«Abstract This paper revisits the debate on the units of Content Analysis (CA) for the purposes of Corporate Social Reporting (CSR) research and also ...»
Movena et al., 2006; Turner et al., 2006). This type of assessment though is limited “as this precludes assessment on scope, coverage, completeness, relevance, reliability, and other such desirable qualities of external financial statements” (Hammond and Miles, 2004, p. 64).
- 26 Whilst both third party verification and adoption of reporting guidelines are relatively new CSR research fields, CA studies employing the basic distinction of quantitative vs. qualitative have a long history and are frequently identified in the CSR literature.
Behind this evidence, there is the intent of researchers to identify whether companies disclose hard-fact, substantial information or not. However, a major limitation of this distinction is that it is not normatively rooted; as Erusalimsky et al. (2006) note, “Content analysis got us so far but more substantive, explicitly normative templates would seem to be essential to guide future work” (p. 19). With regards to this distinction, it is uncertain whether all the quantitative information that companies disclose is of relevance (since they may as in the BA study disclose some CSR data of the industry) or that all declarative statements are of less importance (e.g. description of policies adopted providing specific examples). Likewise, it is unlikely that the general vs. specific distinction that was also identified in the literature (De Villiers and Van Staden, 2006) would be of particular usefulness to CSR, considering that it brings some relatively limited benefits in context (by focusing on more normatively oriented, latent information) comparing to the limitations in validity (in case that an index approach is employed, as in De Villers and Van Staden ) or reliability (from the otherwise possibly vague definitions).
A more useful CA distinction, based on Legitimacy Theory, appears to be the substantive vs. symbolic distinction. This distinction was suggested by Pfeffer and colleagues (Pfeffer and Salancik, 1978; Pfeffer, 1981, see also Meyer and Rowan, 1977; Richardson, 1985; Ashforth and Gibbs, 1990; Suchman, 1995; Weisul, 2002 for related arguments) and has not been employed widely in the CSR context (but see Savage et al., 2000; Day and Woodward, 2004; for exceptions). Substantive legitimation is evident in the works of Rousseau and Habermas and involves “real, material change to organizational goals, structures and processes, or in socially institutionalized practices” (Savage et al., 2000, p. 48). Symbolic legitimation on the other hand traces its roots to the work of Marx and Weber; it involves “the symbolic transformation of the identity or meaning of acts to conform to social values” and is predicated on that “the acceptance of authority resides in the belief in the legitimacy of the order independently of the validity of that order” (Richardson, 1985, p. 143, emphasis in original).
- 27 Drawing on organisational theory and their own research, Savage et al. (2000) have offered 12 legitimation strategies, three substantive and nine symbolic which could also incorporate the oft cited strategies by Perrow (1970), Lindblom (1994) and O‟Donovan (2002) and are presented in Appendix A (I). However, when it was attempted to apply this framework as such in the BA study it was quickly realised that some of the symbolic strategies were easily conflated and it was decided to merge some categories, adopt a more „pragmatic‟ approach and customise it to reflect the set research questions, focusing on the impacts on CSR of detrimental activities. This resulted in the employment of six legitimation strategies, as depicted in Appendix A (II). The new categories were thus less ambiguous, although it should be noted that two categories, the role performance and the identification of symbols, end up being used more often (arguably, there is no way out of this unless complicated, exhaustive and time consuming decision rules are established allowing for the use of more detailed categorisations).
The main benefit of this distinction, albeit categorical, is that it assists in identifying some latent characteristics of the data. Further, since parts of these arguments may also lend support to other theoretical frameworks (such as institutional theory, business ethics theory, resource dependence theory and even image and competitive advantage arguments), the categories may also identify relationships between these theories and synthesise theoretical arguments underpinning CSR research (as in e.g.
Roberts and Chen, 2006). However, as in the case of the positive vs. negative CSD, some theoretical inconsistencies may also arise from the adoption of this approach and e.g. even more „ethics‟- oriented organisations, following a major legitimacy threat, may disclose increased symbolic rather than substantive CSD through admitting guilt and offering apologies. As also pointed out earlier, therefore, if this categorical approach is complemented with a clearly inductive one, in a mixed CA design, this would further bring some triangulation benefits to the analysis. Two of these approaches are now discussed.
