Research method
4.5 Analytical procedures
4.5.2 Within-case description
When analysing data from the embedded cases, the first task is to describe each to demonstrate understanding of them before they can be compared and contrasted with each other (Eisenhardt, 1989a; Patton, 2002; Ragin, 1994). Information gathered from documents, websites and other sources about NZAS – North, NZAS – Central and NZAS – South Island was used to develop background information for each case; this also included network maps, drawn using PAJEK from data entered into UCINET 6 software, which show the various stages of network development (this method is explained later under Use of quantitative techniques). The descriptive stage of developing background data on each NZAS network was audited for accuracy by feeding each embedded case back to each of the CEOs of the NZAS incorporated societies to confirm accuracy and further clarify understanding (Denzin, 1994; Miles & Huberman, 1994; Patton, 2002; Richards, 2005).
Findings from the qualitative data were then added to the within-case descriptions to build a descriptive within-case summary of each network. Quotes from participants were added after this to illustrate important points. Quantitative data that captured the structural aspects of the networks and measured the strength of relational dimensions was also included. Integrating qualitative and quantitative data gives a richer understanding of the cases, and
follows the recommendations of Coviello (2005), Miles (1979), and Miles and Huberman (1994).
Each within-case description was conducted separately across multiple levels – CEO/Board, work-unit and individual – by the construct areas identified from prior network studies, as suggested by Brown and Eisenhardt (1997). These construct areas were role of central broker, informal coordination mechanisms, forms of networks and context. The within-case description formed the first level of analysis of the data and was reviewed as part of the supervision process.
The within-case descriptions meant emerging patterns could be more easily identified, and the descriptions also built familiarity with each case prior to the cross-case comparisons (Brown & Eisenhardt, 1997; Eisenhardt, 1989a, 1989b; Richards, 2005).
Management of data
For the purpose of qualitative data management, QSR N6 (formerly known as NUDIST) software was used. Each transcribed interview was allocated an individual name that allowed for easy identification during the analysis. The interviews were entered into QSR N6, coded by embedded-case type, then by level of analysis, and finally by sentences and paragraphs against the construct areas (see Figure 4.3). This process allowed reports to be run off by level of analysis for each embedded case, and independently by construct area. The reports were then analysed for their embedded meaning.
The list of codes was first developed as part of a gradual process from the construct areas identified from prior network studies (see Chapter Two), and then developed further to include ‘commitment’ and ‘position and power’. This is why the initial coding diagram (see Figure B.1, Appendix B) is different from its final development, shown in Figure 4.3.
The coding scheme was amended and developed further once all interviews were completed and analysis had begun – hence ‘other’ being added as a coding category. Included in ‘other’ were emerging sub-categories from the data consisting of perceptions of
the network, perceived disadvantages of belonging, the role of NZAS, competition and social capital. These categories are clearly depicted in Figure 4.3. These categories emerged as a result of examining the data, with each consecutive interview informing and developing these categories. This illustrates the inductive process of the research, i.e. when being interviewed the research participants described what is meaningful and important to them (Miles, 1979; Patton, 2002; Ragin, 1994; Strauss & Corbin, 1994).
Use of quantitative techniques
To enrich the understanding of the qualitative data and provide triangulation to strengthen findings, two software packages were used: SPSS for Windows, and UCINET 6. SPSS was used to analyse data gathered from completed questionnaires (see Table B.5, Appendix B) for strength of relational aspects between organisations and the measurement of cross-level pressures within each embedded network. UCINET 6 was used to analyse the structure of the networks and also to analyse the data gathered from the questionnaires for strength of relational aspects; this follows the methodology of Borgatti, Everett, and Freeman (2002).
For this part of the study 31 research participants were recruited from multiple levels from 18 organisations across the three embedded, intentionally formed networks. All were invited to complete questionnaires on organisations in their respective network. Questionnaires were completed by research participants on organisations that they had knowledge of. A total of 144 questionnaires were completed by the research participants: 44 questionnaires were completed by 11 participants from NZAS – North network from all five organisations; 35 questionnaires were completed by nine participants from NZAS – Central network from four organisations, and 65 questionnaires by 11 participants from NZAS – South Island network from nine organisations. There were no invalid questionnaires. The statistical tests are explained in the next section.
SPSS for Windows software
Due to the small sample size and the small number of questionnaires gathered for the quantitative part of this research, the following statistical tests were possible using SPSS for Windows: one-sample t-test on the means of all questions for the three networks, Levene test for homogeneity of variance between the three networks, and cross-tabulation analysis on the levels within each of the networks across each variable. There was insufficient data to run a factor analysis and multidimensional scaling technique.
It was not possible to gather data on all actors in each network due to the constraints of time for the study, the nature of the emerging data-collection method, and the lack of knowledge of other actors by research participants in the network.
Simple descriptive statistics showing frequency of responses for each question were not presented because more detailed information is produced by the one-sample t-test on the means of these questions. The procedure for the one-sample t-test is discussed next.
For each within-case description for NZAS – North, NZAS – Central and NZAS – South Island networks a one-sample t-test procedure was used to test if there was a significant variance from the midpoint of 5. The midpoint represents the average on the scale of 0 (worst/less) to 10 (best/highest) for each of the means of the variables listed in the questionnaire, with the null hypothesis being that there is no significant variance from the midpoint of 5. Significance was taken at the 0.10% level (90% confidence level) rather than the 0.05% level (95% confidence level) because the purpose of this study is exploratory rather than confirmatory. (A lower confidence level means a smaller difference may be detected as the sample size is small.) The purpose of calculating t-tests results was to provide a triangulation for qualitative data and to measure the strength of relational aspects. Results are presented in Chapter Five as part of each within-case description. More detailed reporting on this procedure along with the full results is presented in Appendix F.
