Sampling: Summary

A sample is a subset of a population (group of individuals of interest to the researcher). The type of sample selected determines the degree to which research results can be generalized to the population as a whole (external validity).

There are two sources of error that limit generalizability: sampling error (chance variation) and sample bias (constant error) which results from inadequate research design. Sampling error (but not sample bias) can be taken into account using statistics.

Probability samples are representative of the population. They permit generalization to the population from which they are drawn. There are two types of probability samples: Random and stratified.

Random - each individual in the population has an equal chance of being selected for the sample.
Stratified - a miniature representation of the larger population with regard to proportions within selected strata (e.g., gender, education, socioeconomic level). Individuals are randomly selected within strata.

A table of random numbers or the random number function in Excel can be used to select a random sample from a population.

Nonprobability samples are of 3 general types:

In all these types, selection is non-random. It is worthwhile attempting to increase representativeness so as to improve validity.

Quota - numbers within levels are determined by the researcher. Selection may or may not be random within each quota.
Purposive (including snowball) - a select group is targeted with sample obtained in non-random way.
Convenience - sample of available participants, an accidental sample.

Large samples are preferable to smaller ones, but may not be obtainable due to limitations of access, time, or resources. Smaller sample sizes may be sufficient for small populations, strong treatment effects, good response/return rates, and few variables being studied.

The Experience Sampling Method (ESM) uses 3 techniques for having participants sample their own behaviors:

occurrence or nonoccurrence of the specified event in response to a signal - Signal contingent
occurrence or nonoccurrence of the specified event at pre-determined times - Interval contingent
occurrence of the event - Event contingent

The results are limited in that they are subjective reports and lack detail. Richer data may be obtained using diaries. They have the advantage of being less dependent upon memory. The method is less intrusive than direct observation or behavior mapping.

Terms to know (define each before clicking to see the definition in a pop-up message)

 Accidental sample Constant error Convenience sample Event contingent Experience Sampling Method (ESM) External validity Interval contingent Nonprobability sample Nonprobability samples (types) Population Probability sample Probability samples (types) Purposive sample Quota Quota sample Random sample Sample Sample bias Sampling error Signal contingent Snowball sample Strata Stratified sample