- 28 Thematic distinctions Although Boyatzis (1998) considers that “a theme may be identified at the manifest level or at the latent level”, Berelson (1952), Holsti (1969a) and Krippendorff (2004) seem to agree that thematic distinctions, as opposed to the categorical distinction of the oft termed „theme‟ (the subject) of the disclosure described in the CA protocols, such as the one depicted in Figure 1, refer to “unitizing freely generated narratives thematically” (Krippendorff, 2004, p. 107). Two such approaches were identified in the reviewed literature. Both did not take a quantitative form, therefore may consist of a CA in the broader view, as defined by Stone et al. (1966), Holsti (1969a,b) and Krippendorff (1969).
Quantising – Miles and Huberman (1994)
A number of CSR studies explicitly or implicitly adopt the approach to data analysis suggested by Miles and Huberman (1994), which Ritchie and Lewis (2003) have adapted and graphically represented as an analytical hierarchy of the stages and processes in qualitative analysis. This „quantising‟ approach of Miles and Huberman (1994) involves primarily three forms of activity: data management in which the raw data are reviewed, labelled-coded, sorted and synthesised; descriptive accounts in which the analyst makes use of the ordered data to identify key dimensions, map the range and diversity of each phenomenon and develop classifications and typologies;
and explanatory accounts in which the analyst builds explanations about why the data take the forms that are found and presented (Ritchie and Lewis, 2003).
CSR authors have adopted this analytical perspective with variations regarding the precision with which they conducted each stage of the analysis. Owen et al. (2000), Woodward et al. (2001), Adams (2002) and Roslender and Fincham (2004) presented their findings by set preposition, or under interest topic, implying at least the use of the identification of initial themes and sorting data by theme or concepts steps of the hierarchy, and then moved straight to develop explanations. O‟Donovan (2002) identified themes and patterns, detected cross patterns with his quantitative data set and moved to develop explanations, applying rather loosely the analytical hierarchy,
- 29 although he followed a systematic approach to combine the qualitative with the quantitative data in the analysis stage. O‟Dwyer (1999; 2002; 2003; 2004) on the other hand, explicitly adopted the Miles and Huberman‟s (1994) approach by identifying underlying themes, developing a coding scheme, summarising and synthesising data, identifying cross case patterns in the data and detecting regularities and developing explanations in the evidence collected. He was also cautious in the last-generalisation-step to avoid presenting a “smoothed set of generalisations that may not apply to a single „interview‟” (Huberman and Miles, 1994, p. 435) and made efforts to preserve the uniqueness of certain individual interviews (see particularly O‟Dwyer, 2004, for a more detailed and focused description of his approach).
The distinct benefits of this approach include that it offers a simple, grounded and therefore quite valid approach to qualitative analysis, which allows for all variation in themes and topics to be revealed and captured. This is particularly useful when exploratory research is conducted and the widest possible variety of themes is sought.
Further, although this method of data analysis is primarily qualitative, as EasterbySmith et al. (2002), note “it is still possible to introduce some element of quantification into the process” (p. 119), particularly when employing some computer aided qualitative analysis software. However, even more structured approaches to thematic analysis may be adopted, as discussed in the next section.
The variation of Bebbington and Gray (2000)
Bebbington and Gray‟s (2000) methodological approach was similar to that of Miles and Huberman (1994) in a number of ways: all three forms of activity identified by Miles and Huberman (1994) were undertaken; following data management, some descriptive categories were created; and, following the synthesis of the data, explanatory accounts were developed. With regards to data management, though, Bebbington and Gray (2000), implicitly drawing on Yin (1989), follow a three-step approach: the authors firstly describe their research questions and explain what data need to be collected to address them: e.g. to investigate who appears to be educating corporations about sustainable development, the authors explain that this issue “is addressed by examining the various definitions of sustainability which are used by
- 30 corporations and by studying which organisations are influencing companies‟ understanding of sustainable development (Bebbington and Gray, 2000, p. 20). Then, some so-called „semiotic‟ categories are developed, based on the research questions and the literature, by identifying various sustainability definitions from major organisations that might have possibly influenced the examined companies‟ stance.
“an examination of the disclosures using these categories was then undertaken and an initial identification and classification of these disclosures attempted.
These initial categories, however did not prove sufficient for analysis… as a result, the categories were further refined drawing again from the relevant literature… [and] were also added inductively from the analysis of the environmental reports themselves” (ibid., pp. 21-20).
