Observations on the Design and Implementation of Sample Surveys in China

Surveys in China conventionally sample from local area residential registers, which until recently have been of sufficient accuracy to function as de facto population registers. Due to a combination of large scale internal migration and massive replacement of housing in urban areas, a large fraction of the population currently does not live where registered. Individuals not living where registered are thus ineligible for inclusion in conventionally generated samples. Surveys whose samples depend on access to residential registers are inherently based on an under-enumeration of the population, as well as on a biased representation of the population due to exclusion of unregistered local residents. We report conclusions from, and observations related to, a pilot study designed to test the feasibility and effectiveness of a sampling method that does not depend on residential registers. In the pilot we (1) purposively selected small areas; (2) enumerated the small areas; (3) randomly sampled individuals from the enumeration lists; and (4) interviewed sampled individuals. This approach substantially reduced the underenumeration problem. As implemented, however, its point of departure required previously selected small areas. We describe an extension designed to achieve full coverage of the population of China through sampling of small areas as the penultimate stage of a multi-stage design