How NigeriaPolls.ng Conducts Rigorous Public Opinion Research

Published: February 2026 | Methodology: n=2,047–5,000 per poll | Reading time: 8 minutes

Every number you see on NigeriaPolls.ng carries a story — not just about what Nigerians think, but about how we know what they think. In a media landscape flooded with unverified Twitter polls and anecdotal reporting, methodological transparency is not optional. It is the difference between noise and signal.

This post explains exactly how NigeriaPolls.ng designs, fields, and validates every poll we publish. If you are a journalist citing our data, a brand considering custom research, or a citizen evaluating claims about "what Nigerians want," this is your guide to separating rigorous research from guesswork.

1. The Problem with Most "Polls" in Nigeria

Before explaining our approach, let us diagnose what typically goes wrong:

The Twitter Poll Fallacy: A media outlet posts a poll to 50,000 followers. 5,000 vote. The result is treated as "public opinion." The problem? Twitter users in Nigeria skew urban, male, educated, and affluent. The sample is not representative of the 220 million Nigerians who matter.

The Online Survey Trap: A research firm sends a link via email and social media. Respondents self-select. The motivated and the angry over-participate. The busy and the disengaged are silent. The result reflects who has time and internet access — not the population.

The Invisible Weighting Error: A poll collects 1,000 responses but does not adjust for the fact that 60% came from Lagos and Abuja, while Kano, Borno, and Enugu are barely represented. The national headline is effectively a Lagos headline with a misleading label.

These are not hypothetical failures. They are the standard operating procedure for most polling in Nigeria. NigeriaPolls.ng exists because we believe Nigerians deserve better.

2. Our Sampling Philosophy: Who Gets to Speak?

The Sampling Frame

Our sampling frame begins with the most recent official data: the National Population Commission census projections, INEC voter registration records, and NBS household surveys. We do not use convenience samples. We do not use "whoever responds." We build a statistical mirror of Nigeria.

For a national poll, our target sample is proportionally distributed across:

• All 36 states + FCT (not just the big five)
• Urban and rural areas (not just cities)
• Gender (50/50 target, adjusted for state demographics)
• Age groups (18–24, 25–34, 35–44, 45–54, 55–64, 65+)
• Phone type (smartphone vs. feature phone — critical for mode effects)

Sample Size Determination

We do not use arbitrary round numbers. Our sample sizes are calculated to achieve a specific margin of error at 95% confidence:

For high-stakes polls (election tracking, major policy announcements), we increase to 3,000–5,000 to enable reliable subgroup analysis.

3. Multi-Modal Data Collection: Reaching Every Nigerian

Nigeria's digital divide is not a footnote. It is the central design challenge. A smartphone-only survey systematically excludes:

• Rural populations (feature phone prevalence: 60%+ in some states)
• Lower-income Nigerians (smartphone ownership correlates with income)
• Older adults (digital literacy gaps)
• Women in conservative households (phone access restrictions)

Our Three-Channel Approach

Channel 1: SMS (Primary)

• Reach: Universal — works on every phone in Nigeria
• Method: Shortcode or longcode messaging
• Limitation: 160-character constraint, no rich media
• Mitigation: Ultra-short questions, single-letter responses (A, B, C, D)

Channel 2: WhatsApp (Secondary)

• Reach: ~40 million active users in Nigeria
• Method: Conversational bot with rich formatting
• Advantage: Longer questions, images, voice notes for low-literacy users
• Limitation: Requires smartphone and data

Channel 3: IVR — Interactive Voice Response

• Reach: Any phone with voice capability
• Method: Automated call, press 1/2/3/4 to respond
• Advantage: Bypasses literacy barriers entirely
• Cost: Higher per-completion (₦15–25 vs. ₦2–3 for SMS)

Channel Assignment Logic

We do not ask panelists to choose their channel. We assign based on their profile:

This multi-modal approach is expensive. It is also the only way to claim national representativeness with integrity.

