There were two units of analysis in this study: person and “person-in-situation.” The notion of person-in-situation was used throughout the qualitative content analysis, which was the major analytical part of this study. This approach is derived from the micro- moment time line interview technique of Sense-Making Methodology. Dervin (2008) ar- gued that “communicating (internal and external) is assumed to occur in micro-moments in time-space and the bracket of time-space chosen for consideration is itself a method- ological choice” (p.23).
Deductive coding
The deductive coding (i.e., content analysis) is, itself, a systems analysis of various types of content. One of the main characteristics of deductive coding is that the coding scheme is developed before the analysis begins (Spurgin & Wildemuth, 2009). For this particular study, the only code that was pre-defined prior to the analysis was the status of small world membership. The criteria were discussed above (see also: coding scheme in Appendix D). The unit of analysis for deductive coding was person which was used to identify the status of small world membership of an individual (i.e., case). The development of the coding scheme was done right after finishing the collection of public documents. The coding scheme was tested with a small set of data and revised before use with the whole collection of data. As a result, the final coding scheme, considering missing values, includes nine categories. In this coding scheme, one person (i.e., case) can belong to only one category. Figure 3.3 shows the final codes using in the analysis. Table 3.2 illustrates the number of cases by status of small world membership and content type. The majority of bloggers (517 out of 775) aggregated through Google Blog search were absolute outsiders. About 67 percent of Flickr users were identified as outsiders-in. The YouTube videos were uploaded mostly by outsiders-in and absolute
Figure 3.3: Coding categories of small world membership
insiders; 22 and 19 users, respectively. Wikipedia editors were the most challenging group to identify the small world membership. Only 161 out of 462 Wikipedia editors (approximately 35 percent) indicated both nationality and geographical location.
However, it is important to point out that the status of small world membership obtained by deductive coding was not directly used for any quantitative or systematic analysis (e.g., cross-tabulations). It was used only as a reference for open coding and pseudonyms.
The intercoder reliability test was conducted after the development of the codes. Another coder, a Thai graduate student in the United States, was selected. The second coder coded 10 percent of the cases across all types of content. The selection of cases for the intercoder reliability was via systematic randomization. The cases of each content type were sorted alphabetically. The researcher chose one out of every ten cases for this reliability test. The researcher conducted an informal training session for the second coder, including co-coding in order to make sure that the second coder understood the
Table 3.2: Number of cases by small word membership and content type Membership Type Blog Flickr YouTube Wikipedia Total
Absolute insider 98 21 19 36 174 Absolute outsider 517 7 10 116 650 Insider-out 9 1 2 9 21 Outsider-in 108 74 22 0 204 Anonymous-in 0 5 2 21 28 Anonymous-out 0 0 0 136 136 Insider-anonymous 8 0 2 19 29 Outsider-anonymous 35 1 2 29 67 Absolute-anonymous 0 1 2 95 98 Total 775 110 61 461 1407
procedure and the coding scheme. After the second coder finished the coding, the codes were compared. Scott’s pi was used to report the reliability of coding as it is appropriate for a nominal variable and two coders (Lombard, Snyder-Duch, & Bracken, 2002). For this study, the minimum acceptable coefficient (i.e., acceptable level of agreement beyond chance) was at 0.70, which is appropriate for an exploratory study (Lombard et al., 2002). The Scott’s pi value computed for this study based on the above procedure was 0.763, which was deemed acceptable.
Open coding
This study applied qualitative content analysis as a main analytic framework. In gen- eral, the analysis relied on an inductive strategy, which generates and confirms theory based on patterns, themes, and categories emerging from data (Patton, 2002). How- ever, in terms of practical and theoretical process, a grounded theory (Glaser & Strauss, 1967) approach was specifically implemented. Patton (2002) noted that grounded the- ory “operates from a correspondence perspective in that it aims to generate explanatory propositions that correspond to real-world phenomena” (p.489).
Grounded theory emphasizes what is known as open coding (Strauss & Corbin, 1998, p.223), which highlights the discovery of concepts, their properties, and dimensions directly from data. Thus, the categories and a coding scheme are derived from the data, rather than from other studies or theories.
There are three phases for grounded theory coding: 1) an initial phase involving naming each segment of data; 2) a focused, selective phase that derives the codes from the initial phase; 3) an axial coding phase focusing on the properties and dimensions of the categories (Charmaz, 2006).
The initial coding is to ensure that the analysis is based on careful reading and observation. For this study, different techniques were implemented based on content types. For textual data (including blog posts, Wikipedia entries, and associated data of non-textual contents – metadata, tags, comments, etc., and memo notes), the researcher read and observed text (and its context) carefully and created a new code or applied an existing code to appropriate messages. Basically, codes, marks, and annotations were attached to each phrase, sentence, or paragraph of interest. Note that one message could have more than one code. For still image data, position-based coding was applied. NVivo allows position-specific description of image. Certain descriptions refer to the specific area of an image, rather than the overall picture. A researcher can highlight a spot of interest (the size is flexible) and insert a description of that spot. For audio and video files, the coding technique was based on timeline coding. The researcher listened and annotated the codes at the timeline (in range). Partial transcription was applied only for responses that might be used for citing. The transcription consisted of timestamps and text.
The second phase of the coding is called focused coding, referring to the use of “the most significant and/or frequent earlier codes to sift through large amounts of data. One goal is to determine the adequacy of these codes. Focused coding requires decisions about which initial codes make the most analytic sense to categorize data incisively and completely” (Charmaz, 2006, p.58). In this phase, the researcher reviewed the
codes generated from the initial phase. The researcher used NVivo to organize the data by code. The analytical process included sorting, synthesizing, integrating, and organizing data. In addition, data that belong to the same codes were compared to ensure consistency. With the query (search) functionality of NVivo, the researcher used keywords referring to each of the codes to check if there was anything else that might be included.
After finishing the focused coding, the codes were aggregated into a separate space (i.e., coding book). Then the researcher conducted theoretical coding (i.e., axial coding) by making connections among the substantial codes and constructing the hypothesis to be integrated into a theory based on such connections. In this phase, the researcher con- structed a coherent picture (i.e., analytic story) by sorting, synthesizing, integrating, and organizing categories. In addition, the coding involved the construction of relationships between codes and categories (i.e., codes and codes; codes and categories; and categories and categories). At the same time, the researcher also evaluated the convergence – in- ternal homogeneity (the extent to which the data that belong in a certain category hold together or “dovetail” in a meaningful way) and external heterogeneity (“the extent to which differences among categories are bold and clear”; Guba, 1978; Patton, 2002).
As a result, the coding schemes were primarily framed by the three research questions of this study. The code book is shown in Appendix D.
Throughout data collection and analysis processes, the investigator made informal memo-notes. According to Charmaz (2006), memo-writing helps the researcher to con- struct and elaborate the analytic frameworks, and accelerates productivity. The memo- notes will be written throughout data collection and data analysis.
It should be noted that NVivo was used throughout the process of data analysis including coding and annotating transcripts and documents, memo writing, retrieving data and codes, manipulating codes and categories, and creating visual diagrams.