How is the work done, and can you check it?
Our methods, the standards behind them, and the limits of what any method can tell you. Written for people who have to judge research before they commission it.
Methods as of October 2026. Next review January 2027.
How does a study run, from question to release?
At six points, someone other than the person who did the work signs off. If you are commissioning a study, you can ask to see the file at any gate.
Design
Before the inception report
Study lead and statistician
The decision the study serves, the criteria and benchmarks, and the sample or sampling logic.
Ethics and data
Before first contact
Ethics reviewer and data protection lead
The ethics decision, consent forms, data protection assessment and referral pathway.
Pilot
Before full fieldwork
Study lead and language reviewers
The pilot log, sign-off of each language version and device tests.
Field
Weekly, during fieldwork
Field lead and data supervisor
Daily checks, back-check results, contact attempts and any team paused.
Analysis
Before results are shared
Statistician and an independent reviewer
The signed analysis plan, code review, weights and disclosure check.
Release
Before the report leaves
Evidence lead and editor
The limits page, a source and date on every figure, and the report checked against report standards.
In a small team one person may hold several roles, and the study file names who signed each gate. What matters is that the person who signs is not the person who did the work.
Which method answers which question?
The services page says what you receive. This is the standard behind each kind of work.
Surveys and measurement
How many people, where, and what do they need?
MICS, NDHS and LSMS designs. HHFA and Service Delivery Indicators for facilities. EGRA and ASER-style learning assessments. AAPOR reporting.
Qualitative and participatory research
Why do people do what they do?
The approach is chosen by the question: contribution analysis, the Qualitative Impact Protocol, realist evaluation or Most Significant Change.
Evaluation
Did it work, for whom, and at what cost?
OECD DAC criteria with a stated benchmark for each judgement. UNICEF impact evaluation briefs. Registered analysis plans for causal designs.
Monitoring, tracking and verification
Did it arrive?
Unannounced visits, verification against source records, cash monitoring guidance, and third-party monitoring with its limits stated.
Economics and financing analysis
Where does the money go, and what would this cost?
System of Health Accounts, the iDSI Reference Case, PEFA and expenditure tracking surveys, with every assumption written out.
Policy, governance and systems analysis
Why is a sound policy not working?
Published assessment tools such as ISPA, run as written, with conflict sensitivity guides where the setting needs them.
Evidence synthesis and research translation
What is known, and where is it thin?
Reviews reported to PRISMA 2020. Briefs that set out the problem and the options and take no position on the choice.
Data systems and M&E design
Can we trust the data our plans depend on?
WHO’s Data Quality Review toolkit checks completeness and consistency, and verifies against source records.
How do we design a sample?
Most disputes about a survey number start with a sample that was never explained. Every sampling note of ours answers the same eight questions before fieldwork begins.
What every sampling note states
- 01The level you need to report: national, zone, state or local government area. The sample follows it.
- 02The frame, and how we checked it against the ground.
- 03The stages of selection.
- 04The confidence level and margin of error.
- 05The design effect we assumed, and the non-response we expect, each with its source, usually the last comparable national survey.
- 06The sample that results, by reporting level.
- 07How the weights are built, and that every analysis uses them.
- 08What happened to every sampled case, reported with AAPOR’s standard outcome rates.
Ten questions to ask any research company
You don’t have to take our word for any of this. These are the questions we would ask before commissioning research, and what a good answer contains. Use them on us, and on anyone else.
Who signs the ethics decision, and when?
A named committee or panel, with a date before first contact.
What level can you report, and how does the sample reach it?
The reporting level is fixed first, and the sample follows.
What design effect and non-response did you assume, and what happened?
Assumptions stated before fieldwork, and outcome rates reported to a standard definition.
How were the tools translated and tested?
A named method such as TRAPD, a pilot log and native-speaker sign-off.
Who does the fieldwork, and who checks it?
Named teams, a training record, back-checks and a rule for pausing a team.
What happens to my data, and who can see it?
Classification, access by role, a disclosure check and a breach plan.
Is the analysis plan fixed before the data are opened?
A dated plan, registered for causal studies, with changes logged.
How are results broken down by sex, age and disability?
Groups declared in advance, a stated cut-off, and only where the sample supports it.
What can’t this study tell me?
A limits page, in plain words.
What changed from the plan, and why?
A change note, not a silent edit.
What standards do we work to?
There is no single rulebook for social research. A donor or reviewer checks five things at once, and a study can pass one and fail another. This is what each layer asks, and the documents we work to.
Rights and ethics
May we do this, and how do we protect the people involved?
- UNICEF Ethics Procedure, 2021
- UNEG Ethical Guidelines
- CIOMS guidelines, 2016
- NHREC National Code
- Nigeria data protection directive, 2025
Checked by: Ethics boards, donors and the Data Protection Commission
Evaluation norms
Is the study built to judge merit fairly, and be useful?
- OECD DAC evaluation criteria, 2019
- UNEG Norms and Standards, 2016
- EU evaluation framework
- African Evaluation Principles, 2021
Checked by: Commissioners and quality reviewers
Measurement
Can the numbers be trusted and compared?
- Nigeria MICS 2021 design
- World Bank LSMS guidebook
- AAPOR Standard Definitions, 2023
- TRAPD questionnaire translation
Checked by: Statistical offices and technical reviewers
Sector instruments
Which validated tool fits this sector?
- WHO facility assessment (HHFA)
- World Bank Service Delivery Indicators
- Early Grade Reading Assessment
- IPC food security manual
- JIAF 2.0 needs analysis
- Washington Group questions
Checked by: Sector leads at UN agencies and donors
Transparency
Can someone check our working?
- 3ie transparency and ethics policy
- Registry of impact evaluations
- PRISMA 2020 for reviews
- OCHA data responsibility, 2025
Checked by: Funders, registries and peer reviewers
What can’t our methods tell you?
Every method has a blind spot. We state the ones that apply on every study. These are the six we watch for most, with the evidence for each.
Phone surveys
In lower-coverage countries, households with phones were wealthier, less rural and more educated. Weights help, and they widen the margin of error.
We use phone for follow-up from a face-to-face frame, reweight, and report the design effect.
IPC food insecurity
An analysis of more than 10,000 IPC assessments found they miss about one in five acutely hungry people.
We cite IPC figures with that limit beside them.
Expenditure tracking
A public expenditure tracking survey is a snapshot. It isn’t built to show change over time.
We don’t present a repeat as a trend unless the design supports it.
Third-party monitoring
Monitors are good at physical checks, and weak on intangible results such as peacebuilding.
We use them for verification, and triangulate with two other sources.
Stories of change
Most Significant Change works across settings, and it is not a 360-degree evaluation.
We say what a qualitative study can and can’t show.
Breakdowns by disability
In emergencies, not every dataset can support a meaningful breakdown by disability.
We state the cut-off, check the sample can support it, and say when it can’t.
Have a method question we haven’t answered?
Ask us. We answer in writing, and if others would find the answer useful, we add it here.