IV. Methodology A content analysis especially well suited for journalism The news articles produced and distributed by members of the Northern News Service are the data-set upon which this analysis is based. Although a few stories were distributed late in 1988 and distribution continues into 1993, this study is restricted to stories from the years 1989-1992. During that period, a total of 296 articles were distributed in a series of monthly "bulletins" collated, translated and distributed by APN in Moscow. Of that total, 242 are available in English for use in this survey. This is not a sample, but rather all the English-language versions available in Alaska and Moscow. This total represents more than 80 percent of all the articles produced and is relatively evenly distributed over the years of NNS operation, containing 65 from 1989, 76 from 1990, 49 from 1991, and 52 from 1992. Nothing suggests that this collection of articles deviates from a purely random selection, and the articles used are in their original form, not as they were published (and potentially edited) in a member newspaper. Thus the data-set fully represents the material editors received from the Northern News Service and is appropriate for developing inferences about the service as a whole. The quantitative analysis of these articles is done with a computer-assisted system originally used in evolutionary biology and based on Q-analysis, or polyhedral dynamics. This process essentially treats the news stories as multi-dimensional structures for the purpose of analysis, and yields far more realistic content classifications and richer comparisons than most conventional taxonomic approaches, which can be characterized as "traditional partitional" analysis. (See Gould, et al. 1984: pp S1-S97). The ability of modern computers to process the Q-analysis algorithm can generate results far more reflective of real-world patterns than the necessarily more confined constructs resulting from most previous forms of content analysis. This description from Gould, et al. explains the key advantage enjoyed by this system: In the days when information had to be written down and laboriously looked up in card files and catalogues, it was quite understandable that simplifying partitional schemes called classifications were necessary and useful, even if you had to simplify things rather artificially, and so pay the price of losing some information in order to make awkward things fit in one box or another -- things that sometimes wanted to live between the boxes. Given the limitations of the human brain to collate ... and given the time it took to look things up, there simply was not much in the way of alternatives. You really had to simplify. But another problem arose when you forgot that you had simplified complexity, and began to believe that the simplifying and damaging constructs you had imposed were the actual reality 'out there' ... But the technological matrix in which we live has now changed radically. What happens now that we have machines that can store and collate enormous amounts of information in the proverbial twinkling of an eye? "The unfortunate answer: ... we invariably ignore the potentialities opened up by the advent of such machines and go merrily on doing -- that is, thinking -- exactly as we did before. ... Whereas in the past days we hacked the world up into boxes with a rather slow cleaver, we now do exactly the same things with a fast computer, never stopping to think whether we should be hacking things up in the first place ... "It is our contention that the computers we have today ... release us from old constraints and so allow us to think new thoughts about the whole question. (Gould, et al. 1984: S39-S40) Those "new thoughts about the whole question" have resulted in a technique that avoids many of the problems associated with the traditional taxonomic approaches. In a standard text on content analysis, Klaus Krippendorff has noted that, "Unfortunately, most practical uses of systems notions in content analysis have been marred by their still-archaic forms. The study of trends has often been concerned with only one variable at a time ... Patterns have often been conceptualized in binary relation ... and differences have rarely been explained in terms of interactions that may enhance or diminish them in time. One problem is the sheer volume of data required to find characteristics that are sufficiently invariant to warrant systemic inferences" (Krippendorff, 1980: 22). The system of "combinatorial analysis" used in this project specifically addresses those deficiencies. The analysis not only allows but encourages consideration of multiple variables simultaneously, as the results detailed below illustrate. This multivariate capacity seems especially appropriate in looking at a data-set of news stories, which by their very nature involve widely varied subjects and are almost never unitary in focus. Prof. Graham Chapman, who wrote the software used in this analysis and supervised its application, noted in a similar analysis of television content data that "quite often, there is more than one dimension implied by the question, so that we would need more than one classificatory hierarchy" (Chapman, 1982: 8). Indeed, this analytical technique allows precisely the kind of examination in which scholars look less for "the sort of thing that connects planets and pendulums" and more for "the sort of thing that connects chrysanthemums and swords" (Quoted in Pfaffenberger, 1988: 12). Finding connections is a key strength of this analytical technique. "... [T]he fact that people carve up the world into disconnected pieces seems to reflect the roots of our education in the days before computers could be used to sort out the connections at the 'edges' of the sets" (Gould, et al., 