Thus, the data management approach adopted is in accord to both „pattern matching‟ (where “an empirical pattern is compared with a predicted pattern, following the theoretical propositions of the framework”) and „explanation building‟ (where the objective is to build a general explanation that fits each of the individual cases) qualitative analysis techniques, dictated by Yin (1989, pp. 108-109) for multi-case explanatory research designs. Evidently, following „pattern-matching”, by noticing that some data could not have been explained by their initial categories, Bebbington and Gray (2000) revisited those, and thus, modified their theory, in an attempt to explain all the data.
In contrast to the Miles and Huberman (1994), this approach employs originally theory-driven „pragmatic‟ codes (see also Unerman  for a similar approach), and thus, despite the subsequent attempts to explain all text, this still damages the method‟s qualitative orientation. Further, it should be noted that undertaking this analysis is time consuming, albeit considerably more time efficient than the Miles and Huberman approach. An additional limitation may be that this approach is not as easily quantified as that of Miles and Huberman, since usually a smaller number of codes is generated. Still, this approach is in accord to Holsti‟s (1969b) suggestion for “continual moving back and forth between data and theory” (p. 116) and even when not coupled with quantitative analysis, it can still provide the research with a
- 31 particularly useful qualitative perspective. Mainly for the reasons of being a more „pragmatic‟ and time-economic approach than that of Miles and Huberman, it was also employed in the BA study.
Conclusion The conclusions that may be drawn from this review are related to the discussed three areas of concern. Firstly, with regards to the index vs. amount/volume approaches: it seems that nowadays, with so much CSR information disclosed, volumetric studies may contribute more to the analysis than index ones. Index measures, however, do have some distinct advantages over volumetric ones in that, to the extent that the presence or absence of specific information is sought, particularly when “what is not disclosed…[is] seen as important as that which is” (Adams and Harte, 1998, p. 783), they are not significantly affected by contextual issues, such as repetition or grammar and since they further have clearly defined measurement units they are more reliable.
Secondly, with regards to the units of analysis: similarly to Unerman (2000), it should be noted that the selective use of information, relatively to both the sampling units (e.g. exclusive use of Annual Reports) and the recording units (e.g. exclusive consideration of narrative CSD) limit significantly the validity of the findings.
Further, with regards to volumetric recording units, it seems that at least in theory, page size data are superior than proportion of page data (particularly considering that the former count the narrative information in some textual unit whilst the latter measure it in essence, in terms of square centimetres). Future research should focus on monitoring the reliability and validity particularly of these two measures, given that they seem to be the only ones which can consider non-narrative CSD in the analysis. This however in view of the fact that tables, graphics, pictures and text are all different things and to attempt to develop an aggregate measure for all of these is highly subjective. Overall, it seems that the main causes for measurement errors and general discrepancies in validity and reliability of the studies are the misspecifications of the context units, which may be partially addressed by the employment of more meaningful approaches to definitions.
- 32 Thirdly, with regards to the ways to define CA units: it seems that traditional distinctions, such as monetary vs. declarative or even positive vs. negative need to be complemented with more meaningful approaches. These can either take the form of explicitly normative templates that could be included in a CA protocol and be quantified, such as the symbolic vs. substantive distinction, or take a more qualitative form, of a varying degree of structure, such as the two reviewed thematic approaches.
In either case, when conducting qualitative CA analysis it is equally important to follow a clearly justified and specified approach to the analysis, to enhance both validity and reliability of the findings.
Three main potential fruitful venues for further research may be identified. Firstly, a review focusing on the qualitative approaches subscribing to the broader CA view, including the distinct systematic ones, such as grounded theory and discourse analysis, could assist in clarifying their relationships and their potential contribution to CSR. Secondly, a review of contemporary CA studies in other fields and consideration of potential applicability to the CSR practice, as in e.g. the cases of Lasswell et al. (1949), Stone et al. (1966) and Gerbner et al. (1969), could save the field from reinventing wheels (Owen, 2004). Thirdly, a similar review of computer software (such as NUD*IST, as presented by Beattie et al ) in CA would be also of particular interest. With regards to the latter, it should be noted that qualitative analysis programmes do not only assist in the analysis in terms of speed in coding but also bring in a number of validity benefits. Cross-code analyses can be conducted;