In many research studies, the investigator does not begin with a finite group of per-sons, animals, or objects in which each member has a known, nonzero probability of being plucked out of the population for inclusion in the sample. In such situations, the sample is technically referred to as a nonprobability sample. Occasionally, an author indicates directly that one or more nonprobability samples served as the basis for the inferential process. Few authors do this, however, and so you must be able to identify this kind of sample from the description of the study’s participants.
Although inferential statistics can be used with nonprobability samples, extreme care must be used in generalizing results from the sample to the population. From the research write-up, you probably will be able to determine who (or what) was in the sample that provided the empirical data. Determining the larger group to whom such inferential statements legitimately apply is usually a much more difficult task.
We next consider four of the most frequently seen types of nonprobability sam-ples: purposive samples, convenience samples, quota samples, and snowball samples.
Purposive Samples. In some studies, the researcher starts with a large group of potential participants. To be included in the sample, however, members of this large group must meet certain criteria established by the researcher because of the nature of the questions to be answered by the investigation. Once these screening criteria are employed to determine which members of the initial group wind up in the sample, the nature of the population at the receiving end of the “inferential arrow” is different from the large group of potential persons with whom the researcher started.
The legitimate population associated with the inferential process is either (1) the portion of the initial group that satisfied the screening criteria, presuming that only a subset of these acceptable people (or objects) were actually measured; or (2) an abstract population made up of people (or objects) similar to those included in the sample, presuming that each and every “acceptable” person (or object) was
measured. These two notions of the population, of course, are meant to parallel the two situations depicted earlier in Figure 5.1.
Excerpt 5.8 illustrates the way researchers sometimes use and describe their purposive samples. In this passage, notice how the researchers set up inclusion cri-teria for both members of the adolescent/parent dyads recruited into the study. As you can see, only younger adolescents and their parents were allowed into the study.
EXCERPT 5.8 • Purposive Samples
A purposive sample of 94 adolescents/parent dyads was recruited from eight middle schools within a single school district in southern California. The inclusion criteria for adolescents are as follows: between the ages of 12 and 15 years; able to read and speak English; signed informed assent; and signed informed parent consent.
The inclusion criteria for parent participants are as follows: able to read and speak English; have legal custody of adolescent participant; and signed consent form for participation for self and child.
Source: Rutkowski, E. M., & Connelly, C. D. (2010). Obesity risk knowledge and physical activity in families of adolescents. Journal of Pediatric Nursing, 26(1), 51–57.
The full research report from which Excerpt 5.8 was taken included a highly detailed description of the people who composed the sample (including demo-graphic information on the adolescents’ grade point average (GPA), height, weight, and year in school, as well as information of the parents’ age, educational level, and marital status). Such descriptions, along with a clear articulation of the inclusion criteria, are essential in research reports based on purposive samples. The reason for this is simple—the relevant populations associated with purposive samples are ab-stract rather than tangible. As pointed out earlier, the nature of an abab-stract popula-tion is determined by who or what is in the sample. Clearly, you cannot have a good sense of the population toward which the inference is directed unless you have a good sense for who was in the sample.
Convenience Samples. In some studies, no special screening criteria are set up by the researchers to make certain that the individuals in the sample possess certain characteristics. Instead, the investigator simply collects data from whoever is available or can be recruited to participate in the study. Such data-providing groups, if they serve as the basis for inferential statements, are called convenience samples.
The population corresponding to any convenience sample is an abstract (i.e., hypothetical) population. It includes individuals (or objects) similar to those in-cluded in the sample. Therefore, the sample–population relationship brought about by convenience samples is always like that pictured earlier in Figure 5.1b.
Excerpt 5.9 illustrates the use of convenience samples. In this excerpt, the resear-chers clearly label the kind of sample they used. Not all researresear-chers are so forthright.
