Polling Methodology Explained: How NigeriaPolls Conducts Surveys

Every NigeriaPolls survey begins with a clear research question. Are we measuring presidential approval? Tracking party preference? Understanding public attitudes toward fuel subsidy removal? The research question determines everything that follows: the population to sample, the questions to ask, and the analysis to perform.

Defining the Target Population

The target population is the group whose opinions we want to measure. For most NigeriaPolls surveys, the target population is all Nigerian adults aged 18 and above. But sometimes we narrow this: for a pre-election poll, we might target "likely voters" — adults who are registered to vote and say they plan to vote. For a youth-focused survey, we might target 18–35-year-olds only. Defining the target population precisely is critical because it determines who we try to interview and how we weight the results.

Questionnaire Design

Writing good survey questions is harder than it looks. A poorly worded question can shift results by 10 percentage points or more. Our questionnaire design process involves multiple stages:

1. Drafting: Research staff draft questions based on the research objectives and review of prior literature.
2. Neutrality review: Multiple analysts review each question for balance, clarity, and absence of leading language.
3. Translation: Questions are translated into Hausa, Igbo, and Yoruba by native speakers with political research experience. Back-translation checks ensure accuracy.
4. Cognitive pre-testing: We conduct 20–30 one-on-one interviews with Nigerian respondents to test whether questions are understood as intended.
5. Pilot survey: A small-scale pilot (n=200) is fielded to test the full instrument and estimate response distributions.
6. Revision: Based on pilot results, questions are refined, answer categories adjusted, and skip logic verified.

The Order Effect

The order in which questions are asked can influence responses. If we ask about economic satisfaction before asking about presidential approval, respondents who are unhappy with the economy may rate the president lower than they would have otherwise. To mitigate this, we randomize the order of some question blocks across respondents and always place sensitive demographic questions (income, religion, ethnicity) at the end of the questionnaire.

Balanced Alternatives

When asking about candidate preference, we present all major candidates in randomized order (with a random half seeing one order and the other half seeing the reverse). This prevents the first-listed candidate from receiving an artificial advantage — a well-documented phenomenon called the primacy effect. We also include an explicit "undecided" option rather than forcing respondents to choose, because many Nigerian voters genuinely have not made up their minds months before an election.

A poll is only as good as its sample. If the sample is biased, the results are wrong — no matter how sophisticated the analysis. NigeriaPolls uses stratified random sampling, the gold standard for national surveys in countries with diverse populations.

What Is Stratified Random Sampling?

Stratified sampling means dividing the population into subgroups (strata) and randomly sampling within each stratum. In Nigeria, our primary strata are the 36 states plus the Federal Capital Territory (FCT). Within each state, we further stratify by urban/rural residence. This ensures that every state, and both urban and rural areas within each state, are represented in the sample.

Why not simple random sampling? Because if you randomly dial 2,000 phone numbers in Nigeria, you will probably get far too many respondents from Lagos and Abuja and far too few from Borno or Taraba. Stratification prevents this geographic skew.

The Sampling Frame

Our sampling frame is built from a combination of the most recent Nigerian census data, INEC voter registration figures, and satellite-derived enumeration area maps. An enumeration area (EA) is a geographic unit of approximately 50–200 households, roughly equivalent to a city block or a village section. We use EAs as our primary sampling units because they provide a complete, non-overlapping coverage of the country.

The process works as follows:

1. We obtain a master list of all EAs in Nigeria, classified by state and urban/rural status.
2. We calculate how many EAs to select in each state, proportional to the state's adult population.
3. Within each state, we randomly select the required number of EAs, ensuring both urban and rural areas are represented.
4. Within each selected EA, field teams create a complete list of households and randomly select a fixed number (usually 10–15).
5. Within each selected household, we use the Kish grid method to randomly select one adult respondent. This avoids the bias of interviewing whoever happens to be home.

Sample Size Allocation

For a national poll with 2,400 respondents, the sample is allocated across states proportional to population, with a minimum of 30 respondents per state to ensure reliable state-level estimates. Lagos and Kano, as the most populous states, receive the largest allocations (around 200–250 each). The FCT receives approximately 80 respondents. This allocation scheme ensures that our national estimates are representative while still allowing meaningful analysis at the state and zonal levels.

Data collection is where methodology meets reality. Nigeria's size, diversity, and infrastructure challenges make fieldwork one of the most demanding aspects of polling. NigeriaPolls operates a hybrid model combining face-to-face interviews with computer-assisted telephone interviewing (CATI) to balance representativeness and efficiency.

Face-to-Face (F2F) Interviews

Face-to-face interviews remain our primary mode because they reach everyone — including the millions of Nigerians who do not own phones or have reliable mobile coverage. Our F2F interviews are conducted using CAPI (Computer-Assisted Personal Interviewing)tablets, which means responses are entered directly into digital forms rather than recorded on paper. This eliminates transcription errors, allows real-time data validation, and enables GPS verification of interview locations.

