How can sampling errors be reduced?


The so-called non-sampling error is given by the difference between the mean value of the original target population and the mean value of the actual study population:

The target population is the population that the study would like to refer to and from which the sample should therefore be drawn. For the following reasons, however, it can happen that the originally intended target population differs from the actually achievable study population:

  • "Listing and framing": This means the problem that no current, no complete or no clear list of all elements of the population is available or can be constructed. Missing or multiple entries of individual observations on the list can lead to a discrepancy between the target and study population.
  • "nonresponse": "nonresponse" means the complete (unit-nonresponse) or partial (item-nonresponse) lack of information on individual observations. If this behavior is systematically related to the feature of interest or to another feature which in turn is related to the feature of interest, then this also leads to a non-sampling error. Unsystematic "nonresponse" does not result in a non-sampling error, but it does reduce the precision of population estimates by reducing the sample size.
  • Measurement error: If the features of interest can only be recorded with a systematic measurement error, this also leads to a non-sampling error. An unsystematic measurement error does not cause an error, but in turn reduces the precision of population estimates.

We will come back to the potential influence of "nonresponse" in more detail later.