As an analytical procedure that identifies patterns in textual data, qualitative content analysis relies heavily on coding technique (Jensen et al., 2016: 250). In order to establish a firm grounding in the data, and obtain a general direction in which the data could be analysed, all the textual data was looked through before coding. After which, Kelle’s six-step coding procedure was introduced to guide the process of coding in computer-aided qualitative data analysis software QSR NVivo. The six steps are ‘format textual data; open coding of data; memo writing; compare text segments that have assigned the same code; integrate codes, and attach memos to codes and develop a main theme’ (Jensen et al., 2016: 252).
93
with QSR NVivo, including unifying the format and translation of text from the raw data. The translation work for Weibo data, from Chinese to English, was conducted by two independent translators. Although the work done by two tranlsators reached a relatively high level of consistency in meaning expression, a third translator still got involved and made informed judgements on which expression was better, especially when translation discrepancies emerged. The divergences mainly arose from the translation of proper noun and requested certain support of background knowledge, such as the choice between ‘right to know’ (see page 164) and ‘right of knowing’. I have to admit that the work of data translation came across numerous challenges, especially when translating terminologies or the name of academic works. For example, in the data Weibo user ‘Wangxiaojian’ referred to Rachel Carson’s famous work Silent Spring, which, however, was mis-translated to ‘The Lonely Spring’ (see page 128) without timely correction.
For addressing two research questions, the thematic distribution of GM future related discussions and the online interactions taking place among the major stakeholders were investigated. After unifying the format of the text and holding it as an ‘internal source’ in QSR NVivo, the data was analysed and categorised into three sets of codes, namely ‘theme’, ‘identity’ and ‘communication model’. In general, a semi-open coding scheme, rather than the ‘open coding of data’ mentioned by Kelle (Jensen et al., 2016), was employed at this stage subject to the two selected approaches of qualitative content analysis (conventional and directed). Specifically, the data was coded thrice in terms of the themes, the identities of participants and the communication models respectively. After repeatedly reading the original data, the initial thematic codes were completely derived from the text using NVivo software to capture the key issues discussed by the participants. The codes were then sorted into categories based on how different codes are related and linked (see Appendix 1), to see how the perspectives were distributed and associated to each other, which helped produce a description of each category in the following analysis.
94
Predetermined codes were set up for coding the identities of participants and identifying the communication models that the samples follow. As can be seen in Appendix 1, the predetermined code framework of ‘communication model’ is composed of Irwin’s ‘First-order’, ‘Second-order’ and ‘Third-order’. While the framework of identity was originally constructed by ‘Experts’, ‘Non-experts’, ‘Opinion leaders’ and ‘Governments’, inspired by Augoustino et al.’s classification (2010). During the process of coding, any sample involving certain stakeholder (stakeholders) was coded with the corresponding identity (identities). The text that could not be organized with the initial identity coding scheme has been identified with new codes in terms of the routine of directed content analysis (Hsieh et al., 2005). ‘Public intellectuals’ and ‘NGOs’ were therefore included and made up the final coding framework together with the predetermined codes. Correspondingly, all the theories and academic literature identifying or giving an explanation to either the behaviors of participants, both predetermined or newly identified, or the models of communication were picked out and formed a theoretical framework as has been elaborated on in Section 2.3. Key thoughts and ideas relevant to the categorised codes were written down and saved as ‘memos’, to record ‘interesting connections’ between codes, ‘links to theory’ or any ideas that need to be highlighted for the next stage of analysis (Jensen et al., 2016: 252). Till then, the coding scheme of the present research was built up.
After finishing the initial coding work, all the codes were organised and regrouped into new meaningful clusters (‘nodes’) to depict the relevance and relationship between codes/themes. At the meantime, a hierarchical ‘structure of node’ was established (Kaefer et al., 2015). The memo attached to each code was merged and then a new cycle of analysis began; to compare and figure out the connections between different aspects of the textual segments involved in each node. The whole coding process underwent a continuous ‘back and forth’ interplay (Jensen et al., 2016) with the data to ensure the accuracy and consistency until the categories could no longer be integrated any further. Then, definitions for each final category were developed. At this point, the coding procedure of the present study was complete. The memo materials written down during
95
the coding process played a significant role in the following analytical work.