Sampling is an act of generalization that we participate in all the time. Consider the free samples at your local grocery store.
When a representative from the deli offers you a square of pizza, you are being asked to draw conclusions about the taste and value of the product itself. Offering a whole pizza to every customer would be expensive, difficult to coordinate and, in all likelihood, a waste of time and effort. Chosen well, the samples will provide customers with enough information to decide whether a whole pizza is worth purchasing. The sample is a representative part, an extract from which to generalize back to the whole.
Sampling: a scientific process
In practice, identifying a representative part of a subject, event or population of interest is one of the more challenging aspects of study design. Let’s say we want to use an in-hospital survey to measure patient satisfaction. How will we select a group of patients to participate in our study?
To begin, we must differentiate between the theoretical population and the accessible population. The first might include any patient who has ever stayed in a hospital overnight. The second is limited to those who stayed in hospital on a specific night. Since we cannot hope to survey every member of the theoretical population, we must identify members of the accessible population to contact. The resulting subset of individuals will be our sampling frame.
However, we have to be cautious about introducing sampling errors and non-sampling errors into the frame. Sampling errors are the differences between the sample and the population being studied. In other words, they’re errors that occur because the data is from a part rather than the whole. Non-sampling errors are statistical errors caused by human error. These can include data entry errors or biased questions in a survey. In our hospital survey, those who could not or did not respond to the survey could introduce non-sampling errors.
Now that we’ve narrowed our population of interest, we must decide how to select the sample. Probability sampling is one of two primary strategies we might consider. In probability sampling, every member of the sampling frame has the potential to be selected for the study. Selection is random, and the probability of a member being chosen can be calculated. Knowing the probability of selection allows us to generalize to the population.
In non-probability sampling, some members will have a greater chance of being selected than others, while some will have no chance of being selected at all. The probability of a member being chosen cannot be calculated, making it hard for researchers to know how well they have represented the theoretical population. Often researchers will turn to non-probability sampling only when other data collection methods are not possible.
Convenience sampling is a type of non-probability sampling, and it illustrates both the benefits and drawbacks of this approach. In convenience sampling, the most accessible members from the sampling frame are selected. For example, we might find that certain patients completed positive satisfaction surveys one year ago. It would be convenient to survey only those patients who already had a positive hospital experience. Probably they would be more willing to complete our survey. But in choosing only these patients, we must also ask whether it’s reasonable to generalize from their experiences.
While all sampling methods are subject to error, researchers must always keep their objective in view: to obtan meaningful information about the theoretical population. Fundamental to this goal is a workable sample.
Source: At Work, Issue 63, Winter 2011: Institute for Work & Health, Toronto