Research Sampling Methods: Probability and Nonprobability Techniques

A variation of this is the proportionate sample. In this approach, participants are sampled from each stratum in proportion to how common they are in the actual population. For example, if a university is 40% white and 21% Black, a proportionate sample would reflect those exact percentages.

Exam Tip: While a proportionate sample is a probability sample because we can estimate the likelihood of selection, it is not strictly "random" in the same way a simple random sample is. This is because individuals in smaller groups have a higher mathematical probability of being selected to meet the required proportion.

Cluster and Multi-stage Sampling

When a population is spread out geographically, researchers often use cluster sampling. Instead of picking individuals, they randomly select naturally occurring groups, or "clusters" (such as specific cities or universities).

If the researcher needs to go even deeper, they use multi-stage cluster sampling. This is a hierarchical approach where they first sample large clusters, then smaller clusters within those, and finally individuals within the smallest clusters.

Example of Multi-stage Sampling:

1. Randomly select 5 universities (Large clusters).
2. Randomly select 3 majors from each of those universities (Smaller clusters).
3. Randomly select 25 students from each of those majors (Final sample).

Nonprobability Sampling

In many psychological studies, researchers cannot estimate the probability of an individual being chosen. These are known as nonprobability samples.

The most common type is the convenience sample, which involves using participants who are readily available or who volunteer to participate (e.g., recruiting through flyers or online ads). While convenience sampling is often criticized, it is not "bad science". It is highly useful when the researcher is not trying to describe a whole population, but rather looking for relationships between variables.

The Risk of Misgeneralization

The primary danger of nonprobability sampling is misgeneralization. This occurs when researchers apply their findings to a population that differs significantly from the sample they actually studied.

For instance, a massive portion of psychological research relies on college students—often from "WEIRD" (White, Educated, Industrialized, Rich, and Democratic) societies. If a study conducted only on private university students claims to represent all college students globally, it risks misgeneralization because those two groups may differ in important ways.

Accuracy and Error in Sampling

Even with the best intentions, a sample will always differ somewhat from the general population. This difference is known as sampling error.

The Error of Estimation

The error of estimation (often called the margin of error in polling) is the degree to which the sample data is expected to deviate from the actual population values. Three main factors influence this error:

1. Sample Size: Larger samples decrease the error of estimation.
2. Population Size: Larger populations increase the error of estimation.
3. Variance: Higher variance (diversity) in the population increases the error of estimation.

The Non-Response Problem

Even if a researcher selects a perfect probability sample, they may encounter the non-response problem. This happens when selected individuals refuse to participate due to lack of time, distrust, or disinterest. This can bias the sample because the people who choose not to respond might be systematically different from those who do respond, destroying the representativeness of the sample.

Determining Sample Size: Power Analysis

Researchers must decide how many participants are needed for a study. While larger samples are generally better, they are not always feasible due to time or cost. To find the right balance, researchers use a power analysis.

Power refers to the ability of a research design to detect a statistical effect if one actually exists. If a study has low power, it might fail to detect a real relationship between variables. Because larger samples increase power, they make it easier for a researcher to detect even small effects.

Summary Table: Sampling Comparison