A simple random sample is similar to putting the names of all students in a hat, and then drawing 200 tickets without replacement . A more practical procedure is to create a frame , or a list, of all WMU students and have a computer randomly select 200 names from the frame. On a smaller scale, one may use a table of random numbers to select subjects from the frame. The method of simple random sampling has ideal statistical properties in the sense that each element in the frame has an equal chance of being selected. However many target populations have no convenient frames available. Examples are the populations of Kalamazoo residents, customers of Meijer Stores, potential credit card customers, and US registered voters. In these cases, more convenient frames that simply approximate the target population may be used, like phone books and mailing lists.
A k-in-1 systematic sample selects every kth subject on a list or a sequence. For example, every 10th customer entering the store may be approached for a survey, or every 25th number in the phone book may be called.
A stratified random sample partitions the population into subsets called strata , and selects a simple random sample from each stratum. In the survey of WMU undergraduates, we may choose to draw 50 students from each of the freshman, sophomore, junior, and senior strata, instead of 200 students from the combined student population.
A cluster sample would randomly select, say, 5 classes of 40 students each. Each class is called a cluster of students. Cluster sampling works best when heterogenous clusters are available, i.e. each cluster is approximately representative of the population. This means that classes need to contain freshmen and seniors, men and women, etc. Therefore, it seems better to limit the frame to containing general education classes, because classes for specific majors tend to be less heterogenous.
The sampling designs discussed above are examples of probability samples , because their use of random selection (of clusters, or within strata) gives them desirable statistical properties. Informal and less scientific surveys may conduct nonprobability samples , like randomly approaching people at the mall or WMU students at the Berhard Center. Nonprobability samples tend to be less representative of the target population because interviewers tend to choose more friendly-looking people, or more attractive people. Furthermore, the shopping mall sample would tend to exclude segments of the population who don't shop often. Approaching students at the Bernhard Center may tend to exclude dormitory residents who have meal plans.
When populations are small, data may be collected on the whole population instead of a sample. This is called a census survey instead of a sample survey.