Samples should be representative of the population from which they are drawn. A completely unbiased sample is one that is highly representative of the population. How do you create a completely unbiased sample? First, you would randomly sample from a population that contains all individuals in the population. Second, you would contact and obtain completed responses from all individuals selected to be in the sample. Such standards are rarely achieved. Even if random sampling is used, bias can be introduced from two sources: the sampling frame used and poor response rates. Moreover, even though nonprobability samples have more potential sources of bias than probability samples, there are many reasons (summarized in Table 7.3) why they are used and should be evaluated positively.
TABLE 7.3 Advantages and disadvantages of sampling techniques
The sampling frame is the actual population of individuals (or clusters) from which a random sample will be drawn. Rarely will this perfectly coincide with the population of interest—some biases will be introduced. If you define your population as “residents of my city,” the sampling frame may be a list of telephone numbers that you will use to contact residents between 5 p.m. and 9 p.m. This sampling frame excludes persons who do not have telephones or whose schedule prevents them from being at home when you are making calls. Also, if you are using the telephone directory to obtain numbers, you will exclude persons who have unlisted numbers. As another example, suppose you want to know what doctors think about the portrayal of the medical profession on television. A reasonable sampling frame would be all doctors listed in your telephone directory. Immediately you can see that you have limited your sample to a particular geographical area. More important, you have also limited the sample to doctors who have private practices—doctors who work only in clinics and hospitals have been excluded. When evaluating the results of the survey, you need to consider how well the sampling frame matches the population of interest. Often the biases introduced are quite minor; however, they could be consequential to the results of a study.