3. METHODOLOGY
3.1 Research design
Silverman and Marvasti (2008, pp. 376) argue that in especially empirical dissertations the methodology section is expected to show that the researcher understands the strengths and weaknesses of the strategy, design and methods. This thesis aims to elab-orate how complex business decisions are made and how AI will change it. Now I will go through my research design choices based on the model introduced by Saunders et al.
(2016, pp. 124). The research design layers, and my choices are shown in Table 4.
Research design layers (adapted from Saunders et al., 2016, pp. 124) and my choices
Layer My choice
Philosophy Interpretivism
Approach Inductive
Strategy Multiple-case study
Choices Mono method qualitative
Time horizon Cross-sectional
Techniques and procedures Interviews, cross-case analyses
The research philosophy of this thesis is interpretivism. It is a subjectivist viewpoint, which emphasizes the difference between humans and physical phenomena (Saunders et al., 2016, pp. 151). While subjectivity is a weakness, as my own values and beliefs affect the results, it may capture unique circumstances and interactions (Saunders et al., 2016, pp. 140–141). My research approach is inductive, which simply means that I start from the data and generate untested theories (Saunders et al., 2016, pp. 145). Strategy-wise I chose to use multiple-case study as it allows to capture real-life events (Yin, 2003, pp. 2) such as complex human-to-human decision-making processes. However, this de-cision was more or less made for me by others in the research project, which I discuss further in the next section. I decided to use just one method as it is simple, and it may prevent from under analyzing data. Nevertheless, triangulation may improve reliability
when compared to mono method (Saunders et al., 2016, pp. 157–159). My time horizon is cross-sectional, which can be used for describing phenomena or how factors are re-lated (Saunders et al., 2016, pp. 200). This was one of the aims of this thesis – to figure out how complex decisions are made and how MA information initiate and influence them. However, this time horizon does not allow studying change and development as longitudinal horizon would (Saunders et al., 2016, pp. 200). When it comes to techniques and procedures, it was chosen for me by other researchers of the project that I would be conducting interviews. This technique allows both fact- and opinion-based question, but it is a verbal report, which is subject to bias, poor recall and inaccurate articulation (Yin, 2003, pp. 90–92). I chose to use cross-case analyses developed by Bourgeois and Ei-senhardt (1988), which helps the researcher to go beyond initial impressions, but may lead to overly complicated theories (Eisenhardt, 1989). The strengths and weaknesses of the choices are illustrated in Table 5.
Strengths and weaknesses of my research choices
My choice Strengths Weaknesses Source
Interpretivism Captures unique
circum-stances and interactions Subjective: researcher's own values and beliefs affect the results
Saunders et al.
(2016, pp. 140–141)
Inductive Takes human perspec-tives and context of the events into consideration
Risk: lack of emerging data patterns
Saunders et al.
(2016, pp. 147–149)
Allows alternative
expla-nations Time-consuming, managers
are more likely to support de-ductive approach
Saunders et al.
(2016, pp. 147–149)
Multiple-case
study Explanatory studies:
good for answering 'how' and 'why' questions
Predictive studies: not good at enumerating 'what' ques-tions
Yin
(2003, pp. 5–7)
Multiple cases improve transferability compared to single case
Requires more resources Yin
(2003, pp. 47, 53)
Mono method
qualitative Simple, may prevent from under analyzing data
Triangulation may improve
reliability Saunders et al.
(2016, pp. 157–159)
Cross-sectional May be used for describ-ing phenomena or how factors are related
Longitudinal study would en-able studying change and development
Saunders et al.
(2016, pp. 200)
Interviews Allows both fact- and
opinion-based questions Verbal report: bias, poor
re-call, inaccurate articulation Yin
(2003, pp. 90–92) Cross-case
anal-yses Helps researcher to go beyond initial impres-sions
May lead to overly
compli-cated theories Eisenhardt
(1989)
Yin (2003, pp. 33–34) state that research design can be judged by four tests: construct validity, internal validity, external validity and reliability. Next I will go through the tests and case study tactics according to Yin (2003, pp. 34) and give my responses to them.
Construct validity tactics are 1) using multiple sources of evidence 2) establishing chain of evidence and 3) having key informants to review draft reports. I used data triangulation as we interviewed several informants on the same decisions. I intended to get original decision meeting memos to improve this, but I realized they do not exist, as the meetings are not usually that formal. In addition, we utilized investigator triangulation as other in-terviewers checked my work during a research paper writing process and evaluation of this thesis. They also analyzed the data separately and we had some unstructured con-versations on the data. Similarly, external sociologists analyzed parts of the data and we discussed the results with them. The structure of this thesis supports the chain of evi-dence. The empirical findings chapter elaborates the case stories of the informants, which are, together with the literature review, the base for the propositions in the synthe-sis of findings and discussion chapter. I did not send any draft reports to the key inform-ants, which hinders the construct validity.
The second design test, internal validity, is used for explanatory or causal studies. This thesis is mainly descriptive, so there is no need to address this. I tackled the third test, external validity, using replication logic by making cross-case analyses. However, the results need more testing before they are transferable. The last test, reliability, is some-what addressed as I followed a relatively clear case study protocol. Nevertheless, the research process was not completely linear (as discussed in the following sections) and it was not well-designed beforehand. Another reliability tactic is to develop a case study database. We stored the interview recordings, transcripts and notes in a folder where everyone in our research group had access. Some of the data was also handed to the previously mentioned sociologists. The research design tests, case study tactics and my responses are shown in Table 6.
Research design tests, case study tactics and my responses (adapted from Yin, 2003, pp. 34)
Test Case Study Tactic My responses Construct validity Use multiple sources
of evidence I used data triangulation (multiple informants on the same decisions)
Use multiple sources
of evidence We utilized investigator triangulation (other in-terviewers checked my work and did their own analyses. External sociologists analyzed parts of the data)
Establish chain of ev-idence
I base the synthesis of findings on the empirical findings chapter, which tells the case stories
Have key informants
review draft report I did not do this. No excuses.
Internal validity Several This is mostly a descriptive study, so internal validity is not relevant
External validity Use replication logic in multiple-case stud-ies
Replication logic is used as I made cross-case analyses. However, the results need more test-ing before they are transferable
Reliability Use case study pro-tocol
A relatively clear protocol was used, although it was not well structured beforehand
Develop case study
database Members of our research group have access to the case study database
Lukka and Modell (2010) state that there are problems in validating interpretative man-agement accounting research. Nevertheless, they point out that it creates rich emic de-scriptions with etic parts for creating explanations. To further judge the research design, I would like to add that quantitative methods should have been used for RQ2 and RQ3 as they are ‘what’ questions. Yin (2003, pp. 6) suggests that this type of questions favor survey and archival strategies. Alternatively, the research questions should have been changed. In addition, mixing the chosen case-study strategy with action or interventionist research methods would have probably provided results that are more comprehensive.
That kind of research design could have partially captured informal communication, which happens during the decision-making processes. This thesis relied only on inter-views, which could have been supplemented by direct observation. However, it would have imposed a risk of being an over-complicated research design. Despite the chal-lenges and problems in my research design, Granlund (2011) states that there is a need in AIS research for cross-sectional studies in ‘establishing a wider picture of current prac-tices and trends of development’. In addition, Korica et al. (2017) specifically called for elaboration on managerial work in practice with qualitative methods.