A great deal of sociological research makes use of sampling. This is a technique aiming to reduce the number of respondents in a piece of research, whilst retaining - as accurately as possible - the characteristics of the whole group.
The purpose of taking a sample is to investigate features of the population in greater detail than could be done if the total population was used, and to draw inferences about this population. In addition, at the practical level, a sample is likely to be both cheaper and quicker to investigate.
All sampling will involve error and sociologists have developed sampling techniques in order to minimize this error. All methods of sampling make use of a sampling frame.
A sampling frame is the list of members of the total population of interest. From this list a sample to study can be drawn. For example, such a list may be an electoral register, if information about those with voting rights is sought, or the family practitioner committee lists if a health survey is projected, or vehicle registration lists, if car ownership or road transport is under study.
The random sample
For inferences about a population to be valid, the sample must be truly representative, the only way to ensure this is to take a Random sample. This involves using either random numbers or systematic sampling. Random numbers are used to ensure that every individual in a sampling frame has an equal chance of being selected as a member of the sample. Systematic sampling involves randomly selecting the first individual from the list, then subsequently individuals at every fixed interval, for example, every tenth person if a 10% sample is desired.
So the test is, does every person in the group have an equal chance of being selected?
Examples of random sampling include ERNIE, a telephone directory, out of a hat.
An example of systematic random sampling is Willmott and Young's sample of Bethnal Green families.
The technique depends on the mathematical probability that a number of members carefully selected from a larger group will be more or less representative of that group. Of course, the improbable can happen and the sample is unrepresentative of the target population. One way to increase the precision of sampling is through stratification.
Non-representative sampling also occurs in sociology. The logic of this approach is that a non-representative sample might present a more demanding test of a researcher's hypothesis. For example, Goldthorpe and Lockwood tested the embourgoisement thesis (the belief that the working class was becoming more like the middle class) with untypical affluent workers. They argued that if the hypothesis was not true among affluent workers, it was not likely to be true with any other workers. In other words they gave the hypothesis the best chance of being proved correct.
When the population to be studied is large and the sample relatively small it may be efficient to use stratified sampling. Random sampling assumes that the list of the population involved (sampling frame) has no particular order of characteristics, which could have any bearing on the investigation. But in social research, most sample frames are not of this type, but have a definite order. For example, classroom research - school classes are already stratified and research then often takes a sample from each point on the strata.
Stratified samples tend to have smaller sampling errors than random samples of the same size. This is because the sample is divided into several groups in proportion to their known prevalence in an attempt to construct a sample that is representative of the whole.
This is a sampling method in which a sample is selected by quotas from each defined portion of the population. The method does not fulfill the normal requirements of random sampling. It involves breaking down the parent populations into strata according to relevant features and calculating how many individuals to include in each of these categories to reflect the parent population structure. Thus so far this is the same as a stratified sample and randomness can still be achieved.
However, once the size of each of these groups is decided no attempt at randomness is made. Instead, the interviewers are instructed to achieve appropriate selections (quotas) to fulfill the requirements within each group.
Contacts are made until a quota is filled. Therefore, non-response cannot occur. The interviewer makes the final choice of sample. However the choice is limited by availability, and the diligence and honesty of the interviewer. The method is much used by market research and opinion pollsters.
This is a form of longitudinal study, but is usually of shorter duration and more focused. It involves questioning the same sample at regular intervals to observe trends of opinion. With this sort of sample, as opposed to cross sectional (one off) studies, change over time can be monitored. A disadvantage is that respondents are lost through death or lack of interest or moving and that those who remain may become atypical through the very experience of being panel members.
This is a method of sampling which selects from groups (clusters) already existing in the parent population rather than assembling a random sample. This tends to be quicker and cheaper, but may lead to a biased sample if the clusters are not representative of the parent population. For example, polls taken of attitudes to government policy may be carried out in selected areas of the country thought to be representative, but because of local political dynamics this may not be the case.
This is a method of selecting a sample by starting with a small selected group of respondents and asking them for further contacts. This is not therefore a random sample and no inferences about the characteristics of the parent population can be made from such a study. Its use is primarily in the collection of in depth qualitative data, perhaps on sensitive topics, where an obvious sampling frame does not exist and the best method of selection is through personal contacts. Such a method might be used in an investigation of sexual habits or bereavement experiences.
To work effectively a representative sample must be an accurate representative cross section.
If a sample is large enough and properly selected it will increase its representational accuracy. If it fails to do either of the above its results may be less accurate than an educated guess.
Bias develops in several ways:
First, through self-selection. This is where respondents select themselves and is particularly true of postal questionnaires, for example, if 100 respondents reply out of a sample of 200, is this 45% or 90% in favour of a particular action if 90 out of the 100 answer yes?
Second, there will be choice involved at three levels in the sample and all ca introduce bias. The choosing of the sample, the choosing of questions, and the choosing of significant responses.
Finally, there is the judgment of interviewers, especially in quota sampling.
Generally, sampling seeks to avoid the possibility of 'freaks' occurring and the larger the sample, the less likelihood there is of this happening. The greater the variety of characteristics in the population being measured, the larger and more carefully designed the sample needs to be. Ultimately, the operation of a sample survey comes down to a running battle against sources of bias - a battle, which is never won.