Biostatistics: Sampling Methods in Research Explained

Introduction to Sampling in Research

Research activities, whether clinical trial based, experimentally designed, or product oriented, must exhibit and command interest, enthusiasm, and passionate commitment. To this end, the researcher must capture the essential quality of the excitement of discovery that comes from research well done. The first step in attaining a desired research goal is to develop a scientific approach toward what is being investigated. A requirement within the scientific approach that is oftentimes misunderstood, and consequently wrongly applied, is that of sampling.

In a rather philosophical approach to sampling, Ohlson (1998) states that sampling is "…but part of the whole. Check to make sure I fairly represent my larger connection" (p. 27). With these words, Ohlson is informing the research enthusiast that sampling alone can skew testing results, infuse uncontrollable error into statistical processes, and violate the empirical premise under which the research investigation is being conducted. The remainder of this paper will attempt, by way of definition and example, to describe the manner in which sampling is conducted to achieve the necessary investigation results.

Definitions of Sampling Design Types

With Random Assignment Using a Probability Sample

Whenever the word assignment is used within the constraints of a research investigation, it denotes the representation of variables by way of participants. In random assignment using a probability sample, the research investigator has designed a study wherein participants have been drawn or selected from a larger population and assigned to a treatment and/or control group through random application.

Without Random Assignment Using a Non-Probability Sample

In research investigations such as these, the research investigator may not have access to a general population as a result of study limitations, cost factors, limited resources, or simply because the study is so unique and confined that a population is not available from which to draw a representative sample — meaning all identified subjects must participate in the study. In order to control some of the intervening error, the research investigator will assign the non-randomly selected sample randomly to the treatment and/or control groups.

Without Random Assignment Using a Probability Sample

This particular investigative situation is generally reserved for cases wherein the research investigator has a large enough population from which to randomly draw a representative sample to participate in a study that is seeking differences, relationships, and/or effects between specific variables — where the treatment is the focal point of the research investigation rather than the participant. As such, participants are randomly assigned to the treatment procedure to reduce selection error or bias.

Without Random Assignment Using Non-Probability Sampling

This particular research design is reserved for investigative situations wherein a researcher is dealing with case study research or wherein only one independent variable is being investigated. The non-randomization of the sampling occurs when there is relatively no population from which to draw a sample, or when the investigator is embarking upon a new research frontier and conducting what is generally called a pilot study.

Examples of Sampling Methods

Probability vs. Non-Probability Sampling

Generally speaking, probability sampling is the selection of a sample from a greater population based upon the principle of selection by randomization of chance. This type of sampling is generally more complex, time-consuming, and costly. Non-probability sampling, on the other hand, uses an arbitrary selection process, and there exists no means by which to estimate the probability of any one factor being included in the drawn sample. As such, there is no way to estimate sampling variability or identify possible bias. With probability sampling, as opposed to non-probability sampling, the research investigator is able to estimate sample reliability and sampling error, and to apply the research results to the overall population because the units of the sample are randomly selected.

Probability Sampling Using Random Assignment

Several methods of probability sampling are available to the research investigator: simple random, systematic, sampling with probability proportional to size, stratified, cluster, and multi-phase sampling. As discussing all of these methods is beyond the scope of this paper, only one example will be presented — the simple random assignment technique. In simple random assignment sampling, each member of the selected population has an equal chance of being included in the sample. In addition, each possible combination of population members has an equal chance of composing the research sample, and it is these two properties that define simple random assignment.

For example, consider a state lottery. Most mega-lotteries require generating 6 numbers from a population of 49, wherein each of the six numbers has an equal chance of being selected and each combination has the same chance of being the winning combination. Even though lottery players tend to avoid combinations such as 22-23-24-25-26-27, that sequence has the same chance of winning as any other combination, such as 7-13-28-33-37-46. The probability element of the equation would lie within the idea that 100 participants were randomly chosen from a population of 1,000 to determine whether any ticket holder held the winning ticket.

Non-Probability Sampling Using Random Assignment

Because factors or elements are chosen arbitrarily in non-probability sampling, there exists no way to estimate the probability of any one element or factor being included in the sample. The research investigator is also offered no assurance that each item has an equal opportunity of being included, making it impossible to estimate sampling variability or to identify possible bias — that is, the difference between the true parameter value and the expected value of the estimator. The most common types of non-probability sampling with random assignment are convenience sampling (ad hoc), volunteer sampling, judgment sampling, and quota sampling, wherein participants are randomly assigned to a particular group after being arbitrarily selected.

An example of convenience sampling might involve all female students (10) in the front row of a lecture hall, where 5 are randomly assigned to participate in the treatment group and 5 are randomly selected to participate in the control group.

Non-Probability Sampling Using Non-Random Assignment

In experimental research, "populations" are sometimes hypothetical and do not represent a true or potential population. In such situations, the research investigator selects participants on the basis of availability and assigns individuals to certain groups using a random assignment method. For example, suppose a teacher wants to determine which teaching method — traditional or multiple intelligence — is more effective with a class of 26 eighth graders for reading improvement. In this case, the sample becomes the population and the population becomes the sample: a non-probability selection, since all students participate in the study. To achieve the random assignment requirement, the teacher would randomly select 13 students to participate in each of the two teaching method conditions, giving all 26 participants an equal chance of participating in either instructional situation.

The drawback here lies in the area of generalization to larger populations. That is to say, the results of the research investigation can only be applied to those 26 students, and not to any other general population such as all eighth graders in a school district or all eighth graders in a state.

Probability Sampling Using Non-Random Assignment

Researchers primarily use this type of sampling when certain pre-selected independent variables are being measured. What is significant here is to randomly select the sample of participants from a greater population for the purpose of accurate representation, and then to assign each participant to a particular testable sub-sample. For example, consider a research investigator who wants to determine, through comparison, the effects of a new antidepressant drug versus a placebo. The researcher randomly selects 100 individuals from a larger population of people diagnosed with depression. The sample is then assigned to groups on the basis of arrival order: the first 50 participants to arrive are assigned to the treatment (drug) group and the last 50 are assigned to the placebo group. Unfortunately, in situations like this, a considerable degree of investigative validity is lost.