Design and conduct of surveys (cont.)

Sampling

The sample of units included in a demographic survey should be as representative as possible of the population being studied. In nearly all circumstances, this means that it should be a randomly-selected probability sample. Selecting a probability sample is only possible if a sampling frame exists that identifies all the units of analysis in the population and makes it possible to locate them. For nationally-representative household surveys, the sampling frame is usually based on the most recent census. To reduce fieldwork costs, many surveys first randomly select a limited number of smallish areas ("clusters") in which to collect data and then randomly select households to interview from within those areas.

If data are collected successfully from a probability sample of this type, the measures obtained from it will provide unbiased estimates for the population (conditional on those measures themselves also being unbiased). Moreover, standard statistical methods can be used to quantify the amount of uncertainty in the measurements.

Other methods of selecting a sample, such as quota sampling of set of units that matches several demographic characteristics of the population, might yield an unbiased sample. However, it is impossible to determine whether or not the strategy has succeeded. In addition, moreover, it is impossible to quantify the extent to which measures obtained from a non-random sample are likely to differ from those for the population as a whole.