Chapter Five: Methodology
5.2. Philosophical Considerations of Research
5.3.4. Sampling Techniques
Questionnaire surveys usually involve only a proportion (or a sample) of the population being studied, because it is often not feasible, or costly to collect data from every single potential representative of a population. Therefore, it is of a major importance to ensure that the sample characteristics match with those of the population of interest (May, 1993; Veal, 1992; Nachmias and Nachmias, 1992). The sample size, however, is subject to numerous debates. Veal (1992:209), for example, states that “the absolute size of the sample which is important, not its size relative to the population”. May (1993), on the other hand, argues that a large, but poor in quality sample will be less accurate than a smaller one that does have quality. In qualitative
research, where the main focus is on exploring a situation or an issue, Maykut and Morehouse (1994) and Kumar (2005) suggest continuing gathering and analysing information until a “saturation point” is reached - when the newly collected information repeats already collected data. This strategy was also found in Lincoln and Guba’s (1985: 234) work that data should be collected “until redundancy with respect to information” is achieved. Therefore, a carefully conducted study adopting this strategy might reach a saturation point using a small, but precisely selected sample of the population being studied (Maykut and Morehouse, 1994). The sample size of both random and non-random samples can also be affected by the amount of time, money and human resources available (Jennings, 2010).
Several types of samples exist in the literature, but samples can be classified either as probability samples or as non-probability samples (May, 1993;
Nachmias and Nachmias, 1992).
5.3.4.1. Probability Samples
Only probability samples can be used to achieve statistical generalization (May, 1993; Maykut and Morehouse, 1992), representativeness of the results and minimised bias (Veal, 1992). Within probability samples, each member of the population of interest has an equal chance for inclusion in the sample (Veal, 1992, Maykut and Morehouse, 1992, Jennings, 2010). It is vital for a random sample to be based on a complete (or as complete as possible) list of the population called a sampling frame (May, 1993; Nachmias and Nachmias, 1992).
There are four subcategories of probability samples: simple random sampling, systematic sampling, stratified sampling and cluster sampling (Pizam, 1994). In simple random sampling every unit has the same chance of being included in the sample (Jennings, 2010). In systematic selection, however, the researcher sets a specified interval throughout the sample and uses it to select units for inclusion into the sample; therefore, the selection of one unit depends on the previous selected unit (Jennings, 2010). Stratified sampling is another subcategory of random samples where the population is
divided “into mutually exclusive and exhaustive subsets” (Pizam, 1994: 102) and a random sample of units is selected from each strata. Cluster sampling is a method used to split the whole population of interest into clusters from which a random sample is chosen (Kumar, 2005).
5.3.4.2. Non-probability Samples
In practice, a list of the population of interest or some sort of sampling frame hardly ever exists (May, 1993; Veal, 1992) and researchers, in particular social researchers (Nachmias and Nachmias, 1992) must make use of a non-probability sample. In other cases, the statistical precision of random sampling techniques is less important than the criterion of “fit for purpose”
(May, 1993) and of achieving understanding of social phenomenon in depth, not in breadth, by carefully selected group of individuals (Maykut and Morehouse, 1992). This approach is in harmony with the interpretivist postulates about the “multiple” realities consisting of both tangible and intangible elements that resulted from individuals’ minds, past, present and future and in disharmony with the positivist position of generalizability.
Moreover, the researcher leaves his/her mark on the criteria for choosing a sample making its representativeness a subject of subjectivity questioning and showing his/her position as “insider” of the analysed phenomena (Maykut and Morehouse, 1994). There are a number of non-random sampling approaches: convenience, snowball, expert, quota and purposive sampling (Sarantakos, 2005).
Convenience sampling is described as a selection process of participants based on the ease of access the researcher has to them. This sample does not represent the population from which it is drawn and only reflects those study units convenient to the researcher at the time of data collection;
therefore, such sampling is described as incapable to reflect other time periods (Jennings, 2010).
Maykut and Morehouse (1994) define purposive sampling (also called judgemental sampling) as suitable for qualitative research because it
represents a selection of participants based on the possibility that each one of them will expand the variability of the sample. Judgemental sampling also relies on researcher knowledge and judgement on who or what study units to include in the study (Jennings, 2010). Expert sampling, on the other hand, involves people who the researcher identifies as “experts” with specific knowledge and experience (Jennings, 2010).
When a population is characterised a wide distribution, snowball sampling might be the only one possible way of collecting data about the population, where an initial contact is made with a representative of the population who then connects the researcher with other members of the population (Jennings, 2010).
A form of non-random sampling often used in street surveys is that of quota sampling. Here the general characteristics of a population are often known beforehand - the proportion of people in particular age groups, social classes, etc. and the sample consists of quotas of participant having these characteristics (Veal, 1993; May, 1992; Jennings, 2010). However, once the quotas have been determined and calculated the selection process is by convenience. Stratified sampling unlike the quota sampling divides the population into quotas by random, whereas the quota sampling specifies the number of sample units in each quota then follows up with convenience sampling (Jennings, 2010).