EXCERPT 5.9
• Convenience Samples
Our survey targeted a cross-section of students enrolled in various courses in the College of Business Administration across all levels (e.g., introductory courses to senior level courses) and areas of study (e.g., introduction to business, finance, marketing, strategic management). We derived this convenience sample from classes that comprise a part of our pre-business and business core curriculum. We selected the specific classes on the basis of the availability of the researchers to administer the survey in person and the flexibility of the faculty member in the classroom.
Source: Sipe, S., Johnson, C. D., & Fisher, D. K. (2009). University students’ perceptions of gender discrimination in the workplace: Reality versus fiction. Journal of Education for Business, 84(6), 339–349.
EXCERPT 5.10
• Quota Samples
Participants were identified using a national quota-sampling procedure. . . . Quota sampling in the context of this study refers to a sampling method in which the first of a pre-determined number of participants are selected, the number being 4000 for this survey. After the number of participants exceeded approximately 4000, the sur-vey was terminated. The quota-sampling strategy stratified on four age groups:
25–29, 30–34, 35–39 and 40–45 to match 2006 U.S. Census data.
Source: Kronenfeld, L. W., Reba-Harrelson, L., Von Holle, A., Reyes, M. L., & Bulik, C. M. (2010).
Ethnic and racial differences in body size perception and satisfaction. Body Image, 7(2), 131–136.
It should be noted that the statements presented in Excerpt 5.9 do not consti-tute the full description of the convenience sample used in this study. The re-searchers provided information on the students’ age, year in school, ethnicity, political orientation, work experience, and GPA. Unfortunately, many researchers put us in a quandary by not providing such descriptions. Unless we have a good idea of who is in a convenience sample, there is no way to conceptualize the nature of the abstract population toward which the statistical inferences are aimed.
Quota Samples. The next type of nonprobability sample to be considered is called a quota sample. Here, the researcher decides that the sample should contain X percent of a certain kind of person (or object), Y percent of a different kind of person (or object), and so on. Then, the researcher simply continues to hunt for enough people/things to measure within each category until all predetermined sample slots have been filled.
In Excerpt 5.10, we see an example of a quota sample. In this investigation, the researchers wanted to have an overall sample size of 4,000, with a quota for each of four age categories. These quotas were determined by first examining census data
to find out what proportion of women ages 25–45 was in each 5-year category; then, each of those proportions was multiplied by 4,000 to arrive at the needed number of women in each of the study’s age strata.
On the surface, quota samples and stratified random samples seem to be highly similar. There is, however, a big difference. To obtain a stratified random sample, a finite population is first subdivided into sections and then a sample is selected randomly from each portion of the population. When combined, those ran-domly selected groups make up the stratified random sample. A quota sample is also made up of different groups of people that are combined. Each subgroup, however, is not randomly extracted from a different stratum of the population; rather, the re-searcher simply takes whoever comes along until all vacant sample slots are occu-pied. As a consequence, it is often difficult to know to whom the results of a study can be generalized when a quota sample serves as the basis for the inference.
Snowball Samples. A snowball sample is like a two-stage convenience or purposive sample. First, the researcher locates a part of the desired sample by turning to a group that is conveniently available or to a set of individuals who possess certain characteristics deemed important by the researcher. Then, those individuals are asked to help complete the sample by going out and recruiting family members, friends, acquaintances, or coworkers who might be interested (and who possess, if a purposive sample is being generated, the needed characteristics). Excerpt 5.11 illustrates how this technique of snowballing is sometimes used in research studies.
EXCERPT 5.11
• Snowball Samples
We recruited participants for the present study [by] using a snowball sampling tech-nique. Specifically, we e-mailed a Web link to an online survey to approximately 45 individuals who were employed full-time and whom we knew personally or pro-fessionally. . . . We invited individuals to participate [and] we asked individuals to forward the link to other fulltime employees.
Source: Culbertson, S. S., Huffman, A. H., & Alden-Anderson, R. (2010). Leader–member exchange and work–family interactions: The mediating role of self-reported challenge- and hindrance-related stress. Journal of Psychology, 144(1), 15–36.