5.2. Research methodology
This research uses a qualitative research approach and it is interpretive according to Klein and Myers’ (1999, 69) terms. They define research to be "interpretive if it is assumed that our knowledge of reality is gained only through social constructions such a[s] language, consciousness, shared meanings, documents, tools, and other artefacts, … it attempts to understand phenomena through the meanings that people assign to them". The research approach is made according to the settings of Eisenhardt (1989) with the eight basic steps, Table 6.
Table 6. Process of building theory from case study research (Eisenhardt 1989, 533).
| Step | Activity | Reason |
|---|---|---|
| Getting started | Definition of research question | Focuses efforts |
| Possibly a priori constructs | Provides better grounding of construct measures | |
| Neither theory nor hypothesis | Retains theoretical flexibility | |
| Selecting cases | Specific population | Constrains extraneous variation and sharpens external validity |
| Theoretical, not random, sampling | Focuses efforts on theoretically useful cases – i.e., those that replicate or extend theory by filling conceptual categories | |
| Crafting instruments and protocols | Multiple data collection methods | Strengthens grounding of theory by triangulation of evidence. |
| Qualitative and quantitative data combined | Synergistic view of evidence | |
| Multiple investigators | Fosters divergent perspectives and strengthens grounding | |
| Entering the field | Overlap data collection and analysis, including field notes | Speeds analyses and reveals helpful adjustments to data collection |
| Flexible and opportunistic data collection methods | Allows investigators to take advantage of emergent themes and unique case features | |
| Analyzing data | Within-case analysis | Gains familiarity with data and preliminary theory generation |
| Cross-case pattern search using divergent techniques | Forces investigators to look beyond initial impressions and see evidence thru multiple lenses | |
| Shaping hypotheses | Iterative tabulation of evidence for each construct | Sharpens construct definition, validity, and measurability |
| Replication, not sampling, logic across cases | Confirms, extends, and sharpens theory | |
| Search evidence for “why” behind relationships | Builds internal validity | |
| Enfolding literature | Comparison with conflicting literature | Builds internal validity, raises theoretical level, and sharpens construct definitions |
| Comparison with similar literature | Sharpens generalizability, improves construct definition, and raises theoretical level | |
| Reaching closure | Theoretical saturation when possible | Ends process when marginal improvement becomes small |
She draws a roadmap for building theories from case study research using a combination of qualitative grounded theory and case study approach.
Alasuutari (1995) compares qualitative analysis to riddle-solving. According to him “To call a study qualitative research it must be assumed that inferences based on purely qualitative analysis, or other references to excerpts or cases in the data, are used as clues to solve the riddle” (ibid, 8). Alasuutari has defined the qualitative analysis as consisting of three intertwined phases of purification of observations, unriddling and combinations of new viewpoints (ibid).
According to Järvinen and Järvinen’s (2000) terms the grounded theory approach is quite demanding and hard for the researcher to follow completely, therefore it is also done using, as they call it, a light version approach. Though they advocate grounded theory as it suits an explorative approach in searching for new phenomenon. This study can be seen as a combination of grounded theory study and a multiple case study where the cases are put together with interviews and other relevant material collected. In this study the above Eisenhardt’s (iterative) sequence is used appropriately. In Sections 5.2.1 – 5.2.8, the steps presented in Table 6 are discussed in more detail as well as how they were applied in this focal research effort.
5.2.1. Getting started
This study started with initial definitions of research questions in broad terms. It is important to make the research domain as focused as possible, as otherwise the researcher can be overwhelmed with impertinent data and needless work. Also a priori specified constructs are in place to help the design of theory-building research. But the research questions and the constructs should be considered only as tentative because they must not obstruct the findings and analysis of the empirical data. The questions and constructs only help in the research work and they should be defined in a way to allow for modifications and enhancements as appropriate. In Chapter 4, a tentative and simple “Software contracting process” model was first developed. Also the taxonomy used was presented. This step was supported by literature surveys in order to find similar reported research settings, descriptions and results. The researcher participated during this period in several seminars where the legal aspects of software were discussed.
5.2.2. Selecting cases
The main selection criterion for the software companies was that they should fall into one of the three business modes (COTS, tailored and MOTS) identified, as the central strategies in software business, cf. the classification in Section 2.2.1 Software business modes. The data was collected from twelve small and medium size software companies in Finland, and from Finnish companies that have operations in the USA. The selected companies are briefly described in Appendix C. Because of the confidential nature of the collected data and the information that the managers have given during the interview sessions the companies wanted to preserve their anonymity. This request was of course respected. Each of the business categories includes several companies in order to give more reliability and material to compare the contracting processes from different types of software companies. Multiple case approach was selected in order to strengthen the research base and validity of the results as the business line in focus is characterized by companies whose maturity level varies. The firms are small and fast developing, they operate in one or several business modes and they operate in different cultures. In the light of studied and published knowledge about the software contracting process, studying only a few companies could be more unreliable. From Appendix C it can be seen that the research context included four “pure” COTS companies, three tailored companies and only one MOTS company. In addition to this, one company operated in all three modes. However, this company also had during that one-year period changed its operations to COTS/MOTS, abandoning the tailored operations. The remaining four companies operate in the COTS/MOTS modes. The classification basis for these three categories was:
Based primarily on the interview material, i.e. transcribed tapes, where it was possible to find out clearly the business modes.
