• No results found

4. Research design and methods

4.6. Data collection

As there is no archival data available, there is a need to collect primary data. Structured questionnaires and the key informant technique are determined to be suitable when the nature of variables in the theoretical model is considered.

4.6.1. The key informant technique and the number of informants

The key informant technique has been commonly used for collecting data in inter-organizational research. Using this technique, one or a few informants with expert knowledge about the phenomenon of interest are identified (Seidler, 1974). These informants are capable of describing critical factors related to the unit of analysis and are willing to communicate about them (Campbell, 1955; Phillips, 1981). Characteristics of the phenomenon described by informants must also exist independently of the informants (Heide & John, 1995). If informants provide information about themselves, such information does not exist independently of the informants. The research must acquire information from a representative sample of informants (Wathne, 2001).

In this study, critical constructs are related to (a) mode of governance, i.e., level of hierarchical and relational governance, (b) transaction hazard, i.e., level of specific investments, (c) firm power, i.e., type of relationship, (d) negotiation strategy, and (e) relationship performance. All of these phenomena are assumed to be independent ofthe informants. Researchers can choose informants based on their knowledge instead of their representativeness in a statistical sense (Svendsen, 2005).

Researchers can decide to (a) collect data from one or more informants from an individual organization and (b) collect data from one or both sides of the dyad. These two decisions attract a great deal of discussion (see, e.g., Bagozzi, Yi, & Phillips, 1991; Philips, 1981). The first decision is whether researchers should collect datafrom one or more informants from the same firm Philips (1981) recommends the use of multiple informants, because there is a low degree of convergence among informants representing the same unit. However, using single-informant

design has become the more dominant approach due toresource constraints and implacability.

Since researchers may have limited time and resources, it is not always possible to use multiple informants. In the single-informant design, data can be registered directly as a report of the informant. Investigation is kept at the structural level and incurs less cost (Seidler, 1974).

Further, it may be inapplicable to collect data from many informants from the same firm. Some firms may only “establish one person as the focal point for relations with a given supplier”

(Heide & John, 1990, p. 30) or customer.

The second decision is whether researchers should collect data from one side or both sides of the dyad. The choice depends on the degree of potential discrepancy between the peerceptions of each exchange partner on the variables in the model. It may be appropriate to collect data from both sides, because researchers can validate the data from one side, compared with those from the other side, to obtain a more accurate value. Many empirical studies adopt this approach (e.g., Anderson & Weitz, 1992; Haugland, 1988; Reve, 1980).

However, if such discrepancies are assumed to be slight, a single-side design is sufficient.

Collecting data from both sides is time consuming and requires a lot more resources (Kumar et al., 1993). Since this study had limited time and resources, collecting data from one side of the dyad was more appropriate. Moreover, if data is collected from both sides, the process of analysis also requires more time because there are many observations for the same phenomenon (Kumar, Stern, & Anderson, 1993); the data must be analyzed for convergences, and joint understanding should be reported. There may be interpretation ambiguity that does not occur with single-informant data collection. By using data from one side, the results can be directly reported, and analysis of data divergence is not required.

Another benefit of collecting data from one side ofthe dyad is that researchers can focus as many observations as possible. Although a multiple-informant design for each relationship is preferred due to its advantage of avoiding or reducing the risk of biased information (Phillips, 1981), collecting data from multiple informants is time consuming and would be likely to reduce the number of observations.

Literature in the field of inter-organization relationships concludes that it is justifiable to conduct a one-sided approach (Heide & John, 1994). Many authors of empirical studies with the same variable in the models argue that there is correspondence between measures of variables, such as perception of transaction characteristics (Heide & John, 1990; Dyer & Chu, 2003), structural form of the relationship (John & Reve, 1982; Reve, 1980), specific

investments and commitment in the relationships (Anderson & Weitz, 1992), and the perception of performance (Anderson & Narus, 1990). Therefore, there seems to be sufficient evidence that there is correspondence between buyerand seller perceptions of the variable in the model. On this basis, it is justifiable to sample from one side of the dyad.

In this study, data was collected from one side of the dyad. General managers in the relevant companies were contacted and asked if they had the knowledge and willingness to be informants. If the general managers were not ready to participate in the research, they were asked to make introductions to their marketing, sales, product, brand managers, or dedicated salespersons. (Since the unit of analysis in this study is the relationship, the choice of informant is the marketing or sales or product or brand managers, or salespersons with in-depth knowledge of the exchange.) Accordingly, the requirements of Campbell (1955), Phillips (1981), and John (1984) were satisfied. Marketing managers have deep understanding about the exchange, customers, and power asymmetry between their firms and their customers’ firms.

