• No results found

CHAPTER 5 RESEARCH METHODOLOGY

5.8 Data Analysis Procedures: Within-case and Cross-case analyses

Cases are examined as using within-case and cross-case techniques. The textual material was

9

subjected to multiple forms of analysis: axieal coding, latent semantic analysis and classical

10

content analysis.

11

The ‘within-case’ analysis utilizes the theoretical concepts from process metaphysics, while the

12

cross-case analysis is analyzed via assemblage theory. The within-case analysis is a preparatory

13

step for the cross-case analyses. The primary purpose of within-case analysis is to become

14

intimately familiar with the data and with each case as a stand-alone entity. Furthermore, the

15

within-case analysis identified key process elements, and data patterns from each case, thus

16

providing a solid foundation for the cross-case analysis. Once the key processes were identified,

17

they were examined via the concepts from assemblage theory in the cross-case analysis. The

18

purpose of conducting the cross-case analysis was to identify and investigate similarities and

19

differences between the cases along two dimensions: vendor neutral vs. vendor-led CCs and

20

established functioning CCs vs. abandoned or nonfunctioning CCs.

A generic depiction of my data analysis approach diagramed as a three-staged, multi-method

1

approach is found in Figure 5.8.1. This approach allowed for a kind of triangulation and

2

faciliated further analysis of the discovered categories. My initial coding during T1 was informed 3

by my research questions and the concepts from assemblage theory. Thus I did not take a

4

Grounded Theory approach in this analysis. However, when issues were found to emerge from

5

the data they were not ignored. The richer, ongoing interpretation was folded into my analysis.

6

The definitions of the concepts from Assemblage Theory were the focus of my early analysis.

7

Having established an initial set of broad categories of assemblages, I further drilled down

8

during the second stage, T2, of coding using latent semantic analysis. With the aid of a software 9

tool, Leximancer, I identified common themes and idioms arising in the respondent’s narratives.

10

The third stage, T3 is a synthesis and re-examination of the data (See Fig. 11). 11

Figure 5.8.1. Iterative data collection and analysis method (Adapted from (Woodside 2010))

12

Overview of the Analysis Method: I conducted the Interviews later transcribed, cleaned, and

13

annotated the MS Word files. The annotation is a basis for manual concepts coding and assists in

14

later validation of the machine coding and the synthesis of discovered concepts. The data

15

processing was divided into three parts: coding procedures (sorting), data reduction techniques

(categorizing), and drawing conclusions (mapping). Coding procedures dealt with strategies to

1

handle the semi-structured interview data as well as the more open-ended interviews and

2

documents. Interviews were coded using “Open Coding (Glaser and Strauss 1971)”, which

3

enables the examination, comparison, conceptualization, and categorization of data. I mapped

4

the results of the open coding concepts to the theoretical concepts of Assemblage. The Analysis

5

also deployed software machine coding, using both Leximancer, and nVivo. I utilized nVivo as

6

the main research database that contains all research related files, such as recording, manual

7

coding, interview notes, and secondary text documents. Leximancer’s latent semantic analysis

8

generated themes, and interpretations of those themes provided further insights, which might

9

otherwise have been missed(Crofts and Bisman 2010) . The technique is described more fully

10

below.

11

Figure 5.8.2. Overall Data Analysis Process utilizing Leximance and nVivo (Adapted from(Penn-Edwards 2010) )

12

Content Analysis 13

Content analysis is a method for extracting the contextual meanings and concepts from text

14

documents. I performed a specific type of content analysis called Latent Semantic Analysis

15

(LSA). The first step in LSA is to read an input text file. In doing so, the researcher typically

transforms words that contain many spelling variants (e.g., organize, organization, organizing,

1

etc.) into “word stems” – so that various grammatical and spelling variations are recognized as

2

having the same meaning. The second step is the creation of a document matrix-vector – which

3

is comprised of two elements: words and documents being analyzed (see Figure 5.8.3).

4

Documents are anything with a “semantic structure” that an analyst seeks to interpret. For

5

example, documents may be abstracts from research papers, blog posts, or advertising copy.

6 W o rd f req u en cy Documents D1 D2 D3 . . . DN Word 1 1 0 1 0 Word 2 0 1 1 0 .. Word N 1 0 1 0

Figure 5.8.3. Document Matrix

The third step in LSA is dimension reduction. The document matrix yields a large vector that

7

needs to be reduced to smaller sets of meaningful concepts. One of the simplest and powerful

8

dimension reduction approaches is Singular Value Decomposition (SVD). SVD is based on

9

linear algebra, details of which are explained in earlier studies (Landauer et al. 1998; Martin and

10

Berry 2007). SVD finds the obvious patterns and trends within the document matrix by

11

analyzing which words frequently appear in specific documents (frequency count), as well as

12

other words that often appear nearby (known as co-occurrences). These patterns are then

13

presented as concepts.

14

While there are many available software tools for performing LSA, a popular and recognized tool

15

within the IS and computer science literature is Leximancer. Several IS studies employing this toolset

have recently appeared (Crawford and Hasan 2006; Debuse and Lawley 2009; Mindel and

1

Mathiassen 2015; Ridley and Young 2012). In Leximancer, concepts are identified via words that are

2

weighted according to how frequently they occur within two-sentence “chunks” of text containing the

3

focal concept, compared to how frequently they occur elsewhere. The concepts then are clustered into

4

higher-level themes. The themes are comprised of concepts that appear together often in the same

5

chunks of texts. Leximancer provides results in the form of “overall” visual maps, where the analyst

6

can view the concepts, sub-concepts (keywords used in creating a concept), or themes (see Figure

7

5.8.4). Once the initial overall map is created, the analyst can change the theme size to adjust the

8

grouping of concepts on the map. For example, in order to select fewer but broader themes, or

9

conversely, to drill down into more detailed themes, the analyst has the ability to select the

10

desired level of granularity.

11

Figure 5.8.4. Leximancer processing: transforming words to themes

12

Leximancer produces visual diagrams, with certain key terms appearing in different-sized

13

circles. Not only is the size of the key term important, but the color of the circle encasing it is

14

important as well. Specifically, the “hot” colors (hues including red, orange, and yellow)

15

depict that the theme has a stronger relationship with the concepts (many or similar concepts

clustering to make a theme).

1

The strength of Leximancer is not merely the identification of concept tokens and patterns,

2

but in its ability to query, retrieve and further drill down into the texts. During this process, it

3

helps in identifying and excluding from the analysis extraneous terms and false concepts. Of

4

course this is an iterative and human guided process, and the researcher, like a pilot using fly-

5

by-wire guided avionics, manages the entire research process system-machine and manual-

6

and is responsible for the interpretation and sense-making of these analyses.

7

8

5.9 Roles of the Researcher