Methodology

This page sets out how our studies are produced, what our figures rest on, and where our limits are. It applies to every study on this site.

What a study covers

Each study follows one commodity through six stages: input supply, production, post-harvest handling, processing, marketing, and the distribution of margins between actors. It closes with a set of interventions, each costed. The structure is deliberately identical across studies so that findings can be compared between commodities — a loss rate in okra means the same thing as a loss rate in mango.

Where our figures come from

• Official statistics. The statistical yearbooks of the Ministry of Agriculture and Irrigation are our primary quantitative source. The most recent complete series available to us covers 2017–2021, including area, production and yield by governorate.
• Trade data. Customs records and international trade databases, cited to the specific commodity code where one exists.
• Sector literature. Published research, donor assessments and technical reports, cited individually.
• Field knowledge. Practice, prices, market customs and constraints documented through Impact Consulting’s work across Yemen, including named markets and districts.
• Broadcast and market reporting. We maintain a reference corpus of Yemeni agricultural broadcast material — market bulletins, extension programmes, farmer interviews and sector discussions. Every study is checked against it.

How we label a figure

This is the part we would most want a reader to hold us to. Numbers in our studies fall into three classes, and we distinguish them:

• Recorded. Taken directly from an official series or a cited source. Verifiable by going to the source.
• Derived. Calculated from recorded figures by a stated operation — a yield computed from area and production, a margin from two prices. The inputs and the operation are both given.
• Reconstructed. An estimate built where no official figure exists, because the commodity is not separately recorded in national statistics. Several commodities on this site fall into this category, including most leafy vegetables. Where a figure is reconstructed, the study says so and explains the basis.
• A reconstructed figure is not a measurement and we do not present it as one.

Checking against current conditions

Official series lag. A study built on a 2021 baseline can be accurate about structure and wrong about the present.

We therefore check every study against contemporary market and extension reporting, and record where conditions have changed materially — import restrictions, price collapses, disease outbreaks, new planting. Where the current situation contradicts the baseline, we say so in the study rather than quietly reconciling the two.

Conventions

• Currency. Figures in US dollars are converted at the Sana’a exchange rate prevailing at the time of writing, and the rate is stated in the study. They do not translate directly to Aden, Hadramout or Marib.
• Losses are reported by stage, not as a single national figure, because the intervention differs depending on where the loss occurs.
• Margins are given as each actor’s share of the final retail price, with the price at each node shown.
• Intervention costs are given per unit where possible, so they can be rescaled.

What we do not do

These are desk studies. They do not report primary field measurement, household survey data or laboratory analysis. Where a study proposes something that would require testing — a water sample, an oil-content assay, a loss trial — it is identified as a research need, not presented as a finding. Financial projections in our studies are scenarios, not forecasts. Where a projection assumes both a yield increase and a price increase, we flag the tension explicitly, because a large supply increase generally pushes prices down.

Corrections and versions

Every study carries a version number and date. Substantive corrections produce a new version with a dated note recording what changed. Earlier versions remain accessible, and the DOI resolves to the version history. We publish corrections rather than silently editing. A record that changes without acknowledgment is not a record.

Challenging our work

If you believe a figure is wrong, tell us — particularly if you work in the chain concerned. We would rather be corrected than cited wrongly.