4.6.2. Data collection procedures

Data was collected in three phases. First, qualitative data was collected from specially selected supplier firms in the O & G industry. Second, informants from the relevant companies were identified. Third, a structured e-questionnaire wasprepared and attached to the invitation email.

The first phase consisted of becoming familiar withthe empirical setting and making contact with the relevant companies. It also included becoming familiar with the practical use and practical understanding of critical constructs and the hypothesized relationships between them.

In this phase, qualitative data was collected through interviews with three specially selected marketing managers in the supplier companies. Additionally, six academic experts were consulted regarding measurements and questionnaire.

In phase two, informants in the supplier companies were identified. The initial sampling frame was 602 contacts. Subsequently, however, during thetelephone invitation process, it was found that there were 43 duplications and 49 out-of-scope firms. Therefore, the updated sampling frame was 510 contacts. The companies in the sample range in size from small to very large;

and provide a wide variety of products and services that support the O & G industry.

E-questionnaires and reminders were distributed in the third phase. It was necessary to recruit a research assistant whose native language is Norwegian; a bachelor student was found to fill the position. The assistant was trained to approach key informants through telephone calls.

(The telephone call guidelines are presented in Appendix A.) In addition to having a native speaker contact the key informants, two further incentives were included to increase willingness to participate in the survey. First, prospective informants were promised that they would receive the results of the research. Second, every respondent had a chance to win an iPad.

During the telephone conversation, the research assistant asked the prospective respondents if they were knowledgeable, had time, and were interested in joining the survey. They were informed that in return, they might benefit from the research results, and they would have the chance to be the winner of the iPad.

The telephone recruitment began in February 2012 and ended by November of 2012. During two periods, the 17 May national holiday in Norway and the early-June to early-August summer holidays, it was difficult to reach prospective respondents, thereby slowing the process of data collection.

The research assistant made telephone calls to all 602 contacts in the initial sample. He found that there were 43 duplications, 49 out-of-scopes, 32 wrong numbers, 29 answering machines, 11 cases of language difficulties, 43 cases of lackof time, 26 cases of no interest in the survey, 4 cases of organizational constraints, and 16 cases in which the contact could not be reached by phone. As a result, 349 prospective respondents agreed to participate the survey. They were sent an invitation email that included a link to the research webpage. To ensure that the prospective respondents received the invitation email (see Appendix B), the invitation requested that the prospective respondents reply to the email.

Many of the 349 prospective respondents answered without the need for reminder emails.

Reminder emails were sent four weeks after the original email. After a further two weeks, if the prospective respondents still had not answered or had answered incompletely, the research assistant made contact by telephone to persuade therespondents to answer the e-questionnaire.

Some further responses were received. By the end of November, there were 198 usable responses and 151 incomplete responses. Incomplete responses occurred, for example, when respondents answered some questions but did not proceed to the next questions. These incomplete responses were not used in the data analysis. Therefore, the response rate was 38.82 per cent, calculated from 198 usable responses, divided by 510, which was the updated sampling frame.

The link to the research study (www.nhh.no/oil) waschosen as the result of the advice provided by an academic expert, during the consultation in the first phase, to use a URL that would would be easy for respondents to remember (see Appendix C). The webpage included two links to Part 1 and Part 2 of the e-questionnaire. (Actual links to e-questionnaires were not interpretable and hard to remember). The e-questionnaire was developed using Qualtrics.

The original questionnaire had only one part with 139 questions. However, during the consultation in the first phase, several academics suggested that it be divided into two parts, so that the respondents did not have to finish the questionnaire in one session (see Appendix D).

With a two-part format, respondents could complete each part in 10 to 15 minutes. Since the questionnaire was divided, a way to link between the two replies was needed. Therefore a question was added to obtain each respondent’s email address. This step was also necessary for the respondents to receive the research resultsand join the draw for the iPad.

4.7. Summary

This chapter reviews types of research designs and the criteria for selecting research designs.

Based on the review, the correlation design (cross-sectional design) is chosen. Validity concerns, including internal validity, external validity, statistical conclusion validity, and construct validity, are described. The empirical setting that requires the existence of key variables in relationships is also described, with the Norwegian oil and gas industry selected as the research context. Sample frame and sample procedures are explained, and measurement issues are addressed. The final section of the chapter addresses issues and procedures of data collection.