1984: S60). One key aspect of this system is that it relies on classifications that originate from the needs of the researcher and the data-set itself rather than using a set imposed from outside. At first glance, this seems at odds with the typical impulse to arrange things into standardized classifications, as in a library. (See Roberts, 1960). But this apparent conflict is easily resolved by realizing the different purposes of the two classification schemes. A library is dedicated to providing a wide variety of people of different interests and backgrounds with a common language with which to organize and retrieve information. As Roberts said, "Classification is an art ... The aim is not to specify the contents of a particular document, but to be able to produce the relevant documents, from whatever source they may have come..." (Roberts, 1960: 4). But the purpose of classification in this analysis precisely is "to specify the contents of a particular document" -- and to do so with subtlety, and in relation to all the other documents involved. Thus the use of a classification scheme developed by the researcher to meet precisely the needs of this study and to match precisely the attributes of this data is appropriate. This method of combinatorial analysis yielded the customized classifications detailed below. These grow from the data, and are constantly checked against it and against a "usefulness test" to determine whether they are meeting the demands of the research. "This research strategy follows a definite pattern or sequence, with 'feedback' loops at those points in the analysis where we pose the question 'Do these results make sense?' We feel it is very important to enter a new area of research with as open a mind as possible, and not to approach a new area of experience with a rigid a priori scheme" (Gould et al, op. cit.: S64). The definition procedure is illustrated above. After reading and writing characterizations for the 242 news stories in this study,18 basic subject categories were established: communications; culture; demographics; education; economics; energy; environment; extractive resources; exchange/visit; health; indigenous peoples; language; pollution; politics; renewable resources; science and technology; transportation; and wildlife. (Definitions of subjects included in these second-level classification categories may be found in Appendix D.) By way of illustration, note that a classification like "exchange/visit" would never appear in any standardized list of classifications, but that in a study specifically focused on news shared between eight different nations, such a classification becomes a very useful category, indeed. In addition, this study recognizes that in media analysis one must consider not only the subject matter, but also the treatment of stories -- that is, whether the subject was treated as a news story, a commentary, a biographic profile, or what-have-you. This distinction between subject matter and treatment -- and the ability to obtain results that take this important distinction into consideration -- provides a far more sophisticated and realistic result than standard partitional content analysis. The forms of treatment used in this analysis include news story; feature story; sports story; profile; history; commentary; and ofitsioz, a category describing a kind of "official news" prevalent in traditional Soviet newspapers, detailed in the following section. (Treatment categories are detailed in Appendix E). The use of only a single individual for initial coding and decisions about aggregation of data in this analysis eliminated the necessity for detailed coding instructions to ensure standard treatment by different coders. The first stage involved reading the 242 stories and describing each with descriptive phrases about subject matter, written in plain English with no attempt to force them into predetermined categories. Each story was also assigned to one or more treatment categories, and was labeled by date, author, author's affiliation, story size, country of origin, locale of subject, and story title. Later, a "dictionary" of these subject matter descriptors was produced and each was aggregated into one or several of the 18 second-level categories outlined above. It is critical to note that none of this coding or aggregation seeks to identify exclusive categories for the data. In fact, the opposite is true. Since each of the articles could and often did include more than a single subject, each was given as many descriptive phrases as necessary, and these could easily be aggregated into more than one second-level category. [The figure below illustrates a sample aggregation process.] Likewise, those stories that transcended treatment classifications -- by involving, say, both history and commentary -- were likewise coded to reflect that. The software that processed this data then provided a series of analytical print-outs revealing the scope of the data; more than 1,300 lines of subject matter/treatment combinations resulted from the overview analysis alone. A sample of this printout is attached as Appendix F. Here the strength of the combinatorial process becomes evident, for it is in illuminating combinations, patterns and connections that the richness of the data emerges. For example, rather than simply noting that X percentage of the articles deal with wildlife, this combinatorial analysis can show what percentage of articles dealing with wildlife also involved indigenous people and politics. Beyond that, we see how many of those were treated as commentaries, and how many as news. We can also analyze the wildlife-indigenous-politics stories by country of origin, or by date, and so forth. Examples of how that analysis of subsets works is demonstrated in the results in the following section.