Glossary

Synthetic population

Also called synthetic populations

The generated set taken as a whole, built to match the known statistical margins of a territory. The term comes from microsimulation, not from research.

A synthetic population is a set of generated individuals arranged to match the known statistical margins of a territory: the age pyramid, the spread of social categories, the geography. The term is older than this market — it comes from microsimulation, where transport planners and epidemiologists have been building whole populations for decades.

Inside a research project it means the sample once assembled. The difference from a respondent is the difference between the whole and the unit: the population is what you control statistically, the respondent is what you interview.

Open the animation full screen

What it is not

Matching every margin separately is not enough, and this is the part vendors tend to skip. A generation can reproduce the age distribution perfectly and the income distribution perfectly, and still produce age × income crossings nowhere near the real thing. That is the structural limit of the exercise, and it is where methods genuinely differ from one another.

What a buyer can check

Which margins were targeted, against which reference source. Whether the crossings were constrained or only the simple margins. And whether a gap to target was measured after generation: a described method is not a controlled one. See calibration.

Frequently asked questions

What is a synthetic population?

A set of generated individuals arranged to match the known statistical margins of a territory: the age pyramid, the spread of social categories, the geography, and ideally how they combine. The term comes from microsimulation, where transport planners and epidemiologists have built whole populations for decades, long before market research used it.

What is the difference between a synthetic population and a synthetic respondent?

The population is the whole, the respondent is the unit. The population is what you control statistically; the respondent is what you interview, the answers a model gives for one profile.

Does a well-built synthetic population guarantee good answers?

No. The structure of the population and the quality of the answers are two different layers. A population can match every margin and crossing while its simulated respondents still agree with each other far too much. Each has to be measured separately.