equivalent ratios, animal productivity.
  • Economic. Income, profit, gross margin, cost and benefit measures.
  • Environmental. Biodiversity, input use efficiency, greenhouse gases, soil carbon.
  • Social. Food security and nutrition, labour, equity and inclusion, resilience.

Social outcomes and under represented regions are named priorities, because that is where existing syntheses are thinnest.

What it is not

The Hub is not a recommendation engine. Effects depend on soil, rainfall, season and management, so a positive median does not mean a practice will work on a given farm. Agroecology is also a gradient rather than a binary, and source datasets rarely label their arms that way, so deciding which arm counts as agroecological is sometimes an analytical judgement rather than a fact inherited from the source.

Developed by
With partners
Reporting standard: ROSESFAIR metadataCC BY 4.0
Methodology

Evidence recycling, then one shared template.

The Hub does not read the primary literature afresh. It harvests the primary study datasets that earlier meta-analyses already extracted, harmonises them into one shared template, and analyses them again, filling gaps by hand only where a source dataset was incomplete.

Step oneTwo registries

Syntheses are screened and routed. The second order registry records what each synthesis concluded. The first order registry holds the individual primary studies pulled out of those syntheses, which is where the measurements live. A synthesis can feed one registry or both.

Step twoHarmonisation

Each source dataset is mapped into the shared template. Studies already fully extracted by the source are harmonised by script. Partially covered studies are topped up by hand from the original paper. Practice names, outcomes, units, crops and animals come from a controlled vocabulary rather than free text.

Step threeOne row, one comparison

A row in the template is one control against treatment comparison for one outcome. Most fields come in control and treatment pairs, with study, site and outcome attributes shared. That is the unit a meta-analytic effect size needs.

Step fourEffect sizes

The analysis uses log response ratios and standardised mean differences, assigned per outcome by the vocabulary. This site shows the log response ratio, lnRR, and its standard error for every comparison, computed from the reported means, dispersion and sample sizes, which travel with each record so you can recompute anything yourself. The download also carries the plain percentage change, as a reading aid.

The log response ratio and its standard error are the numbers to use. The percentage in the download is for reading only.

Standards

  • Reporting follows ROSES, the standard for systematic evidence syntheses in environmental management, not PRISMA.
  • Scoping uses the population, intervention, comparator, outcomes and context framework.
  • Metadata follows FAIR principles, aligned to the ROSES checklist with DataCite and Dublin Core.
  • The review is living: periodic re-search, re-screening and republication, each release tagged with a changelog and a persistent identifier.

Known limits

  • Coverage reflects which syntheses have been harmonised, not the whole literature.
  • Positive results are more likely to be published than null results.
  • Sites cluster around long running research stations.
  • Dispersion is not reported by every source, so standard errors are available for a subset of records.
  • Which arm counts as agroecological is decided by a written rule, not inherited from the source. Comparisons where neither arm is agroecological are left out. Read the comparator field before pooling.
Effect sizes: lnRR and SMDLiving reviewROSES
Get involved

Have data that belongs in the Hub?

The Hub grows by contribution, and the most useful contribution is a dataset that already exists. If you have pulled comparisons out of field trials, or published a synthesis with its extraction table, tell us about it below. Harmonising data that already exists adds far more evidence per hour than extracting papers one by one.

What we can use

Any table with one row per control-versus-treatment comparison, at any stage of tidiness.

Before you send

Check you are allowed to share it. We cannot include data you do not have the right to pass on.

What happens next

We read it, tell you whether it fits, and credit you in the release that uses it.

Tell us about you

Tell us about the dataset

Where does it come from?

A website cannot send a file by itself. Choosing it here names it in the message, and you attach it to the email that opens. A link to the data is easier for everyone if you have one.

Licence
May we republish it as part of the Hub?

What does it cover?

These lists are the Hub's own vocabulary. Tick whatever applies, and use the box underneath for anything that is missing.

Practices
Outcomes
Crops
Regions

How was it made?

A treatment arm is not automatically agroecological, so the comparator matters as much as the practice.

Does the dataset include

Anything else?

Contact

Tell us what you are working on

One address reaches the data team. Say which crop, country, practice or outcome you are working on and we will point you at the relevant records, or tell you honestly that there are none yet.

Data team

m.lolita@cgiar.org

Host

Alliance of Bioversity International & CIAT

Response time

To be confirmed