Research Wiki · Comparison

Primary research vs synthetic data

Primary research vs synthetic data compares approaches that may look interchangeable but answer different questions or rely on different assumptions. The right choice depends on the decision, population, data, realism and cost of error.

Direct answer

Primary research vs synthetic data compares approaches that may look interchangeable but answer different questions or rely on different assumptions. The right choice depends on the decision, population, data, realism and cost of error.

Side-by-side comparison

DimensionPrimary researchsynthetic data
Primary purposeCollect new evidence for a specific decisionGenerate artificial records from a model
Evidence formNew interviews, surveys, observation or experimentsSource data or assumed generative process
Main strengthDecision-specific and controllable designPrivacy, simulation and augmentation use cases
Main riskCosts time and can still be biasedModel bias and fidelity limits

How to choose

Choose Primary research when

Collect new evidence for a specific decision and its assumptions fit the intended population and decision.

Choose synthetic data when

Generate artificial records from a model and its assumptions fit the intended population and decision.

Do not choose by familiarity

A method used in the previous study is not automatically the right method for the next decision.

Predefine the action

State what different outcomes will cause the organisation to do before seeing the result.

When a combined design is stronger

Many comparisons are false binaries. One method may establish structure or prevalence while another explains context, trade-offs or mechanisms. A sequential design is useful when the first stage improves the instrument, alternatives or interpretation of the second.

Common comparison mistakes

  • Comparing labels while ignoring different estimands, populations or task formats.
  • Assuming the method with more data is automatically more valid.
  • Using cost or speed as the only selection rule.
  • Combining outputs that were generated under incompatible definitions.

Selection checklist

Editorial and methodology note

This reference is written by Ninth Atlas as a decision-oriented explainer. It separates definition, design, analysis and limitations so a method is not mistaken for an answer. Final study specifications should be reviewed against the actual population, evidence, risk and regulatory context.

Sources and further reading

Need the method applied to a live decision?

Share the market, customer, product or investment question that needs a defensible evidence design.

Brief Ninth Atlas