Choosing your sample (cont.)
Once you have identified a sampling frame, the next step is to take a representative sample from this list.
How you should not select a sample
In many studies, samples of individuals are chosen by non-probability sampling:
Haphazard
Individuals are selected without any plan, e.g. the first 50 people you meet, selecting houses with no prearranged plan as you walk through a village, or the patients who attend the investigator’s clinic.
Quota sampling
In quota sampling, a pre-determined number of individuals are selected from specific sub-groups of the population (e.g. sex or age groups), from among those easily available.
Judgmental (purposive) sampling
Individuals are selected by the investigator to give what he/she considers to be a 'balanced' or 'representative' sample.
Snowball sampling
The first interviewee - who may or may not be sampled at random - refers to other individuals, who in turn refer to more individuals and so on. While snowball sampling is a form of non-probability sampling and therefore may introduce bias to your sample, there may be situations where it may be useful. For example, if you were looking at a "hidden" population such as drug users or sex workers, it may not be possible to draw up a sampling frame, and so you may have to rely on referrals from one or more individuals who you know or can locate.
All of these non-probability sampling methods have in common that the probability by which sampling units are selected cannot be determined and may be zero. They can be criticised because they introduce the possibility of selection bias: some people may be more likely than others to get into the sample, and so the sample becomes unrepresentative of the population.
Activity
List some reasons why this might occur.
Selection bias may occur due to a number of reasons when using non-probability sampling methods. Some of these are listed below:
- Interviewers may (consciously or unconsciously) avoid unfriendly looking people or poorer housing areas, or they may allow personal prejudices decide who will be ‘useful’ for the study or will give the result they are hoping to find.
- If all sample members are selected from a particular location or service, they may differ in some way from the rest of the population. For example, people attending a health clinic may be different to the people who do not attend the clinic.
- Snowball sampling will bias the sample towards those with wider social networks and who are more ‘visible’ in the population.