Population, sampling design, and sample size

To solve the research problems, data have to be obtained from the right people or objects to provide correct answers (Cavana et al. 2000). Good samples should provide adequate representation and accuracy for decision-making. A well-defined procedure is proposed to achieve high representation and accuracy (Hair et al. 2003 p.209):

Defining the target population Choosing the sampling frame Selecting the sampling method

Determining the sample size Implementing the sampling plan

4.4.3.1 Target population and sampling frame

The target population is the complete group of objects or elements that the researcher wishes to investigate (Canava et al. 2000; Hair et al. 2003). The main research objective of this research is to apply knowledge management practices into the shipping industry in the context of China. Therefore all organisations in the shipping industry in China are the target population and each organisation is understood as an element of the population in this research.

A sampling frame is a comprehensive, accurate list of the elements from which the sample is drawn (Hair et al. 2003). To find a reliable list of shipping organisations in China, various different sources of information have been consulted. These sources include the World Shipping Directory of the Lloyd‟s Register, the China Shipowners Association, and the China Classification Society. It was found that the World Shipping Directory had the most comprehensive lists of ship operators and ship managers. In addition, the Directory was published in 2007 and is believed to contain

124

the latest information available when the mailing list of this survey was compiled. It is acknowledged that in reality, a list may not be necessarily up-to-date. It may include elements that do not belong to the target population or may exclude relevant elements. It may contain duplicate elements.

From the World Shipping Directory (Lloyd‟s Register 2007), 928 (SO11889-10977; SM1063-1077) shipping organisations were found. These were then considered the total target population for the survey from which the sample would be drawn for the final mail survey.

4.3.3.2 Sampling methods

Sampling design and sample size are important to establish the representativeness of the sample for generalisation. A large sample itself does not necessary produce findings that can be generalised in the population if the sampling design is inappropriate. Similarly, no sampling design can be useful in achieving research objectives if the sample size is inadequate (Cavana et al. 2000). The task of this section is therefore to use the correct sampling methods to determine the right sample size.

Probability and non-probability sampling are two traditional sampling methods. While in a no-probability sampling, the inclusion or exclusion of elements in a sample is determined by the research; in probability sampling, sampling elements are selected randomly and the probability of being selected is known. The main differences of these two methods are the representativeness of the sample and the ability to generalise the result. The probability method minimises the selection bias and is considered to be representative of the target population. In addition, findings based on a probability sample can be generalised to the target population with a specific level of confidence (Hair et al. 2003). Thus a probability sampling method is used in this thesis.

125

research. The sample size is determined by three main factors (Zikmund 2003), that is, a) the variance or heterogeneity of the population (standard deviation); b) the magnitude of error, or the confidence interval (range of random error); and c) the confidence level. The standard deviation can be obtained from a pilot study or as a rule of thumb the standard deviation is one-sixth of the range, while the magnitude of error E and confidence level Z are based on managerial judgement.

2

E ZS n

Where Z = standardized value corresponding to a confidence level

S = sample standard deviation or an estimate of the population standard deviation

E = acceptable magnitude of error, plus or minus an error factor

A majority of questions in the survey uses the five point Likert scale (for example, strongly disagree to strongly agree). As a rule of thumb discussed earlier, the value of the standard deviation for calculation is one-sixth of the range:

S = 6 1 x (5-1) = 3 2

A 95 per cent confidence level (Z) and a magnitude of error of 0.07 are used to calculate the sample size:

348 67 . 18 07 . 0 3 2 96 . 1 ₂ 2 X n

When referring to the influence of the population size on sample size, Zikmund (2003) argues that in most cases the size of population does not have a major effect on the sample size. However, correction is needed if the sample size is more than 5% of a finite population (Zikmund 2003) or when the total population is less than 10,000 (Saunders et al. 2000). As the target population of this study is 928, a sample size of 348 represents more than one third of the population, and therefore an adjustment is required. The adjusted sample size is calculated by the following formula (Zikmund 2003): ) 1 /( ) ( ' n N n N n

126

Where n’ is the adjusted sample size n is the initial sample size N is the population size

Thus: 275 79 . 0 348 ) 1 928 /( ) 348 928 ( 348 ' X n

This sample size is consistent with sample sizes for the different sizes of population at a 95% confidence level provided by Saunders et al. (2000) (see Appendix I).