5.4 Representativeness and random sampling

The main problem with representativeness is that to achieve technical representativeness of a community that has more then a few dozen members, the sample size would have to be far too large to analyse. Neuman (cf. 1997: 222) says that for a population of 1,000 a sample of 300 people and for a population of 150,000 a sample of about 1,500 people would be appropriate. But it has been established that smaller sample sizes are sufficient for linguistic surveys to be demonstrably representative (cf. Milroy and Gordon 2001: 28; Milroy 1987a: 27; Tagliamonte 2006: 22-23). This is quite important as weighing the increase in statistical representativeness against the additional time and cost of achieving it is always a problem (cf. Tagliamonte 2006: 32; Milroy and Gordon 2001: 26). But mainly, it is fundamental to have a sample that can withstand some statistical criticism and to be careful about what claims to make about the data (cf. Milroy 1987a: 20).

In order to have a representative sample, there must be some kind of sampling frame, for example a telephone-book or an electoral register. The researcher must randomly pick people from this frame by, for example, selecting every nth person. However, there are some drawbacks to this approach as each frame only offers a selection of certain people, which according to the respective sampling frame must meet certain requirements. For instance, while a telephone-book will only list people with a telephone connection, an electoral register will only include people who are legally entitled to vote. Also, the fieldworker him/herself may not be completely unbiased in his/her choice of frame, due to personal preferences. And even if the frame should represent the whole population, some people may be unable or unwilling to participate (cf. Milroy and Gordon 2001: 24-26; Tagliamonte 2006: 19).

More information on how many people to choose is found here.

Created with the Personal Edition of HelpNDoc: Easily create iPhone documentation