Blackwell Reference Online

Extract

Convenience samples are best described as what they are not. They are non-probability samples. That is, no attempt is made in their construction to sample randomly from any well-defined population. Random sampling is almost always difficult and expensive, often prohibitively so. Convenience samples, as the name implies, are more easily obtained. They may be self-selected respondents to a mail-out survey. Or they may be readily to hand, patrons of a local gay bar who agree to be interviewed. The non-randomness of the convenience sample militates against straightforward inference from sample to population. The percentage of those who call in to a local talk show host and voice opposition to the proposed location of a halfway house for parolees cannot be taken as an unbiased estimate of the proportion opposed, for any population that a researcher might define. Berk and Freedman (2003) amply describe the mismatch between classical statistical inference and convenience sample data. In particular, they point to the difficult task of trying to link the social processes that lead to the convenience sample data and the assumptions underlying statistical inference. Non-randomness severely limits, but does not completely preclude, the possibility of gaining scientific knowledge from data contributed by a convenience sample. As a prime exhibit, the “gold standard” in medical research, the ... log in or subscribe to read full text