Each field team consists of one supervisor and 4–6 trained enumerators. Teams are deployed to their assigned EAs with tablets, printed maps, and household listings. Enumerators conduct interviews in the respondent's preferred language (English, Hausa, Igbo, Yoruba, or Pidgin) and are required to obtain verbal informed consent before beginning.

Enumerator Training

Enumerators undergo a rigorous 3-day training before each major survey. Training covers:

• Sampling protocols and household selection procedures
• Question-by-question review of the questionnaire
• Neutral interviewing techniques (how to read questions without influencing responses)
• Handling refusals and sensitive topics with cultural sensitivity
• CAPI tablet operation and troubleshooting
• Security protocols for high-risk areas

Enumerators are evaluated through supervised practice interviews before being certified for fieldwork. Those who fail the certification are reassigned or retrained.

Mobile (CATI) Interviews

For some surveys, we supplement F2F interviews with CATI (Computer-Assisted Telephone Interviewing). CATI is faster and cheaper, but it introduces coverage bias because not all Nigerians have phones, and those who do are not representative of the full population. We address this by:

• Using CATI only for urban areas with high phone penetration
• Relying on F2F for rural areas and conflict-affected zones
• Weighting CATI samples to correct for known demographic skews (overrepresentation of educated, urban respondents)

For election polls, we typically use 100% F2F to maximize representativeness. For rapid-response or flash polls, we may use a higher CATI share to deliver results quickly.

Security and Access Challenges

Some parts of Nigeria are difficult or dangerous to access due to insurgency, banditry, or communal conflict. In these areas, we work with local partners and security advisors to assess risk and determine whether fieldwork can proceed safely. We never send enumerators into active conflict zones. When certain areas are inaccessible, we document this limitation transparently in our methodology notes and adjust our weighting to account for the missing geographic coverage.

Raw survey data is never perfect. Some respondents speed through the questionnaire. Some give inconsistent answers. Some interviews happen outside the assigned area. Data cleaning is the process of identifying and addressing these issues before analysis begins.

Data Validation

Our CAPI software enforces real-time validation rules during interviews: range checks (age must be 18–120), skip logic (if you say you are not registered to vote, you should not be asked who you will vote for), and consistency checks (a respondent who says they are unemployed should not report a monthly salary of ₦500,000). Despite these checks, some issues are only caught during post-field review.

Our cleaning protocol includes:

• Speed checks: Interviews completed in less than one-third of the median time are flagged for review.
• Straight-lining: Respondents who give the same answer to every question in a battery are flagged.
• GPS verification: Interview coordinates are checked against assigned EAs. Outliers are investigated.
• Duplicate checks: We check for duplicate phone numbers, device IDs, or respondent names.
• Supervisor review: Field supervisors review a random 10% of interviews from each enumerator.

Flagged interviews are reviewed individually. Those with confirmed quality problems are removed from the dataset. Typically, 2–5% of interviews are excluded at this stage.

Weighting

Even with perfect sampling, the people who actually complete interviews may differ systematically from the population. For example, women and older adults are often more likely to be at home and available for interview, leading to overrepresentation in the raw data. Weighting is the statistical correction that adjusts the sample to match known population benchmarks.

NigeriaPolls weights our samples to align with the most recent Nigerian Bureau of Statistics (NBS) data and INEC voter registration figures. Our weighting variables include:

• State: Each state is weighted to its share of the national adult population.
• Gender: The sample is adjusted to 50% male / 50% female (or NBS-estimated sex ratios where they differ).
• Age: Respondents are weighted into age brackets (18–24, 25–34, 35–44, 45–54, 55+) matching national distributions.
• Urban/rural: The sample is weighted to NBS urban/rural population shares per state.

We use rim weighting (also called raking), an iterative procedure that adjusts the sample to match all target distributions simultaneously. The weights are trimmed at the 1st and 99th percentiles to prevent extreme weights from distorting results. The effective sample size after weighting is always reported alongside the nominal sample size.

Once the data is cleaned and weighted, we move to analysis. Our analytical approach is designed to be both rigorous and accessible: we use statistical methods that meet academic standards, but we present findings in plain language that any Nigerian can understand.

Descriptive Analysis

The foundation of every report is descriptive statistics: percentages, means, and distributions. For categorical questions (like candidate preference), we report the percentage of respondents in each category. For scale questions (like approval ratings on a 1–5 scale), we report the mean, median, and distribution. All percentages are calculated using the weighted data, and confidence intervals are computed for key estimates.

