Study V was aimed at determining whether carrying a backpack and pulling a trolley with different loads influenced the
5.6 Kinematics adaptations while pulling the trolley and carrying the backpack under static and dynamic
My research objectives were to find the processes by which exercise of power affect creation of a working IS. In other words, the intention is to build a longitudinal process theory as opposed to a variance theory (Van de ven, 2007). Unlike variance
models, process models do not identify causal factors and predict outcomes, but instead explain how the process unfolds, identifying the mechanisms that move it along. Explanatory process theories explain an event by neither predicting what will happen (e.g. variance theories) nor describing what did happen (e.g. descriptive case studies), but by identifying the mechanisms that generate what we observe in the empirical domain. Grounded theory is especially meant for building such longitudinal explanatory process theories that focus on social processes (Glaser & Strauss, 1967).
Therefore, I employed grounded theory to perform data analysis and build the theory.
3.6.1.5. 1 Data analysis and coding
Concurrent with grounded theory and the qualitative case study approach, the data collection and analysis occurred iteratively. As field notes and interviews were transcribed, they were coded. However, while I was in the field, I did the coding manually and briefly. I could not perform a detailed coding and analysis at that point in time. I performed the detailed coding and analysis only after I got back to Cornell from my field. The detailed analysis followed Strauss and Corbin’s (1998)
recommendations for open, axial, and selective coding. I coded my field notes
(research journal), memos I generated in the field, most of the archival data including emails and excluding some company documents such as manuals that were not
significantly relevant to the research, and all interview data. Specifically, each passage (from one to several sentences in length) in every interview was assigned one or more codes reflecting what I perceived the speaker to be talking about. The code labels were the words used in the interviews (code in vivo) or were suggested by them. For
example, below in Table 3.3 (next page), is an interview excerpt from my interview with a production supervisor along with the codes:
Interview excerpt Codes The main point is the visibility
to the (group’s) owners so that they can control things from there. They sit far away from us. As owners, they need to know so many things such as the variation in production, profitability, production turn over, etc. If they can come to know on real time basis, they can trace and control things better. ERP provides that visibility.
Open codes: visibility to the owner, visibility to the management, owner’s control
Family: visibility to the owners
Super family (subcategory): visibility to the management
Axial codes: “real time visibility through traceaility”, “visibility to owners for controlling purpose”, “owner’s virtual control through increased visibility”
Category: visibility increaser; increaser of discipline
Selective code:
Category: disciplining tool Core category: technology frame
During open coding, I coded this passage in many ways such as “visibility to the management”, “visibility to the owners”, “owner’s control”. These are called
concepts. I grouped all similar concepts (e.g. “owners’ expressed desire to gain more visibility”) under a family “visibility to the owners”. Similarly, I had another family named “visibility to the top management”. Clubbing these two families together I formed a super family, named “visibility to the management” (see figure 3.1, next page).
TABLE 3.3: AN ILLUSTRATION OF CODES
I used Atlas-ti 5.0 software package to track the coded interviews and field data. These super families formed subcategories. Theory development occurred iteratively with coding. Thus, as coding progressed, I organized codes into families and trees (for examples, see figure 3.1), compared similarly coded passages to generate more abstract theoretical concepts, and wrote memos to propose abstract concepts and
Disciplining Agent Disciplining tool
Forced through visibility
Visibility to peer Visibility to the management
Time breakdown
Visibility increaser Increaser of discipline
Sub category
FIGURE 3.1: AN ILLUSTRATION OF THE CODING PROCESS
potential relationships. Field memos formed the part of these memos. Next, I describe these steps.
During the open coding stage the codes were primarily substantive, mostly using the vocabulary of the interviewees, e.g., “visibility to the owners”, “expected change”. As concepts emerged from the open coding, I started categorizing them and grouping them into families (e.g. “visibility to the owners”) and super families (e.g. “visibility to the management”). The super or super-super families (depending on the levels of break down) are called categories (Strauss & Corbin, 1998:114). In my case super-super families were categories. For example, I had “visibility increaser” as the category at this point in time. At this point in time, I grouped the super families of
“visibility to the management” and “visibility to peers” (subcategories) under two super-super families, named “visibility increaser” (category) and “disciplining tool”.
This is the inductive part or forward moving. The open coding stopped here. Then I proceeded to axial coding, which is more deductive and backward moving. Strauss and Corbin (1998) recommend axial coding once a phenomenon (category) is identified and further explanation is desired. Similar to Glaser’s (1978)
recommendation of coding around a core category (or subcategory), I started axial coding once open coding had uncovered phenomena of interest (category or
subcategory), in my case, for example, visibility to the management (subcategory) and visibility increaser and increaser of discipline (category).
I followed Strauss and Corbin’s (1998) axial coding recommendations, but treated this technique as a method through which to discover the relationships in the data, rather than as a set of restrictive rules. This was to avoid Glaser’s concern that axial coding could be overly restrictive, with questions emanating from the researcher rather than
the data (Locke, 2001). Suddaby (2006) expressed similar concerns, noting that grounded theory techniques require active interpretation by the researcher, not mechanical application of techniques. I used Glaser’s (1978) dimension family, i.e., coding for different types within a category or subcategory, for example different types of anticipated changes (by implementing ERP). As an example of my axial coding, consider the same excerpt I reproduced in table 3.2. Now I coded the passage around the subcategory, “visibility to the management”. I generated additional codes such as “real time visibility through traceaility”, “visibility to owners for controlling purpose”, “owner’s virtual control through increased visibility”. Similarly, during axial coding, I coded a passage as “discipline forced through visibility” (which formed a subcategory) during open coding (see figure 3.1). In relation to this new code, now I had additional codes (that formed subcategories) such as “visibility to peer”,
“visibility to managers”, “visibility to the top management”, and more specifically visibility to peer that occurred by increased breaking down of work (“work break down”) or increased breaking down of time of execution (“time break down”), which were concepts. In this manner, I went through each super family and family of codes, re-checked whether the codes within the family (and super family) belonged to that particular family, and coded further (axial codes) to show its relationship within the family (or categories) with the subcategories. To avoid the risk of fitting data into a pet theory (Glaser, 1998), I deliberately sustained a willingness to reassign the codes to a different family (or super family or super-super family) or even to change its label. I did make such changes.
After axial coding of the interview and field data, the code list had expanded to 895 codes. This large number of codes reflects two characteristics of my coding process.
First, because I did not know what theory might emerge from the data, I coded broadly
to cover several possible theoretical approaches to understanding role of exercise of power to create social consensus about technology and implementation. Second, perhaps, I coded for more depth than needed for the resulting theory, e.g., 68 codes were labeled consequences with some qualifier such as time, workload, visibility, and master scheduling. Although such detailed coding was useful in the constant
comparison process, not all of it contributed to the resulting theory.