How we collect, check and label data

Sampling frames and sample design

A sample is only as good as the list it is drawn from. We choose the frame to fit the question and say in every release which frame we used.

• Phone frame. Our Nigerian phone sampling frame holds 11,000+ numbers. It supports fast national phone fieldwork in our home market, and we draw from it with quotas and weights so that results are not driven by who is easiest to reach.
• Household samples. For baselines and evaluations we use a clustered design: areas are selected first, then households, then one adult in each household. Clusters are chosen from the latest census geography.
• Programme populations. For beneficiary studies the frame is the programme's own register, checked for duplicates and gaps before we sample from it.
• Web panels and online groups. Used as a supplement or for specific online populations. Never used alone for a national claim.

Sample size is set from the precision the buyer needs for the smallest group they want to report, not only for the national total. We state the sample size by group in the release.

Questionnaire design and translation

We write questions in plain language and test them for one meaning only. Standard items are reused where a comparable question already exists, so that results line up with earlier rounds. The named country expert reviews the instrument before it is finalised.

The phone script and the web form are built from one source, so wording matches across modes. Translation is done by a native speaker and checked by a second person who translates it back to English. Differences are resolved before the pilot. Exact wording is published with every release, in English and in each fielded language.

Piloting

Every study runs a short pilot before main fieldwork. We look at timing, routing, refusals, items people misunderstand and interviewer notes. Wording and routing are fixed before launch, and pilot interviews are not mixed into the final dataset unless the questionnaire is unchanged.

Interviewer training and supervision

Interviewers are trained on the questionnaire, consent, neutral probing and the rules for recording refusals. They practise on mock interviews before they call or visit anyone. Supervisors listen to a sample of calls and review field notes. Interviewers who fail a check are retrained or removed. Field partners sign our data protection and anti-fraud annex before award.

Live quality checks and back-checks

The table above lists the checks that run on every wave. Live checks run while interviews come in, so a problem is caught within the day. Back-checks come afterwards: a share of completed interviews is re-contacted and a few key answers are asked again. Failed cases are dropped and count against the field partner.

Weighting and margin of error

We weight to the latest census by age, sex, region and urban or rural, so that the sample matches the adult population. Weighting corrects who answered. It cannot correct who was never reachable, which is why we state the mode and its limits.

The margin of error is calculated on the effective sample size after weighting, not the raw count. Weights reduce the effective sample, so the margin is wider than a headline interview count suggests. Both weighted and unweighted counts are reported. Margins are for the whole sample; groups within it carry wider margins.

Disclosure standard

Every release states the sponsor, fieldwork dates, sample, population, mode, sampling frame, weighting, margin of error and the exact question wording. If a release cannot state any of these, it does not publish.

Data protection and ethics

Respondents are adults aged 18 and over. Studies with minors go ahead only when the client funds a full consent process. Consent is recorded with time, channel, privacy notice version and country. We do not collect ethnicity, religion or party affiliation unless an approved research need requires it and local law allows it. We work under Nigeria's Data Protection Act 2023, Kenya's Data Protection Act 2019, Ghana's Data Protection Act 2012 (Act 843) and Uganda's Data Protection and Privacy Act 2019. We take no work from political parties or candidates, and we follow each country's rules on publishing polls before an election.

Corrections

If we get something wrong we correct the page and add a dated note that says what changed. Report an error to corrections@pollsafrica.com with the page link and a source. The full policy is in standards and corrections.

Reader polls are not surveys

A reader poll on this site is open to whoever visits and chooses to vote. Respondents select themselves, the sample is not weighted to the population and the result says nothing reliable about a country. Reader polls are labelled as such and never carry a margin of error. A scientific survey uses a designed sample, quality checks and weighting, and carries the methods box above.

Related

See how these methods apply on development partner studies, in our research services and in insights. Country context is on the country pages, including Nigeria.

Country Pulse

Country Pulse shows what a country's news outlets are reporting today. It is a reading list, not a newsroom. We do not write, edit or summarise any story.

How sources are chosen

We list national outlets that publish a public feed, the newsrooms of official institutions such as the presidency, parliament, the central bank and the electoral commission, and independent fact-checking organisations. An outlet is in the update only while its feed can be read. Owners are shown on the sources page so readers can weigh each outlet for themselves.

What we collect

Every half hour we read each public feed and keep the outlet name, headline, link and time. We keep the feed's own short description only to help match headlines, and we never store or publish article text. We follow each site's robots.txt.

How topics are formed

Headlines from the past day are compared by their shared words and names. Headlines about the same event are grouped into one topic. A topic is listed only when more than one outlet covers it, and the topics are ranked by the number of different outlets. The headline shown for a topic is a real headline from one of those outlets, the one closest to the rest of the group.

The ranking measures how widely a story is reported. It says nothing about whether a story is important or accurate. Official releases and fact-checks are listed apart from the ranking and are not counted in it.

Links to our data

A topic gets a related-data link when its headlines use words that match one of our data pages, for example budgets, elections or the economy. The match is by keyword and is not an editorial judgement.

To report a wrong link or a missing outlet, write to corrections@pollsafrica.com.

Reader reactions

Under each story, readers can say how it makes them feel and whether it affects them. We show the share of readers who chose each answer, and how many readers have answered.

Not a scientific poll

Anyone who visits the page can answer, so the readers choose themselves. They are more likely to be online, urban, younger and interested in the news than the country as a whole, and a story shared in a group chat can draw a crowd of like-minded people. The figures describe the readers who reacted. They are not an estimate of what the whole country thinks, and they carry no margin of error. Our surveys, which use designed samples and are weighted, are described above and are the place to look for national results.

One reaction per device

You do not need an account. The first time you react, your browser stores a random code that we sign, and each device can react once to each story. Clearing your browser data, or using another device, can allow a second reaction, so the counts are not fully protected from repeat voting. To limit abuse, we also cap how often one connection can submit reactions. We keep a scrambled version of the connection address for that cap only. It cannot be turned back into an address, it is made with a key that changes every day, and it is not stored with your reaction.

The mood gauge

The mood gauge adds up the feelings readers chose over the last day. Worried and angry count to one side, hopeful and happy to the other, and "don't care" is left out of the needle. It is shown only once enough readers have reacted.

Which stories can be reacted to

We pick stories on public-interest topics, by keyword: politics and government, the economy, security, public services, education, health, infrastructure, energy and sport. Foreign news with no link to the country is left out. Entertainment gossip and other lighter stories are listed apart, with reactions closed. The rules are plain keyword lists and no one writes or rewrites a story. Nigeria's stories and reactions are on NigeriaPolls, which uses the same service and the same rules.