Sampling

Sampling is the selection of a given number of units of analysis (people, households, firms, etc.), called cases, from a population of interest. Generally, the sample size (n) is chosen in order to reproduce, on a small scale, some characteristics of the whole population (N)

Sampling is a key issue in social research designs. The advantages of sampling are evident: feasibility of the research, lower costs, economy of time, and better organization of the work. But there is an important problem to deal with: that is, sampling error, because a sample is a model of reality (like a map, a doll, or an MP3) and not the reality itself. The sampling error measures this inevitable distance of the model from reality. Obviously, the less it is, the more the estimates are close to reality. Unfortunately, in some cases, the sampling error is unknowable.

There are two main families of sampling methods: probability (random) sampling and nonprobability sampling, respectively typical of (but not exclusive to) quantitative and qualitative research.

Probability sampling, definitively codified in the 1930s by the Polish statistician Jerzy Neyman, is characterized by the condition that all units of the population have an (theoretically) equal, calculable (i.e. known), nonzero probability of being included in the sample.

Probabilistic samples are considered representative of reality: What can be said about the sample can be extended to the reality of what is sampled by statistical inference. Another advantage is that the sampling error, which is a crucial datum to assess the validity of the sample, is calculable: This is possible only for probability samples. The main problem, however, is that researchers need the complete list of the target population (i.e. the sample frame), though sometimes the exact number of the population is sufficient, to extract the sample, and often this is impossible to obtain (e.g. when a researcher wants to study the audience of a movie).

There are several types of probability sampling. The most common are simple, systematic, and stratified random sampling. Other types of probability samples are multi-stage, cluster, multi-phase, and spatial sampling.

In most cases, the size of a probability sample is determined by the following formula:

where z, refers to the confidence level of the estimate (usually fixed at 1.96, corresponding to a 95% confidence level), pq is the variance (that is unknown and then fixed at its maximum value: 0.25), N is the size of the population, E is the sampling error (often < 0.04).

Nonprobability samples are generally purposive or theory driven. This means they are gathered following a criterion the researcher believes to be satisfying to obtain typological representativeness. This latter is achieved, when the researcher has sufficient members of all the main categories of interest to be able to describe with confidence their patterned similarities and differences.

Being purposive, nonprobability samples are rather heterogeneous. Up to 16 different qualitative sampling strategies have been listed for choosing a nonprobability sample. It is almost impossible to give an exhaustive list, because they are continuously open to integrations and new solutions. However, quota, snowball, purposive, theoretical, and accidental sampling are among the most common types of nonprobability sampling techniques.

...