If you plan to run this kind of study via WhatsApp, see how Yazi works to operationalise the sampling and fieldwork below.

The foundation: what is a master sampling frame?

A master sampling frame is the comprehensive list or structure from which every survey sample is drawn. In a multi-country study, the goal is a consistent, efficient probability sample in each country, so every person in the target population has a known, nonzero chance of selection.

Master sampling frame design

Each country typically builds its own frame, coordinated across countries for comparability. A national statistics office often uses the latest census to build an area-based frame, a list of enumeration areas (EAs) or villages. The advantage of a master design is that certain sampling stages can be reused across surveys or over time, which meaningfully cuts the cost of building a fresh frame for every single study.

The power of an integrated master sample

An integrated master sample takes this further: a large sample of primary units (EAs, say) selected once and reused for multiple surveys over several years. A country might select 500 EAs after a census, then use that same set as the foundation for a labour force survey one year and a health survey the next. Bangladesh has used exactly this approach, reusing its Labour Force Survey units across other studies for major cost savings. For practical examples in African markets, see Yazi's case studies. Build once, use many times, is the core of an efficient frame.

Building blocks of the frame

Frame units: definition and hierarchy

A frame unit is the basic element listed in a sampling frame, often a geographical area in household surveys. The most common are small areas like census enumeration areas: an EA in Tanzania, for example, contains an average of roughly 86 households. Units sit in a clear hierarchy, households within EAs, EAs within districts, districts within regions, which enables multi-stage sampling and reporting at any of those geographic levels. Each frame unit needs to be unique and clearly defined to avoid overlaps or omissions.

Grouping small frame units

When basic frame units are very small, visiting each one individually is inefficient, so researchers group several neighbouring units into a larger primary sampling unit (PSU). Travelling to one cluster of three adjacent villages is usually cheaper than visiting three villages scattered far apart. This grouping has to happen before selection, by ordering units geographically and combining consecutive ones, to preserve statistical integrity.

Frame documentation and maintenance

Frame documentation covers maps, population counts, and unique codes for every unit; maintenance is the ongoing work of keeping that information accurate. Populations shift constantly, new homes get built, people migrate, boundaries change, so an outdated census frame can miss whole new settlements entirely (a problem known as noncoverage). For sources that support keeping a frame current, see Yazi's data resources for Africa.

The gold standard: multi-stage stratified sampling

The most trusted method for large-scale surveys is the multi-stage stratified area probability sample, used by nearly every major survey in Africa, including the Demographic and Health Surveys and Afrobarometer. It balances statistical rigour with what's actually practical on the ground.

Stratification by region and urban/rural status

In most African countries, region and urban or rural residence strongly predict lifestyle, opinions, and behaviour, so stratification is almost always built around both. This creates strata like "Northern Region, Urban" or "Southern Region, Rural," with a sample drawn from each separately. If a country is 70% rural, the design ensures roughly 70% of selected survey locations are rural too, guaranteeing every major region and urban/rural split is represented in the correct proportion.

PSU and SSU selection using PPS

Once strata are defined, Primary Sampling Units, villages or EAs, get selected using Probability Proportional to Size (PPS), which gives larger PSUs a higher chance of selection in proportion to their population. That keeps every individual's overall chance of inclusion equal. After a PSU is chosen, the team selects a fixed number of Secondary Sampling Units, typically households, usually via simple random or systematic sampling. PPS for PSUs plus a fixed number of SSUs per PSU produces a self-weighting design, which simplifies analysis later. Afrobarometer uses exactly this combination: PPS to select EAs, then a fixed number of households within each one.

Fine-tuning your sample for better insight

Allocating the sample across strata

Once strata are set (region by urban/rural, say), you decide how to spread the total sample across them, proportionally to population size, equally per stratum for easier country-level comparison, or oversampled where a small but important subgroup needs enough respondents for reliable analysis on its own.

Sample size and margin of error

Sample size and margin of error are directly linked: a larger sample shrinks the margin of error, but the gains diminish as sample size grows. A sample of around 1,000 typically yields a margin of error of about ±3% at 95% confidence, which is why it's such a common standard for national surveys; many multi-country polls, including Pew Research Center's, target roughly 1,000 interviews per country for this reason.

Remember that margin of error only accounts for sampling randomness. It says nothing about other sources of error, like question wording or nonresponse bias.

From theory to reality: fieldwork and quality control

Selecting households and respondents

After a PSU is chosen, fieldwork needs an unbiased way to start: a sampling start point, a random landmark or GPS coordinate, followed by a strict protocol such as selecting every fifth household in a set direction. This "random walk" spreads interviews unbiasedly across the whole area. Once a household is selected, the actual respondent still needs to be chosen randomly, using the "last birthday" technique or a Kish grid, rather than whoever happens to answer the door. Many surveys, Afrobarometer included, also alternate a gender quota between successive households to keep the sample balanced.

Handling inaccessible areas and noncoverage

Some areas are inaccessible due to conflict, natural disaster, or extreme remoteness, producing noncoverage, where part of the population is excluded from the sample entirely. Reputable organisations are transparent about this: Afrobarometer documents exclusions due to insecurity directly in its technical reports for each survey. That's a real limitation, but documenting it lets anyone using the data understand its potential impact.

Interpenetrating subsamples for quality control

One classic quality check is drawing two or more independent subsamples, each a miniature version of the full sample, and assigning them to different fieldwork teams or timeframes. Since they should produce similar results within the margin of error, any large, unexplained difference signals a problem, interviewer bias or a procedural error, and gives you a built-in way to measure the consistency of the fieldwork itself.

Making sense of the data: the role of weighting

Weighting for cross-country comparability

Combining data from multiple countries raises an immediate question: how much influence should each country have? Pooling raw data lets a country with a larger sample or population, Nigeria, for instance, dominate the overall average. To prevent that, multi-country datasets include a combined weight. Afrobarometer's "Combinwt," for example, standardises every national sample as if it were equal in size, so a cross-country average treats each country fairly.

Within-country vs. combined weights

A multi-country dataset typically carries two weight types, illustrated below with example figures:

Using the correct weight isn't optional. It's a fundamental part of making cross-country comparisons fair and accurate.

Frequently asked questions