During the interviews the interviewees description of the company’s strategies in relation to the business mode.
The use of other relevant material received from the company, e.g. company brochures, newspaper articles and www-sites.
With hindsight, it can be said that the interview questions could have been formulated more clearly to give a more accurate classification scheme, e.g. direct annual turnover connected to the COTS, tailored and MOTS business modes might have been sufficient.
5.2.3. Crafting instruments and protocols
The data collection methods included interviews at the twelve companies. During the visits, as well as after the visits, other material was also collected, e.g. contracts, contract templates, company presentations and quality standards. The unit of research was the software company as most of the interviewees represented the managerial level and they had a good overall view of the company and its operations. After the interviews, interview protocols and field notes were written up where the situation and other non-verbal ideas and observations were put down to prepare for subsequent case analysis. Two interviewers, working separately, carried out the interviews. Only in one case did they do an interview together and this was at the beginning of the project in order to standardise the following interview sessions and learn interview techniques from each other. According to Eisenhardt the use of multiple investigators is justifiable as it enhances the creative potential of the study and the convergence of observations from multiple investigators enhances confidence in the findings (ibid). The interviews were completed using the semi-structured interview approach where the interview themes and questions were prepared beforehand, see Appendix A. The semi-structured approach was selected in order to have an adequate focus for the discussions. However, space was given to follow new leads and other interesting topics emerging from the conversations.
5.2.4. Entering the field
Eisenhardt notes the "striking feature of research to build theory from case studies is the frequent overlap of data analysis with data collection" (ibid, 538). If (as usual) the data gathering process is implemented during a longer time period and the researcher starts to analyse the field notes and the empirical material simultaneously with the ongoing data collection process he or she is able to adjust the data collection instruments on the fly, e.g. adding new questions to the interview protocol. Even new cases can be added if they give more insight into the focal object. According to Eisenhardt this is a "legitimate" procedure, as in the case of qualitative theory-building study when the researcher tries to understand the focal phenomenon in as much as depth as possible. But she, understandably, adds that this flexibility does not excuse being unsystematic.
In this focal research the interview period lasted from 16.2.2000 to 16.5.2001. Some new topics were added during the data collection process and two new interviews were carried out in order to get more specific data from issues that had emerged during the analysis work. These interviews were also valuable from the data triangulation perspective as they were carried out in one company with different levels and responsibilities. From the interviews, short memos were written with details of the company name, place, interviewees, time, duration, and other possible observations of interest.
5.2.5. Analysing data
Strategies for sensemaking as Langley (1999) calls the seven generic strategies described in her explanation and thought provoking paper. These seven strategies cover in depth the analysis of the process data (Table 2). She describes two different philosophies evaluating organizational processes, the first group elaborates a priori process theories which they subsequently test using a longitudinal time series and event-history methods. The second group gets off the ground from the opposite direction, i.e. they analyse first the processes of collecting data with various methods and specificity, even in real time. Using this material they try to create an explicit picture – model or theory – to comply and reflect the real world as exactly as possible, by being well-founded and universal. This focal research could be placed between these extremes, as the researcher has pre-understanding about the research area and some interventions were made before the conceptual model was elaborated. Quoting Orton (1997) the study has features of the "iterative grounded theory research, in which researchers cycle back and forth between process theory and process data to produce process knowledge" thus taking advantage of both inductive and deductive approaches.
Langley further notes that "In theorizing from process data, we should not have to be shy about mobilizing both inductive (data-driven) approaches and deductive (theory-driven) approaches iteratively or simultaneously as inspiration guides us" (ibid, 708). Next these approaches and their usages are explained in Table 7 in more detail.