Cross-Tabulations

Cross-tabs (short for cross-tabulations) show how different groups answered the same question. We produce cross-tabs for gender, age, region, education, religion, urban/rural residence, and — for election polls — past vote and party identification. Cross-tabs are where the real insights live. They reveal whether a candidate's support is broad or narrow, whether an issue divides urban and rural Nigerians, or whether young people think differently from their parents.

We apply statistical significance testing to cross-tab differences. A difference is flagged as significant only if a chi-square test or z-test shows it is unlikely to have occurred by chance (p < 0.05). This prevents us from over-interpreting small, noisy differences.

Reporting Standards

Every NigeriaPolls report includes the following elements:

• A plain-English executive summary of the key findings
• The full questionnaire (so readers can see exactly what was asked)
• Methodology notes (sample size, dates, method, margin of error, response rate)
• Topline results tables
• Selected cross-tabs (gender, age, region, education)
• Limitations and caveats

We do not cherry-pick favorable findings or bury uncomfortable ones. If our data shows something unexpected — a candidate surging in a region where they were previously weak, or an issue uniting voters across party lines — we report it. Our job is to describe public opinion, not to shape it.

Quality control is embedded at every stage of our process, not added as an afterthought. Here are the specific measures we use to ensure data integrity:

1. Supervisor Oversight

Each field team has a trained supervisor who accompanies enumerators, observes interviews, and reviews completed questionnaires daily. Supervisors are required to conduct at least one observed interview with each enumerator per survey wave and provide feedback on technique.

2. Back-Checking

After fieldwork concludes, our quality control team randomly selects 10–15% of respondents and calls them back to verify that the interview took place, that the respondent matches the recorded demographics, and that key answers are consistent with the original interview. Discrepancies are investigated and, if fraud is confirmed, the enumerator is dismissed and all their interviews are removed from the dataset.

3. GPS and Time-Stamp Verification

Every CAPI interview records the GPS coordinates of the interview location and the start/end timestamps. We map all interviews to verify they occurred within the assigned EA. We also check that interview durations are reasonable — interviews that are impossibly short (suggesting the enumerator filled in answers without asking) or suspiciously long (suggesting the enumerator was interrupted or distracted) are flagged for review.

4. Audio Recording (Where Permitted)

In some surveys, we ask respondents for permission to audio-record the interview for quality assurance. These recordings are reviewed by senior researchers to check that enumerators are reading questions exactly as written, not paraphrasing or adding their own commentary. Audio review is especially valuable for detecting subtle interviewer bias.

5. External Validation

We compare our results against other credible polls, election outcomes, and administrative data where available. For example, after the 2023 elections, we conducted a retrospective accuracy analysisto see how our pre-election polls compared to actual results. This external validation helps us identify and correct systematic biases in our methods.

No poll is perfect. We believe that openly discussing our limitations builds more trust than pretending they do not exist. Here are the major sources of error and uncertainty in Nigerian polling:

1. Sampling Error

Sampling error is the unavoidable uncertainty that comes from interviewing a subset of the population rather than everyone. It is what the margin of error quantifies. Even with a perfectly random sample, there is a 5% chance that the true population value lies outside the reported confidence interval. This is why we never claim a candidate "will win" based on a poll; we say they "lead by X points, with a margin of error of Y."

2. Non-Response Bias

Not everyone agrees to be interviewed. If the people who refuse differ systematically from those who participate, the poll will be biased. For example, wealthy or politically connected Nigerians may be less willing to share their opinions with strangers. We mitigate non-response bias through careful weighting and by documenting response rates, but we cannot fully eliminate it.

3. Social Desirability Bias

Some respondents give answers they believe are socially acceptable rather than their true opinions. In Nigeria, this is particularly relevant for sensitive topics like corruption, ethnicity, and candidate preference. A respondent might say they support a "unity" candidate in public but vote along ethnic lines in the privacy of the ballot box. We use neutral question wording, private interviewing environments, and anonymous response methods to reduce this bias, but it cannot be fully eliminated.

4. Coverage Error

Our sampling frame does not perfectly cover the entire population. Some remote areas lack updated EA maps. Some households are nomadic or displaced. Phone-based polls miss people without phones. We do our best to address coverage gaps, but we acknowledge that some Nigerians — particularly the most marginalized — are underrepresented in our data.

5. Timing

A poll is a snapshot of opinion at a specific moment. Events that occur after fieldwork — a scandal, a policy announcement, a security incident — can shift opinion rapidly. A poll fielded in March may be irrelevant by May. This is why we date-stamp every result and caution against treating old polls as current reality.

6. The "Shy Voter" Phenomenon

In some elections, supporters of a controversial candidate are reluctant to admit their preference to pollsters. This "shy voter" or spiral of silence effect has been documented in elections worldwide, including in Nigeria. If it exists, polls will understate support for the controversial candidate. We test for this by comparing phone and face-to-face results (shy voters may be more honest in anonymous phone interviews) and by asking indirect questions about "who people you know" support.