Services
Surveys and measurement
We count and measure. Surveys of households, facilities, schools and providers tell you how many people are affected, where they are, and how far a service reaches them.
Questions this answers
- How many people are affected, and how does it differ by state, sex, age and disability?
- What share of facilities, schools or households meet the standard?
- What do people say about the service they use?
The kinds of survey we run
Most studies combine two or three of these.
Household and individual surveys
Split by sex, age and disability. The standard way to measure how many people are affected, and how that differs by group and place.
Facility, school and provider surveys
Visits to health facilities, schools or other service points, to check what they can actually deliver, not just what is on the register.
Phone and other remote surveys
Used where face-to-face access is unsafe, too slow, or unnecessary for the question. We say plainly what a phone sample can and cannot represent.
Community perception surveys
What people say about a service they use, gathered systematically rather than anecdotally.
Specialised modules
Learning and skills assessments, time-use and unpaid care, financial and digital inclusion, and food security and welfare, added to a household survey where the question needs them.
Secondary analysis of national data
Where a national survey already asked your question, we analyse the existing data before proposing a new one.
How we sample, and how we weight
The method changes the confidence you can place in the number. We state ours in every proposal, before you commit.
A probability sample, not a convenience one.
Every unit in the population has a known, non-zero chance of selection. This is what lets a sample of a few thousand people stand in for a state or a country.
The mode fits the coverage.
Phone and web surveys only represent the people who have a phone or a connection. Where that share is too low or too uneven for the question, we use face-to-face interviews instead, and we say which we used and why.
Weighted for who was reached, and who was not.
Every sample slightly over- or under-represents some groups. We weight the results to correct for the chance of selection and for who did not respond, and we weight to match the known population where a reliable count exists.
Translated and tested before fieldwork.
Every questionnaire is translated using the TRAPD method: two independent translations, a review that reconciles them, a final decision by an adjudicator, a pretest with real respondents, and a written record of each choice. Then the whole survey is piloted before fieldwork runs.
What the numbers on a report mean
Three terms appear on almost every estimate we give you. Here is what each one means, with an illustrative example.
Margin of error
How far the true figure could plausibly be from the number we report, most of the time. A finding of “42%, margin of error ±3 points” means the true figure is probably between 39% and 45%.
Confidence level
How often that range would contain the true figure, if the survey were repeated. We report at the standard 95% level unless we say otherwise: run the same survey 100 times, and about 95 of those ranges would contain the true value.
Design effect
Some sample designs (for example, interviewing several households in the same cluster) are less efficient than a pure random sample of the same size. The design effect adjusts the margin of error to account for this, so the figure you see is not falsely precise.
From our own work
Real work of ours that used this kind of research, not an invented example.
A cross-border energy infrastructure operator
Real exampleMeasuring community impact along a cross-border energy corridor
308 households in each of two rounds, 2019 and 2022. 33 communities in Lagos and Ogun States, along a 56km right-of-way.
Adebayo Adeleke LLC
Real exampleMapping the real cost of moving food across Nigeria
195 respondents (98 farmers, 95 suppliers/retailers). Seven states: Benue, Kano, Borno, Oyo, Edo, Lagos and Abuja (FCT).
Reading an estimate
Illustrative42% of households in the example zone had access to an improved water source (n=1,204; margin of error ±3.1 points at the 95% confidence level, design effect included).
A real report from us states the sample size, the margin of error, the confidence level and the design effect for every headline estimate, not only in an appendix.
Questions worth asking any survey provider
Buyers of survey samples have a standard set of questions they can ask a provider, published by ESOMAR, the global research association. We have adapted five of them for a fieldwork study rather than an online panel, and we answer all five in every proposal.
How was the sample selected, and from what frame?
We name the sampling frame (for example, the latest census enumeration areas) and the selection method, in the proposal, not only in the final report.
What is the expected response rate, and what happens if it is low?
We state a target response rate before fieldwork, and we report the actual rate achieved, including how non-response was handled.
Who checks the data while fieldwork is still running?
A supervisor back-checks a sample of interviews during fieldwork, not only after it ends, so problems are caught while they can still be fixed.
How is the questionnaire translated, and by whom?
By the TRAPD method, in every language the study needs: two independent translators, a reviewer, an adjudicator who settles the final wording, a pretest with real respondents, and a record of every choice.
What happens to the data and the questionnaire after the study?
Set out in writing before fieldwork begins: who owns it, what is anonymised, and what you receive.
What you receive
Estimates by group and area with the margin of error stated, the questionnaire, and an anonymised dataset where consent and contracts allow.
The estimates
Split by state, sex, age and disability wherever the sample allows, with the margin of error on every headline figure.
The questionnaire
Exactly as fielded, in every language it was fielded in.
The dataset
Anonymised, and documented, where consent and your contract allow. We say in advance if it will not.
Questions people ask about surveys specifically
How big does the sample need to be?
It depends on how precise the answer needs to be, and how many groups you need separate results for. A bigger sample narrows the margin of error, but the return shrinks as the sample grows, so we size it to the decision, not to a round number.
Why not just call people?
Only where phone coverage is high enough and even enough across the groups you need results for. Where it is not, a phone survey would tell you about people with phones, not about the population as a whole, and we would say so rather than let the number pass as if it were.
Which languages can the questionnaire be fielded in?
Hausa, Yorùbá, Igbo and Nigerian Pidgin, alongside English, and other languages where a study needs them. Every language version is translated by the TRAPD method, pretested with respondents and signed off by a native speaker.
How long does a survey like this take?
For a typical household survey: four to eight weeks to design, translate, programme and pretest the questionnaire; about two weeks to train and pilot the field team; two to six weeks of fieldwork, depending on the sample and how widely it is spread; and four to six weeks to clean, weight, analyse and report. Most studies take three to five months end to end. Repeat rounds with an existing questionnaire are faster, and the proposal gives you the timeline for your study, stage by stage.
Have a number you need to be sure of?
Tell us the decision it would inform. We will say whether a survey is the right tool, and what sample and method it would need.