Glossary

Calibration

Also called empirical anchoring

The operation that ties a generated population to observed real-world data — and the measurement of the gap that remains once it is done.

To calibrate is to hold a generated population against an observed one: you start from real data — census, public surveys, sector statistics — translate it into targets, and build under those targets.

The half that matters comes afterwards. Once the population is drawn, a gap to target remains. Those gaps, measured and written down, are what separates a method that is described from one that is controlled. A vendor who explains the machinery at length and never publishes a single gap has shown you nothing.

Open the animation full screen

What it is not

Statistical conformity is not behavioural validity. Recovering the right age pyramid says nothing about whether the answers are any good. These are two separate controls, and the second is much harder — which is precisely why it is published far less often.

What a buyer can check

Which reference sources, for which market, from which year. Which gaps were measured, on which variables. And whether the calibration was done on the market in question or borrowed from another country — a model fitted on US data does not describe France.

Frequently asked questions

What is calibration in synthetic research?

It is the operation that ties a generated population to observed real-world data: you start from the census, public surveys or sector statistics, turn them into targets, and build the population under those targets. Then you measure the gap that remains. Those gaps, measured and written down, separate a method that is described from one that is controlled.

Does calibration prove the answers are right?

No. Statistical conformity is not behavioural validity: recovering the right age pyramid says nothing about whether the answers are any good. That is a second, harder check, run on questions the system did not see during calibration.

What should a buyer ask about calibration?

Which reference sources, for which market, from which year. Which gaps were measured, on which variables. And whether the calibration was done on the market in question or borrowed from another country: a model fitted on US data does not describe France.