Representative Sample

The notion of a representative sample is often misunderstood. The general intent usually is to draw a sample from a population so that particular properties of that population can be estimated accurately from the sample. For example, political scientists may draw samples from the population of voters to predict with some certainty the outcome of an election.

In general, only properly drawn probability samples such as EPSEM samples will guarantee that the population to which one wants to generalize is properly "represented." On the other hand, a generally erroneous notion is commonly expressed that, in order to achieve "representativeness," it is desirable to draw a stratified sample using particular "quotas" (quota sampling) where demographic characteristics such as age, gender, race, etc., are properly "balanced," to match precisely the makeup of the underlying population. This notion is false. The precision of the estimates (such as voting margins) for a population computed from such a sample will only be enhanced if the variables that one is attempting to match (age, gender, race, etc.) are (strongly) related to the outcome variable of interest (e.g., voting behavior). However, in practice such a-priori knowledge is usually elusive, and applying such quota sampling methods may yield grossly misleading results.

Refer to, for example, Kish (1965) for a detailed discussion of the advantages and characteristics of probability samples and EPSEM samples. In STATISTICA, you can create probability (EPSEM) and stratified random (probability) samples using the Random Sampling options.