The two main types of sampling method - probability methods and non-probability methods - are shown in Figure 4.3.
Statistically speaking, these are the best types of sampling method as each respondent has a known chance of being selected, so bias is minimized. They also allow the accuracy of the results to be estimated statistically. Sometimes probability sampling methods are referred to generically as 'random sampling' methods. In fact, this refers to a specific type of very precise probability sample. There is often some confusion over the use of the term 'random'. Selecting people in the street at random is not technically random sampling, but more often refers to selection of respondents by interviewers for quota sampling.
The main types of probability sample are simple random sampling, systematic random sampling, stratified random sampling and cluster sampling.
Simple random sampling Items can be selected from the sampling frame by using the lottery method, e.g. taking numbers out of a hat. In the UK, ERNIE the computer selects Premium Bond winners, and does so by using simple random sampling. Random number tables are generated by computer and often used in marketing research.
Systematic random sampling With larger samples it is more convenient to divide the population by the sample size to calculate the sampling interval (n). A random starting point is selected using random number tables and every nth time after that is selected.
Example If the sample size is 50, and the population size is 3000, then the sampling interval is calculated as:
If the random number picked from the tables was 35, for example, a then the first item selected from the sampling frame would be 35. i
Every 60th number after that would be selected until a sample size of 3
50 was achieved. a
As the first number was selected randomly, this method is some- tt times called a 'quasi-random' method. J
The advantage of these methods is that they are relatively simple e to carry out and sampling error and confidence levels can be calcu- $
lated statistically. The main disadvantage is that samples may be pro- c duced that do not reflect the characteristics of the survey population. For example, if a sample of students were drawn from a list of all students at a university, it is possible that all the students in the sample might be design students. This is clearly not representative of the student population as a whole.
Stratified random sampling One way to try to overcome this type of sampling error is to use stratified random sampling. This is used when it is felt that different groups within the population have characteristics that are likely to lead to different types of answers. The population is divided into distinguishable groups (strata) who have similar characteristics. Stratification factors should be as relevant as possible to the survey (e.g. consumer surveys are often stratified by age, gender, socio-economic group, and so on). A random sample is then taken from each stratum.
There are two main methods used to stratify samples. First, with a uniform sampling fraction (proportionate sampling), or secondly, with a variable sampling fraction (disproportionate sampling).
Proportionate and disproportionate sampling If all the strata are equally important to the survey, a proportionate sample would be taken, i.e. the same number selected from each stratum. Frequently, some strata are more important to the research than others. For example, if you were conducting a survey into the purchase of outsize garments (size 18+), it would be reasonable to assume that most of these items would be purchased by those who were larger than size 16 rather than those who were not. It is logical that more of these people should be included in the sample. In other words, a disproportionate sample would be taken. If a proportionate sample were taken, too few of the people who took larger sized clothes would be included in the survey and it would be difficult to extrapolate the results to the general population with any degree of accuracy.
Cluster sampling Cluster sampling is a variation of stratified random sampling and may be used when the survey population is concentrated
in a relatively small number of groups (clusters) that are considered typical of the market in question. A random sample of these clusters is then taken. A random sample of units from within these clusters is then taken. If the number of units within a cluster is small, a census may be carried out. In a national survey of specialist bridal wear retailers, for example, sales areas could be identified by geographical region and a random sample of these taken. Within each selected sales area, all or a sample of the store managers would be interviewed.
There is a problem with cluster sampling that occurs if the clusters are not sufficiently representative of the survey population. For example, in a small geographical area, it is likely that it will consist of people with similar housing, incomes and lifestyle. Although cluster sampling can be more cost-effective than some other methods of probability sampling, there is a danger that sampling error will increase if the clusters are not carefully defined before the first stage of sampling.
Sampling frames When using probability sampling methods it is necessary to use a sampling frame. This is a list of every element in the survey population. The sample is drawn from this list. A sampling frame is essential for probability-based techniques, as each element must have a known chance of selection, and so must be included in the sampling frame. According to Webb (1999), a sampling frame must have the following characteristics:
♦ Each element should be included only once.
♦ No element should be excluded.
♦ The frame should cover the whole of the population.
♦ The information used to construct the frame should be up-to-date and accurate.
♦ The frame should be convenient to use.
Examples of sampling frames include electoral rolls, the telephone book, the Royal Mail's lists of postcodes and other similar databases.
In practice, most sampling frames are not perfect. Not everyone with a telephone is in the phone book, for example. Finding a sampling frame that is suitable for your research can occasionally prove difficult.
With non-probability sampling methods, some element of judgement enters the selection process. The extent to which judgement is used, and therefore the element of bias introduced, varies in these methods. Non-probability methods do not require a sampling frame and the chance of each unit being selected is unknown. Statistical estimates of a the size of the sampling error cannot therefore be made. i
The methods are convenience sampling, judgement sampling and 3
quota sampling. a i
Convenience sampling Items are selected that are close or easily g available. This is useful in the exploratory stage of research, giving e the researcher a 'feel' for the subject. Despite being very cheap and a quick to carry out, the level of error and bias with this method is likely h to be very high and so it should be used with caution.
Judgement sampling Items are selected by the researcher that are felt to be representative of the survey population. This method attempts to be more representative than convenience sampling. Experts also may be consulted for advice on which items are likely to be more appropriate for the survey. For example, in a survey of textile manufacturers, a staff specialist such as a product developer may provide useful advice on which manufacturers would be suitable for selection.
Quota sampling This is the most likely non-probability method to produce a representative sample as items selected are based on known characteristics of the population.
Example Assume that your survey population has the following characteristics:
If we wanted to interview 150 people who were representative of the above population in terms of the two quota controls (age and gender), we would calculate the quotas as shown in Table 4.1.
This is more conveniently represented as shown in Table 4.2.
A survey's accuracy of representation can be increased by narrowing the bands and including more characteristics, e.g. social class. Interviewers are then allocated a number of interviews (quotas) with specific types of respondent.
The advantages of quota sampling are that it is relatively quick to carry out and easy to administer from a fieldwork point of view. It is also cheaper to use than probability sampling methods. The disadvantages of quota sampling involve problems of bias and sampling errors. The responsibility for selection of respondents lies with the interviewer, which may introduce bias. There is the added problem that there is no probability mechanism with quota sampling, so the sampling error cannot easily be calculated.
Table 4.1 Quota sampling frame (A)
16-29 = 26% 30-64 = 58% 65+ = 16% Female = 52% Female = 52% Female = 52%
Quota = 20
Quota = 45
Quota = 12
Table 4.2 Quota sampling frame (B)
19 42 12 73
20 45 12 77
Quota samples are often used in surveys where fine degrees of accuracy are not required, for instance in product testing for preference between products.
Although many companies who provide continuous research services use probability sampling, the majority of ad hoc marketing research is conducted using quota samples. If this method gave consistently biased or misleading conclusions, it would not be used.
Was this article helpful?
You won’t want to miss this. Get Paid Taking Surveys In Your Spare Time. I know what you’re thinking, ‘Oh great, another get rich quick scheme’. WRONG! This is your guide to making money from home by participating in paid surveys on the internet.