Phase IV: Selecting the Sample

One of the tasks in survey research is to identify the research population. The population represents the total units from which the sample is selected. The sample frame is a list of all the units from the population from which the sample is chosen. The sample is a segment of the population chosen for investigation, and is often derived from the sampling frame as it is not always possible or practical to involve the entire population. For a sample to be representative, it must exhibit aggregate characteristics that closely approximate those in the population (Babbie 2008). To obtain a sample for this study the following issues were considered and procedures undertaken: defining the population and sample frame; selecting the sample; determining the sample size; implication of sampling errors and bias; and non- response bias.

The first research task was the identification of the population. The population that is relevant to this study is all superannuation defined contribution fund members. The sample frame chosen for this study is Unisuper defined contribution fund members. Under the current university-wide industry fund, university staff can belong to a defined benefit superannuation plan or a defined contribution fund plan. They also have the choice to move to other superannuation plans outside the university sector fund. Statistics derived from the university sector superannuation fund reveal that approximately 25 percent of members belong to the defined contribution fund and around 75 percent have a defined benefit plan.

4.5.4.2 Selecting the Sample

Non-probability and probability sampling are the two broad types of sampling methods referred to in the literature. Probability samples are considered more representative when developing sample frames. That is, probability sampling ensures that all the members of a population have an equal chance of being included in the sample (DeVaus 2002). A non-probability sample is defined as “any technique in which samples are selected in some way not suggested by probability theory” (Babbie 2008, p. 203).

In this study a non-probability sampling approach was adopted to meet the requirements of the research. A non-probability sampling approach was adopted as it was impractical to generate a probability sample. To generate a random sample the co-operation of the superannuation fund management is required and several attempts to gain their assistance were not successful. Efforts made to gain the fund manager’s co-operation commenced with an initial telephone call to the Chief Executive Officer’s (CEO’s) office followed by a detailed email explaining the purpose of the research. A telephone conversation with the CEO resulted from the email and a more detailed submission was made on the CEO’s recommendation. The submission included the draft survey instrument and a detailed summary of the study together

with another fund, a purposive (judgement) sampling technique which is considered a form of non-probability sampling, was adopted. This method uses judgement in obtaining a representative sample through the inclusion of typical groups in the sample (Kerlinger 1986), and proceeds on the basis of judgement.

The sample was chosen from 26 universities identified in Appendix Four and included a cross-section of university staff. Particular effort was made to ensure that these universities represented various geographical locations across Australia. More specifically, steps were taken to ensure that rural and city-based university campuses were included in the sample: this ensured geographical diversification. The sample was then derived from the 26 chosen universities. Systematically, different departments and faculties within each university were chosen to obtain a mix of academic and non-academic university staff in the sample. Particular effort was made to ensure the profiles of university staff included in the sample were representative. That is, the sample was chosen from academics and non-academics. The non- academics included staff from administration, library services, human resources, information technology, finance and accounting, and building and grounds (Refer to Appendix Four for more detailed information).

4.5.4.3 Sample Size

With non-probability samples no statistical inferences exist by estimating a sample size as it is not possible to determine the probability of any particular participant being selected for the sample (Ritter & Sue 2007). It is suggested that increases in sample size reduce the non-representativeness associated with the use of judgement in generating the sample (Kervin 1992). It is shown that sample sizes beyond 350 produce estimates of population mean and correlation that have less fluctuation and more stability (Kervin 1992). Perry (1995) argues that as a rough rule of thumb, doctoral quantitative research requires at least 350 respondents. Other guidelines that have been suggested for determining a suitable sample size for non-probability samples include samples of somewhere between 30 and 500 (Hill 1998) or 10% of the

at least 100 cases and minor sub-groups contain between 20 and 50.

It was decided to choose a sample of 6000 university staff to ensure that an adequate sample of accumulation fund members was reached. However, only 25 percent of the 6000 sample would, statistically, belong to the defined contribution fund. Accordingly it was expected that approximately 1500 (25% x 6000) of the sample would belong to the university sector defined contribution fund. That is, one of every four surveys sent would be received by a defined contribution fund member. From a potential sample of 1500 it was considered that a sufficient number of responses would be generated to produce appropriate quantitative analysis.

4.5.4.4 Implications for Sampling Errors and Bias

Sampling errors result from selecting a sample that is not representative of the population as a whole. In a probability sample, it is possible to calculate a sampling error. However, in non-probability sampling statistical means of estimating sampling errors cannot be used. It is said that statistical estimates are valid only for samples where every member of the population has an equal chance of selection (Sapsford 2007). It is argued that if the particulars of the sampling are made clear, then the reader is in a position to make a judgement as to the extent to which the statistics are relevant to the population (Sapsford 2007). This point is further reinforced by Schofield (1996, pp. 99-100) who suggests that:

In preventing standard errors in circumstances such as this, researchers are in effect saying: O.K, I know I haven’t got a random sample and so can’t estimate sampling error. But this is the best I could do. It could be the case that it hasn’t mattered very much, and thus I have calculated the standard errors and have used them in further tests. My finding has support from the literature and looks useful. It’s up to you, dear reader, to decide how much reliance you will place on it. Perhaps you’ll think that the result is important and will be able to replicate it without the sampling difficulties which I have had and have reported.

nevertheless deliver robust samples (Jensen & Jensen 2002). The purposive sampling method was adopted in this study to produce a judgement sample that is a fair (close) representation of the Unisuper membership within the defined contribution fund. This should be regarded as one limitation of the data where sampling error cannot be strictly (theoretically) calculated from a non-probability sample. Statistical inferences have been drawn from the analysis of the data and conclusions and recommendations follow. However, only tentative generalisations can be made from data collected from such a purposive sample.

4.5.4.5 Non-response Bias

The aim of survey research is to achieve an acceptable number of responses from the chosen sample. This is rarely achieved in either mail or web-based surveys. The problem then arises as to whether or not the responses received are in some way different from those where there was no response. Non-response arises as a result of the sample not being contactable, or simply refusing to take part in the survey (Sapsford 2007): it may also result from unusable questionnaires. Non-response is considered a problematic and important source of survey error (Fowler 2008). The existence of non-response means that it may not be possible to assume that the responses received are representative of the sample. To combat the impact of non- response bias researchers should endeavour to maximise the response rate (Fowler 2008). The techniques used to maximise response rate in this study have been discussed in Section 4.5.3.8.