4. Panel Recruitment: Building the Foundation

How We Find Panelists

We do not buy phone number lists. We recruit through:

1. Partner organizations: Cooperatives, trade unions, religious organizations, community associations
2. SMS opt-in campaigns: Shortcode advertisements with clear value proposition ("Your opinion matters. Earn ₦100 airtime per survey.")
3. Referral incentives: Existing panelists earn ₦50 for each verified recruit
4. Geographic targeting: Deliberate over-recruitment in under-represented states, then weight down

Verification Protocol

Every panelist passes a three-step verification:

Step 1: Phone ownership verification

• OTP sent to registered number
• Must respond within 24 hours
• Duplicate phone numbers flagged and merged

Step 2: Demographic confirmation

• State, LGA, ward, age, gender, education
• Cross-checked against known distributions
• Outliers flagged for manual review

Step 3: Engagement test

• Sent a 2-question "practice survey"
• Must complete within 48 hours
• Speeders (< 30 seconds) and straight-liners rejected

Panel Maintenance

A panel is a living organism, not a static list. We continuously:

• Rotate inactive members: Panelists inactive > 90 days are deprioritized
• Recruit for under-represented segments: If Kano women 55+ drop below target, we launch targeted recruitment
• Purge fraudulent accounts: Duplicate phones, impossible geolocations, bot-like response patterns
• Update demographics: Annual re-verification of state, age, education changes

Current panel size: 50,000+ verified panelists (as of February 2026)

5. Weighting: Making the Sample Match Reality

Even with perfect recruitment, no sample perfectly mirrors the population. Weighting corrects this.

What We Weight By

How We Weight: Iterative Proportional Fitting (Raking)

We use raking — an iterative algorithm that adjusts the sample to match known population targets across multiple dimensions simultaneously.

Example: If our sample has 65% men but Nigeria is 51% male, each male response is weighted down (multiplied by ~0.78) and each female response weighted up (multiplied by ~1.47). The algorithm repeats this for all variables until convergence.

Weighting Limits

We cap extreme weights at 3.0 (no single respondent counts more than three times). If a subgroup is so under-represented that it requires weights > 3.0, we flag this in our methodology note rather than hide it.

6. Quality Control: Catching Bad Data Before It Pollutes Results

Every response passes automated validation before entering analysis:

Human Review Layer

Automated flags are reviewed by our Research Director before publication. In 2025, 8.3% of responses were excluded due to quality concerns — a figure we publish transparently.

7. Margin of Error: What ±2.2% Actually Means

Our standard national poll reports a margin of error of ±2.2% at 95% confidence. Here is what that means in plain English:

If we repeated this poll 100 times with different random samples, 95 of those polls would produce results within ±2.2 percentage points of the true population value.

When Margins Are Larger

We always report subgroup margins. Headlines that say "State X thinks Y" without noting the ±6% margin are misleading.

8. Transparency: Our Non-Negotiable Commitment

For every poll, we publish:

• ✅ Exact question wording (in English, Hausa, Yoruba, Igbo where applicable)
• ✅ Field dates (start and end)
• ✅ Sample size and response rate
• ✅ Margin of error (overall and by subgroup)
• ✅ Weighting targets and procedure
• ✅ Quality control exclusion rate
• ✅ Raw data download (anonymized, CSV)
• ✅ Pre-registration (for hypothesis-testing polls)

We do not:

• ❌ Alter methodology after seeing results
• ❌ Exclude inconvenient responses post-hoc
• ❌ Accept client direction on public poll design
• ❌ Publish without peer review by our statistician

9. Limitations We Acknowledge

No poll is perfect. Here are the honest limitations of our approach:

Non-response bias: The 40% of Nigerians who never respond to surveys may systematically differ from respondents. We mitigate this through multi-modal outreach and aggressive weighting, but we cannot eliminate it entirely.

Social desirability: Respondents may not admit unpopular views (e.g., corruption tolerance, ethnic bias). We use indirect questioning techniques for sensitive topics, but some bias persists.

Mode effects: SMS respondents give different answers than WhatsApp respondents. We test for mode effects in every poll and report when they are significant.

Temporal validity: A poll captures opinion at a moment. Events can shift opinions overnight. Our pulse polls (48-hour turnaround) address this for high-volatility topics.

10. How to Evaluate Any Poll in Nigeria

Use this checklist when you see any polling claim:

If a poll fails more than two red flags, treat it as entertainment, not evidence.

Conclusion

Rigorous polling is expensive, slow, and technically demanding. It is also the only way to know what 220 million Nigerians actually think — rather than what a vocal minority tweets.

At NigeriaPolls.ng, we have chosen rigor over speed, transparency over convenience, and representativeness over virality. Every number we publish has survived sampling design, multi-modal fieldwork, weighting, quality control, and human review.

The result is not just data. It is trust.

About this post: This methodology overview is updated quarterly. Last updated February 2026. For the full technical methodology guide, download the PDF. For questions, contact research@nigeriapolls.ng.

Related: Understanding Margin of Error | SMS vs. WhatsApp Surveys | Our Panel