Table 7. Seven strategies for sensemaking (Langley 1999, 696).
| Strategy | Key anchor points | Specific data needs | “Good theory” dimensions (Weick) | Form of sensemaking |
|---|---|---|---|---|
| Narrative strategy | Time | One or few rich cases. Can be helped by comparison. | High on accuracy. Lower on simplicity and generality. | Stories, meanings, mechanisms |
| Quantification strategy | Events, outcomes | Needs many similar events for statistical analysis: one or few dense cases is best. | High simplicity, potentially high generality, modest accuracy. | Patterns, mechanisms |
| Alternate templates strategy | Theories | One case is enough. Degrees of freedom come from multiple templates. | Each theory can be simple and general. | Mechanisms |
| Grounded theory strategy | Incidents, i.e. (units of text), Categories | Needs detail on many similar incidences. Could be different processes or individual-level analysis of one case. | High on accuracy, moderate simplicity. May be difficult to go from substantive theory to more general level. | Meanings, patterns |
| Visual mapping strategy | Events, orderings | Needs several cases in moderate level of detail to begin generating patterns (5-10 or more). | Moderate levels of accuracy, simplicity, and generality. | Patterns |
| Temporal bracketing strategy | Phases | One or two detailed cases is sufficient if processes have several phases used for replication. | Accuracy depends on adequacy of temporal decomposition. Moderate simplicity and generality. | Mechanisms |
| Synthetic strategy | Processes (e.g., decisions, change efforts, new products) | Needs enough cases (5+) to generate convincing relationships. Moderate level of detail needed for internal validity. | Modest accuracy. Can produce simple and moderately general theories. | Prediction |
The key anchor points depicted in the table are according to Langley to fix attention on some point(s) that help to structure the material, but also to determine which elements will receive less attention. Interestingly the different approaches produce different levels of accuracy, generality and simplicity (ibid, 694). The data analysis process carried out in this study exploiting the above-depicted Langley’s strategies is subsequently described:
All the cases were analysed several times in order to minimize the possibility of the understanding changing during the reading and analysis process. When using transcriptions the categorization process took five weeks, but the tentative computerized analysis work used somewhat less time. Hirsjärvi and Hurme describe this process as depicted in Fig. 28.
During the open coding phase which was first done assisted by software several tentative categories were established beforehand and during the analysis process itself new (sub) categories were added. Among these categories were e.g. business processes, software processes, legal processes and contracts but as this approach was not fruitful it was abandoned, ref. below Alternate templates strategy. Next the approach was totally turned around and the analysis work started by defining all process activities from the data, and there was a change to use the axial coding method in this phase (Strauss & Corbin 1998). In axial coding the researcher selects the main characteristics as the object of the analysis and the coding is done using these axles. Strauss and Corbin (ibid, 124) define the axial coding to be "the act of relating categories to subcategories along the lines of their properties and dimensions". The activities thus form the smallest unit of analysis and these were subsequently classified with four attributes:
Phases comprised of negotiation, definition, production and assessment classes.
Interaction process comprised of exchange, adaptation and coordination classes.
Ford’s stage comprised of pre-relationship, exploratory, developing, and stable classes.
Motor comprised of prescribed and constructive classes.
Further to these elements the following attributes to help to scrutinize the processes were attached: inputs, outputs, owner and events, see Chapter 4.
All the process activities were marked and translated into English and subsequently they were inserted case by case into individual Excel tables in order to help the analysis work by enabling the data to be viewed in more compact form, Appendix D. After this the different classes were inserted together in one Excel table to make it possible to order the activities in different orders to see any interrelationships and patterns, Appendix E. These shorthand tables were later modified to become countable in order to be able to derive some coarse statistics out of the data.
Narrative strategy was used translating short English synopses from each transcribed case material. The synopses included contractual issues and contract process activity descriptions that later formed the base for the detailed and numbered process activity lists, Appendix D. In one case a short written narrative and descriptive story was written in order to highlight the interesting situation where the focal company had a completely immature contracting process, Appendix H.
Quantification strategy was used to some extent when the process activity lists were transformed into calculable forms, Appendix D and E. Tabularisations were done using the Excel software and the worksheets were then capitalized to produce simple 3x3 matrixes from different interesting phenomena. In this context it was not the case of using formal and extensive statistical methods, as the scope of the available data did not permit this.
Alternate templates strategy were used during preliminary tentative computer assisted analysis work. The data categorization was started by reading the transcriptions through several times. First a robust tree structured pre-categorization was created, and during the analysis process augmented up until 15 main categories at the root level and approximately 65 subcategories at the so-called child level after becoming acquainted more with the material. However, soon this seemed to be too hard to use, as well as being too atomized. There was an evident risk of losing the overview of the whole data. So the next step was to drop the number of the main categories to 10 and the subcategories to 51. This pre-categorization was also found to be too detailed. The second classification approach was to classify all incidents and expressions on different contracts, but this perspective would have been useful if the viewpoint of the research would have only been from the legal and contractual perspective. It did not help the dynamic processual study. Reading and pondering about the empirical data with different tools and from different angles gave more and more insight into it. However, two problems could be expressed and they are, first, that the computerized analysis easily produces too many categories and, secondly, technically the computer screen is not so agreeable and handy to use in the process as normal A4-paper. Although the time spent with the empirical material and the software was not in any way wasted time. After this experiment the researcher started to read the transcription papers in order to categorize the process elements activities. The analysis made reading the papers was also more illustrative and it was also easier to keep track of the whole – maintaining a helicopter perspective – when analyzing and classifying the smaller process activities.