A cross-tabulation using SPSS was carried out for each level (CEO/Board, work-unit and individual) in each network for each variable in the questionnaire to discover if there was any variance in the means between the three levels. The reason behind this analysis was to investigate cross-level pressures within each NZAS network. Results are presented in Chapter Five as part of each within-case description. More detailed reporting on this procedure, along with the full results, are presented in Appendix F.
UCINET 6 software
UCINET 6 software (Borgatti et al., 2002) was used to analyse the completed questionnaires from the research respondents and to analyse network structural aspects. Originally it was anticipated that UCINET 6 software alone could be used to analyse the questionnaires but there was insufficient data. To use UCINET 6 software for this purpose requires all research participants to rate all actors and that there be no missing data. Given the limited number of research participants, these criteria could not be met. Instead a
limited UCINET 6 analysis was conducted by entering data into this software from the actors selected from each of the three networks. Research participants rated other actors in their network on each of the eleven variables shown in Table B.5 (see Appendix B). Data for the question “How much does belonging to the network help you with your business?” was calculated manually and this is explained in Appendix H. Where data was collected from more than one research participant residing in the same actor (this may be because an actor operates at more than one level within the network), an average rating was calculated and used for that actor.
UCINET 6 was used to compute similarities and then cluster those similarities at two different levels, namely the network level and actor level. The aim of this analysis was first to determine which relationships the research participants within a network viewed in the same manner at the network level and, second, to ascertain which actors viewed other members of the network in a similar way in terms of these relationships. This analysis could indicate, for example, that if all but one member views the others as strongly committed it may be that the network is not working as well for the member with the divergent viewpoint. Similarly, when considering the network as a whole, some relationships may be seen as effective (narrow range, high ratings) or ineffective (narrow range, low-to-moderate ratings), whereas relationships with a wider range of ratings across the participating organisations may work well for some but not for others – indicating problem areas requiring attention by these organisations. A fuller explanation of this analysis along with the results is presented in Appendix H.
UCINET 6 uses mathematics in the form of graph theory or sociograms to analyse the structure of each of the networks at their various stages of development (Hanneman, 2001; Iacobucci, 1995; Wasserman & Faust, 1995). Therefore, this software was chosen to analyse the structural aspects of each network in this study. The analysis consisted of ego- network density, structural holes and brokerage. Again, the results from these calculations are presented in the within-case descriptions. An explanation of the terms used in the UCINET 6 calculations is presented in Appendix G and the methods for calculating network structural aspects follow the instructions of Hanneman (2001).
To enable the analysis, data was gathered from network maps drawn by the research participants (described in more detail below in Cognitive-mapping technique). From this data a matrix was constructed using binary data (0 and 1) to represent directionality of ties: if there is no tie, then 0 is used, but if there is a tie, it is represented by 1; if a tie is reciprocated, then 1 is used in the corresponding actor box, whereas if a tie is not reciprocated, then 0 is used. (See Table 4.2. for an example of this type of binary matrix.)
Table 4.2: Matrix second stage of network for NZAS – North
Actor
WINTEC MISH AUT UA
WINTEC 0 0 1 0
MISH 0 0 1 1
AUT 1 1 0 0
Actor
UA 0 1 0 0
WINTEC = Waikato Institute of Technology MISH = Millennium Institute of Sport and Health AUT = AUT University
UA = University of Auckland
Cognitive-mapping technique
The cognitive-mapping technique was used to define the network structure over time, and the patterns of connections. This information was collated from each participant within each separate network and used to produce network maps or pictures using PAJEK software (see Figures 5.2 to 5.4, Chapter Five). Network pictures are the backbone to understanding interactions within networks and are a central concept for network management (Henneberg & Mouzas, 2006). The research participants were invited to draw the network by identifying the network organisations and then drawing the connecting lines between each organisation at the various stages of the network’s development. Development stages were identified by the research participants.
A matrix of these connections for each network using this technique was also produced. A list of actors identified by participants was used to form each axis. Where a connection exists with another actor, a ‘1’ was used to indicate this; a ‘0’ indicated no connection. This enabled analysis of density (awareness of connections between actors) at the three levels of each of the three networks. (See Table 4.3 for an example of the matrix which can be
derived from cognitive mapping. This table reveals NZAS – North network is dense because all research participants at all levels know of all other actors in the network.) All results for cognitive mapping of density are presented in Appendix I.
Table 4.3: Cognitive mapping of NZAS – North network to show density by level
Known actors in network
Level Actor NZAS – N MISH UA WINTEC AUT
NZAS – N 1 1 1 1 1 MISH 1 1 1 1 1 UA 1 1 1 1 1 WINTEC 1 1 1 1 1 CEO/Board AUT 1 1 1 1 1 Work-unit AUT 1 1 1 1 1 NZAS – N 1 1 1 1 1 MISH 1 1 1 1 1 UA 1 1 1 1 1 WINTEC 1 1 1 1 1 Individual AUT 1 1 1 1 1 Key 1 = know about 0 = don’t know about
NZAS – N = NZAS – North Inc.
MISH = Millennium Institute of Sport and Health UA = University of Auckland UniSports Centre WINTEC = Waikato Institute of Technology AUT = Auckland University of Technology