Chapter 3 Theoretical Research Framework and Hypothesis Development
4.3 Data Collection and Preparation
4.3.1 Sampling
The population for this research was wineries in South Africa and New Zealand. As earlier noted, wineries either have contracts with grape producers, produce grapes for their own use or buy grapes on the spot market. They are therefore key decision makers in the governance form, and hence were the study‟s unit of analysis.
According to SAWIS (2009), there were 585 wineries in South Africa as at the end of 2008. SAWIS is an organisation that that collects, processes and disseminates South African Wine Industry information (see http://www.sawis.co.za/). SAWIS provided the contact details of the wineries and as per their industry policy, offered to help coordinate the research. Questionnaires were sent to all the wineries in South Africa. A total of 111 responses were usable, representing an 18.9% response rate. There were 683 wineries in New Zealand in 2009 (NZWINE, 2009). However, NZWINE (2009) could provide contacts of only 580 wineries (85%). Hence 580 questionnaires were sent out in New Zealand. A total of 116 responses were usable, representing a 20.0 percent response rate. Further details of the responses are discussed in Chapter five.
4.3.2 Data Collection Procedures
A number of methods for collecting the data for this study were considered. These methods included self administered questionnaires, personal and telephone interviews, field studies, internet based surveys as well as an assortment of library based approaches (Zikmund, 2003). All these methods have weaknesses and strengths. The choice of each therefore depends on a number of considerations regarding their strengths and weaknesses. For this research, major
consideration was given to the costs of collecting data from South Africa and New Zealand. Each country has a number of regions that had to be covered and this added to the costs. Thus, the mail survey was seen as the most appropriate data collection method for this research because it allowed for the coverage of both countries at reasonable costs. It also allowed for respondent flexibility since respondents could fill the questionnaire as and when they had time (Zikmund, 2003). Further, it allowed respondents enough time to think and reflect on their responses.
The design and implementation of the mail survey followed Dillman‟s (2007) mail survey method. This method puts emphasis on improving the response rate and improving data quality. This is achieved through among others, the cover letter, a detailed yet short and to the point questionnaire, pre-paid reply envelope and a post card reminder two weeks after the initial mailing. All these steps were followed, except that instead of a post card, a reminder letter was sent with a new questionnaire and a free post envelope three weeks after the initial mailing. In addition, respondents were given the choice of whether or not to be sent the summary of the results once the research is completed.
4.3.3 Non-Response Bias
Despite following best practice with the intention of increasing response rates, some wineries did not return the questionnaire. Such potential respondents may have deliberately declined the questionnaires due to, among others, proprietary nature of the data or they simply did not have time to answer the questionnaire. This could potentially result in non-response bias, a statistical difference between a survey that includes only those who responded and an ideal survey that would also include those who failed to respond (Zikmund, 2003). This difference may be due to the fact that subjects that do not return the questionnaire or answer certain questions may have opinions that are substantially different from those who returned the questionnaire or answered all questions.
Results drawn from data with non-response bias do not allow the researcher to say with confidence how the entire sample would have responded (Armstrong & Overton, 1977). Li (2005) states that a standard way to test for non-response bias is to compare the responses of those who return the questionnaires after the first mailings with those who return it after the second and subsequent mailings. The assumption is that those who do not return the questionnaires after the first mailing have almost similar opinions with those who do not respond at all. For this research, non-response bias was tested through the independent
samples t-tests to compare means from first mailing responses with means of the after reminder responses (Pallant, 2007). Results are presented in Chapter five and they show no significant evidence of non-response bias.
4.3.4 Data Coding, Entry and Cleaning
The questionnaire was pre-coded through the likert scale format. Each question had a coding range that guided data entry or transfer into the electronic format. All data were handled through Statistical Package for Social Sciences (SPSS) for windows (version 17). Data cleaning was then undertaken to ensure that there are no data entry errors. This included checking for scores that are out of range. This included checking the minimum and maximum scores as well as the mean to ensure that they fall within the coding framework (Pallant, 2007). It also included the inspection of the box plots to ensure that there are no outliers. Any values out of line with expectation were corrected by going back to the questionnaire to re- enter the correct values. This phase also took care of missing data. Descriptive data was generated and checked for the proportion of missing data. Only a few data points were missing. When only a few data is missing, Pallant (2007), recommends excluding the missing cases through SPSS‟s pairwise exclusion of missing data option. Consideration was given to replacing the data with the mean through SPSS “replace with mean option”. This option was not taken because Pallant (2007) cautions researchers against it. She argues that it should “never” be used as it can severely distort the results of the data analysis. This resulted in excluding three and five data points for South Africa and New Zealand, which yielded the final sizes of 111 and 116 for the two countries respectively.