Grounded theory strategy was used extensively as can be seen in Chapter 6. This was the main method used in this study. It was also best suited for this type of research where the researcher approached new and untouched fields. The objective of the grounded theory based research approach is to generate theory rather than to test it, this gives the researcher analytical tools for handling a large empirical data base, and helps him to find alternative meanings for the phenomena (Rantala 1999). This enables the work to be at the same time systematic as well as creative and during the whole research process the researcher strives to identify, develop and proportion the concepts, i.e. the building blocks of theory.
Visual mapping strategy was used to some extent by describing every case as the example in Appendix I shows.
Temporal bracketing strategy is used to visualize the contracting process as well as other contractual issues in the different business modes as shown in Chapter 6, and see especially Figures 29, 30 and 33.
Synthetic strategy is put to use in Chapter 7 to elaborate the Dynamic contracting process model. According to Langley the process theorizing method produces an enriched understanding and explanation, however, it often lacks predictive power (ibid). The synthetic method uses the process as a whole (unit of analysis) and constructs global measures from the detailed event data in order to be able to describe and compare the corresponding processes from different cases. The synthetic strategy requires: 1) clearly defined process boundaries – in this focal case it is the software contracting process, 2) abstraction level that is high enough to permit the comparison between several cases – in Chapter 4 elaborated conceptual model fills these requirements well, and 3) the strategy requires a sufficient number of cases – this study used twelve different cases that gave enough data to infer rational conclusions about the contract process unfolding.
To use the Möller and Wilson model as a base in constructing the software contracting process model was quite a straightforward effort. In this study the customer’s viewpoint was deliberately left out, as the main focus was on the software company’s contracting process and its behaviour linked to software development processes. Therefore the customer context in the Möller and Wilson model was left for less scrutiny. From this model the environmental context, supplier and task specific characteristics, organizational characteristics, interaction processes and outcomes elements were used. In the elaborated software business model these elements were specified as reflecting the business at issue.
This study tries to capitalize on the above-mentioned seven methods to a varying extent in order to cover and get the relevant issues from the empirical data from as many angles as possible and in as many ways as possible. This data “massaging” helped the researcher to get better acquainted with the data as thoroughly as possible. To sum up, from the above-discussed strategies depicted by Langley, the grounded theory was used vitally in this research. The other strategies were more or less applied to enhance the understanding of and familiarize with the focal data.
5.2.6. Shaping hypotheses
According to Eisenhardt this step is a two-part process where the researcher tries to refine the definitions of the constructs and to find evidence to measure and support the construct in each case. If the data and the theory work well together then the construct converges and advocates reality. Weick proposes a concept of disciplined imagination that consists of problem statements, thought trials and selection criteria (Weick 1989).
The conceptual models used in this study were first constructed and introduced in Chapter 4. These models were based on the literature and pre-understanding of the subject. The analysis of the data is then presented in Chapter 6 and following this analysis the results of the iterations and refinements are depicted in Chapter 7. These descriptions include the Augmented basic contractual model, the modification of the Software contracting process model and lastly the Dynamic Augmented basic contractual model.
5.2.7. Enfolding literature
According to Eisenhardt (ibid, 544) “an essential feature of theory building is comparison of the emergent concepts, theory, or hypothesis with the extant literature”. She further notes that this is especially important and valuable if the results of the emergent theory conflict with the literature findings.
The literature on this focal research subject of the contracting process in software developing company is quite sparse. In this study the results are to some extent comparable with the NTNU model for the implementation of a software development project (Kilde 1998, Kilde 1999). This is described more in Chapter 7. In Chapter 8 the main theoretical, empirical and managerial implications are presented and discussed.
5.2.8. Reaching closure
The researcher has two central problems that he or she must pay attention to and these are when is the empirical material sufficient and when to stop iterating between the theory building and data analysis. Eisenhardt contemplates e.g. the number of sufficient cases and as she notes that no optimal number exists because the adequacy is of course dependent on the focal subject as well as on the saturation point (ibid). But in practice the researcher has set in advance some limits that also depend on the available funding and time. In theory the research process described above is a coherent unfolding process, but in reality the process always considerably overlaps (Alasuutari 1995).
In this study the number of cases grew to twelve, even though the last cases started to resemble the previous ones, the researcher cannot state that there would not have emerged any new contractual issues concerning the contract process itself as the business environment is so multifaceted.