Design-based estimation

Indian Economy glossary

Also called: Design-based estimates, Model-based estimation · Topic: Economic Data: Census, NSS, Surveys and Statistical Tools · NCERT: Beyond NCERT

Meaning

Design-based estimation is a way of getting survey figures straight from the sample design and the data collected. The sample design is the plan that decides how units are picked. Each sampled unit gets a sampling weight (multiplier), and the weight is the inverse of its chance of selection: weight = 1 ÷ probability of selection. So a household picked with a 1-in-500 chance stands for 500 households. The estimate is the weighted total or weighted ratio of what people actually reported. No assumed statistical model is used to fill gaps or smooth the result. This keeps the estimate transparent, and its margin of error (sampling error) can be worked out from the design itself.

Example

The Periodic Labour Force Survey (PLFS), run by NSO under MoSPI, was revamped from January 2025. It dropped model-based estimates and now publishes only design-based ones. The sample is about 2.7 lakh households a quarter. A monthly unemployment rate is the weighted count of unemployed people ÷ the weighted labour force, using only that month's interviews. Because the method changed, PLFS results from after January 2025 cannot be compared directly with the older series.

Don't confuse with

  • Model-based estimation: it relies on an assumed statistical relationship, for example with other data or past trends, to produce or smooth figures. This is useful where the sample is too thin. But the result is only as good as the model, while a design-based estimate rests only on the sample plan and the responses.

Read more