Key Details
A paper in the National Statistics Office’s 121st issue of Sarvekshana journal tests an alternative sampling design for the Annual Survey of Industries (ASI). It is a simulation, not an announcement that ASI sampling has changed.
Measure | Relative Standard Error: Existing Equal-Probability Design | Alternative Size-Based Design |
|---|---|---|
Total workers | 0.2820% | 0.1056% |
Total gross value added (GVA) | 0.1562% | 0.0442% |
GVA per worker | 0.1709% | 0.1065% |
A lower relative standard error indicates a more precise estimate in this comparison. The study draws 1,000 repeated samples under each design from a simulated population based on Uttar Pradesh’s ASI 2017–18 manufacturing data.
Factory Size Helps Estimate Some Totals More Precisely
The existing design gives factories in a sampling group an equal chance of selection. The alternative — probability-proportional-to-size sampling — uses each factory’s recorded number of workers to set its selection chance. Larger factories are more likely to be picked, and the estimates are weighted to account for those different chances.
In the simulation, that approach reduces error for total employment, total GVA and GVA per worker. These are important ASI measures, so the result gives survey designers a reason to investigate whether information already held in the factory list can improve them.
The Trade-Off Depends on What Is Being Estimated
The size-based design does not win across the board. The existing equal-probability approach performs better for the other five of eight measures tested, including workers per factory and GVA per factory. It also reproduces the number of factories in the study’s frame exactly, whereas the size-based estimate varies. The authors note that the factory count can be taken directly from the frame rather than estimated from a sample.
The distinction is therefore not between a good and a bad sampling method. Precision changes with the indicator, and gains for prominent totals must be considered alongside losses elsewhere.
The Result Is a Test, Not a National Survey Finding
The researchers built a simulated population using data from one state, one survey year and selected manufacturing industries. Repeated sampling shows how the two methods perform under those study conditions. It does not establish that the same balance of gains and losses would hold across India’s industries, states and later ASI rounds.
What Is Probability-Proportional-to-Size Sampling?
It is a way of selecting survey units in which a known measure of their size affects their chance of selection. Here, that measure is the number of workers recorded for a factory. Larger factories are sampled more often, then statistical weights are used so the results represent the wider factory population.
Policy Relevance
The paper offers a specific route to testing better industrial statistics, not a case for immediately replacing ASI’s design. The next methodological question is whether the employment and GVA gains persist with real data across different states and years—and whether they justify weaker precision for other indicators. That assessment should be made measure by measure, because ASI serves several policy uses at once.
Follow the Full Paper Here: Sarvekshana, 121st Issue, “Impact of In-Kind Social Transfer on Level of Living of Indian Households”