// Methodology
The method behind the number.
A figure is only worth publishing if it can be defended. This page sets out how we field, how we check, and where the limits of each approach lie. It is written so that a client, an editor or an opponent can all read the same account.
Two standards, not one
We do two different kinds of work and they are held to different standards. Conflating them would flatter one and misrepresent the other.
Public opinion polls
Open to anyone with the link. Self-selecting, non-probability samples, de-duplicated server-side and time-gated to a fixed window. They describe the people who took part. They are published without a margin of error, because the design does not support one, and they are not forecasts.
Commissioned research
Quota-controlled samples designed against a defined population, fielded in whichever mode the question requires, with independent back-checking. Where the design supports it, results are weighted and reported with a margin of error.
How we field
The mode is chosen by what the question needs, not by what is cheapest. A trade-off exercise on finishes cannot be run down a telephone line, and a buyer with the income to transact is not sitting on an online panel in useful numbers.
Face-to-face interviewing
CAPIAn interviewer works through the questionnaire in person on a tablet. Skip logic runs itself, answers are validated as they are entered, and every interview is GPS-stamped and timed.
- Best for
- Long interviews, anything the respondent has to see, and people no panel or phone list reaches.
- Where it falls short
- The most expensive mode, and the slowest to field.
Telephone interviewing
CATIA trained interviewer works from a screen in a supervised call centre, calling from a sample frame or by random digit dialling.
- Best for
- Geographic spread at speed, short verbal questionnaires, and repeat tracking waves.
- Where it falls short
- Nothing visual. Beyond about eighteen minutes, refusals rise sharply.
Online and mobile self-completion
CAWIThe respondent completes the questionnaire themselves on a phone or laptop, reached by panel, link or open publication.
- Best for
- Volume, speed, image-based questions and low cost per response.
- Where it falls short
- Reaches the connected and the willing. Without a controlled panel and quotas it is a self-selecting sample and cannot represent a population.
Intercept interviewing
Interviewers approach people at a specific location: a mall, a showroom, a transport node, a comparable development.
- Best for
- Reaching people who are actually in-market at the moment they are deciding.
- Where it falls short
- Represents who was at that place at that time, and nobody else.
Depth interviews
IDIA 45 to 60 minute structured conversation with someone who holds specialist knowledge: an agent, a letting manager, a prescriber, an offtaker.
- Best for
- Explaining why the numbers look the way they do, and reaching trade knowledge no survey holds.
- Where it falls short
- Qualitative. Small numbers, never reported as statistics.
Observation and audit
Counting rather than asking: footfall counts, retail and distribution audits, facility surveys, shelf and availability checks.
- Best for
- Measuring what happens rather than what people report happening.
- Where it falls short
- Tells you what, not why. Pairs with interviewing rather than replacing it.
Registry and desk research
Official and commercial records: construction registrations, county approvals, utility connections, transaction and listing data, published indices.
- Best for
- Supply pipelines, market sizing and anything that already exists in a register. No survey can count competing supply.
- Where it falls short
- Only as current as the register. Vintages differ and are stated in the report.
Trade-off analysis
ConjointRespondents choose between whole packages at a stated price rather than rating features on a list, which is how a real purchase decision is made.
- Best for
- Establishing which features carry a price and which do not.
- Where it falls short
- Stated willingness to pay reliably overstates real willingness. We report it as a ranking with a conservative premium, not as a forecast of revenue.
Most studies use more than one mode. The same question asked in different modes produces different answers, so we do not silently pool them: either modes are kept separate and compared, or the study is designed for mixing and weighted accordingly.
How we check it
Quotas set before fieldwork
Sample structure is agreed with the client and fixed before a single interview. Achieved sample against target is published in the report, including where a cell fell short.
Independent back-checking
Ten per cent of interviews are re-contacted by a supervisor who did not conduct them. Interviews that fail a back-check are voided and replaced.
GPS and duration stamping
Every face-to-face interview records where and when it happened and how long it took. An interview that took four minutes when the median is thirty is investigated, not counted.
Server-side de-duplication
For online and open studies, each response is de-duplicated by a durable browser key and an independent device signature, so neither clearing cookies nor switching network defeats it.
Minimum reporting base
Any subgroup below thirty responses is reported as indicative and never as a percentage. Every figure carries the base it was calculated on.
Weighting, where the design supports it
Where a study is designed to represent a population and the sample frame allows it, results are weighted to known demographics and reported with a margin of error. Where the design does not support that, we say so rather than implying a precision we do not have.
Where we field
54 African markets, held to ward level where the administrative geography allows it. Kenya is mapped to all 47 counties, 290 constituencies and 1,450 wards, from the IEBC register. Elsewhere we use the OCHA Common Operational Datasets maintained with national statistical offices. A catchment is defined by named administrative units, not by a radius drawn on a map, so a figure can always be attributed to a real boundary.
Independence and ownership
Whoever commissions the work, the fieldwork, the checks and the count are ours. For commissioned studies the data, the report and the underlying dataset belong to the client outright, and we do not resell them. For published polls, the questions, the window and the counts are ours and we publish them whatever they show.
Kenyan electoral polling is additionally governed by the Publication of Electoral Opinion Polls Act, 2012, which requires the sponsor, the fieldwork dates, the sample size and the method to be published alongside any result, and prohibits publication on election day and in the five days before it. Our system enforces that window automatically.
What research can, and cannot, tell you
A study is a measurement, not a prophecy. It captures what a defined group said or did within a defined window. It does not predict an election, and it does not tell you what someone will pay when their own money is on the table, only what they say they would.
Where our numbers are soft, we say so in the report rather than letting a reader assume otherwise. A finding we cannot stand behind is one we will not release.
Questions about the method are welcome and answered in full. Write to hello@voicepointafrica.com. Voicepoint Africa, voicepointafrica.com.