Whether sampling is necessary or not depends on the research question(s) and objectives. In some cases a census is possible where data from the entire population is collected and analysed. If a sample represents the entire
population a census would not necessarily provide more useful results than the sample (Saunders et al 2009).
There are practial considerations that cause the researcher to work with a sample instead of the entire population: Budget and time constraints and other practical aspects prevent the researcher from conducting a census but
examining a sample only (Sekaran 2003, Saunders et al 2009).
A sample is a subset of the target population (Thietart et al 1999, Remeyi et al 2000, Sekaran 2003). If the sample is representative of the entire population the conclusions drawn from it should be generalisable (Sekaran 2003). However, a sample will never be exactly the same as the entire population. Another sample generated in exactly the same way would be different from the first one. To which extent a sample is regarded to be representative in the end is a subjective judgment of the researcher (Remenyi et al 2000). According to Thietart et al (1999) the validity of a study is influenced by three characteristics of the sample: (1) whether the sample elements are of a heterogeneous or a homogeneous nature; (2) by which method the sample was generated and; (3) the sample size compared to the whole population.
Sampling techniques can be divided into two major types: Probability and non
probability sampling (Thietart et al 1999, Remenyi et al 2000, Sekaran 2003, Saunders et al 2009).
Probability sampling relates to positivistic research (Remenyi et al 2000) and survey and experimental research strategies (Saunders et al 2009). In probability sampling it is assumed that each sample has the same known probability to be randomly selected from the population (Sekaran 2003,
Saunders et al 2009). Samples can be analysed by applying statistical methods (Remenyi et al 1999). The results are considered to be generalisable for the entire population (Saunders et al 2009). Common types of probability sampling are simple random sampling, systematic sampling, stratified sampling, cluster sampling and multi-stage sampling (Thietart et al 1999, Remeny et al 2000, Saunders et al 2009).
Non-probability sampling methods are the domain of phenomenologist
researchers (Remenyi et al 1999). They are based on subjective or judgmental selection from the population (Remenyi et al 1999, Saunders et al 2009).
According to Remenyi et al (1999) non-probability samples are primarily relevant in exploratory research. The following techniques are commonly used to generate non-probability samples.
Quota sampling is usually used for interview surveys. It is assumed that the sample is representative of the population as the variability in the sample for
various quota variables equals that in the population. To select a quota sample the population is divided into specific groups. For each group a quota is
calculated based on relevant data. From each quota a defined number of cases must be researched, i.e. data must be collected. The collected data then is combined to provide the full sample (Saunders et al 2009).
Purposive sampling allows the researcher to use own judgment to select cases considered to best help answer the research question(s) (Saunders et al 2009).
A purposive sample is not aimed to be statistically representative of the population (Remenyi et al 1999). The purpose of the sample here is to e.g.
catch best practice experiences by selecting individuals who are considered to provide the desired information (Remenyi et al 1999, Sekaran 2003).
Snowball sampling is mainly used when the researcher has difficulties to identify members of the desired population or to get access to them (Remenyi et al 1999, Saunders et al 2009). Here the researcher makes initial contact to the first case(s) which then indentify(ies) further cases. Bias is considered to be huge as the sample will normally be homogenous due to respondents most likely identifying other potential respondents who are similar to themselves (Saunders et al 2009).
Self-selection sampling occurs when each subject can identify their desire to participate in the research. Therefore the demand for participants is usually published in appropriate media or the potential subjects are asked to take part (Saunders et al 2009).
Convenience sampling refers to the selection of subjects that are most conveniently available to participate in the research (Remenyi et al 1999, Sekaran 2003). The bias problem occurring here is justifiable when the population is rather homogenous with little variation (Saunders et al 2009).
In this study the possibility of conducting a census was excluded from the beginning. Currently the International Air Transport Association represents some 240 airlines located all over the world (International Air Transport Association 2012a). These are mostly international carriers not counting the various regional and small airlines. Not only budget and time constraints would
have made it impossible to include all airlines of the world in this research but also the researcher’s own experience that getting access to airlines without personal contacts is nearly impossible.
XYZ Cargo approached the researcher and therefore no sampling was
conducted with regard to the client company. However, the researcher felt that additional participating companies would be helpful. First, with the researcher having started the research and then including companies, this would have helped in terms of credibility of the Action Research strategy as this sequence would support Schein’s (1995) view that this would clearly distinguish Action
Research from consultancy.
Therefore the researcher decided on sampling. In theory a probability sample was possible but knowing that most airlines would refuse access and
participation in this study the researcher decided for a non-probability sampling method. Snowball and self-selection sampling did not provide useful cases.
Although some individuals were willing to contribute to the study they were not able to grant the required level of access.
In the end only XYZ Cargo was willing to participate in this study by allowing the researcher to take the position of a participant researcher who was also allowed to actively interfere in the company’s processes. Contact to this company had already been established before the start of this research project based on a previous working relationship. As in this case XYZ Cargo approached the researcher the selection was not done by sampling.
However, convenience sampling was applied for the additional field work when the researcher obtained additional feedback from aviation experts on the developed Operational Factor Approach during the field work at the client company. The sample of 25 airlines conveniently picked by the researcher can be regarded as being representative. Airlines with different fleet sizes (between 1 and 70 aircraft), different fleet types (jet aircraft, turboprop aircraft and mixed fleet), and different business experience (between 3 and 40 years) are included.
Some of the airlines are very similar to the client company, others are very different.