Sampling

Probability versus Nonprobability Sampling

It is widely agreed that probability sampling—ensuring that every member of a population has an equal chance of being selected for participation—is most representative of a target population, and will result in the most accurate estimation of the population parameter for a given construct. However, this sampling method is [Page 1208]sometimes not possible, or at the very least prohibitively impractical, for health communication researchers. As one example among many, due to confidentiality reasons there are often no comprehensive published lists of individuals dealing with particular health concerns that a researcher may want to study, making it impossible to be certain that all members of a given population have the opportunity to participate in a research study.

For this reason, much research in health communication utilizes nonrandom sampling methods, including volunteer sampling (any sampling procedure in which the participant elects to be a part of the study), convenience sampling (participants selected on the basis of their accessibility to the researcher), quota sampling (the researcher selects some variable as the basis for a quota and purposefully samples to ensure adequate representation of that quota), and snowball sampling (typically reserved for small or difficult to access populations, the researcher asks an initial participant to recommend others who meet the criteria of inclusion). However, these sampling procedures can become suspect when trying to make global attributions to the population from the sample.

Generalizability

Nonrandom sampling methods, particularly those involving convenience samples of college students, have historically been looked upon by some as lacking in representativeness and therefore generalizability. While this certainly can be the case, there are steps that researchers can take to approach nonprobability sampling in systematic ways that may strengthen arguments for external generalizability.

Ideally, sampling should be conducted in a way that minimizes sampling bias, or systematic differences between the sample and the overall population. As sampling bias increases, the likelihood of sampling error also increases, leading to larger differences between a sample statistic and the actual population parameter for a given variable. Thus, maximizing the similarity between the sample and the overall population can increase the probability that the sample statistics will generalize to the population of interest.

One strategy to accomplish this is to be aware of the demographic breakdown of a population of interest and purposefully collect a sample to mirror that breakdown. This technique is known as quota sampling. A requirement of quota sampling is that the demographic proportions of the population are known. For instance, knowing the proportion of skin cancer survivors that are male or female, those within particular age categories, racial/ethnic categories, and so forth would represent quota sampling. The researcher then constructs a matrix of the relevant demographic characteristics of the population. When collecting data, each cell in the matrix is weighted according to its percentage of the overall population. This method limits sampling bias, in that the researcher ensures that the sample is as reflective of the population as possible. This type of data collection is useful for both quantitative and qualitative methods, as quota sampling is often used in focus group research to match characteristics of participants to the overall population.

...