The sample is the set of cases from which the results were gathered. Most research aims to generalize the results from a sample to a larger population, i.e. to form abstract conclusions from specific instances. Research posits the existence of a population domain from which one takes a specific sample in order to infer conclusions about it. This requires the sample to be generalizable.
Generalizable. A sample is generalizable if results from it apply to a larger population. Every study collects data from a sample but conclusions are invariably made about a population, e. g. a study of college students using browsers might conclude how “people” use browsers. The sample is the students who gave the data but the population to whom the conclusions apply is people in general. One argues the generalization from sample to population by showing the sample is:
- Big enough. Many subjects generalize better than few.
- Representative. Reflects population distribution of gender, age, … e. g. 50% female.
- Unbiased. That the sample was chosen from the population randomly.
So if the sample is not big enough, not representative, or biased, then the generalizability of the research results is threatened. For example, a case study based on one person is rarely enough to generalize to all people. Equally a sample of just men does not represent a general population that is 50% women. Finally, a biased sample is not a good basis for a general conclusion, e. g. to conclude that “Everybody likes me” based on my mother, wife and sister is a biased sample. Describe why the sample has implications for the population concerned based on its size, representativeness and lack of bias.
Sample size. For an experiment, the sample size needed depends on the number of treatment value combinations, e.g. a study of how Usefulness (High, Medium, Low) and Ease of Use (High, Medium, Low) affect web usage is a 3 x 3 study with 9 value combinations or cells. So by the rule of thumb that each cell needs 15 subjects, this research needs about 9 x 15 = 135 subjects. More independent variables and/or values would require more subjects, e.g. a study of how Usefulness, Ease of Use and Gender (Male/Female) affect web usage is a 3 x 3 x 2 study that needs 270 subjects! Check the sample is big enough for the design of your experiment.