Identification of extended chains necessitated marking the section boundaries in the texts. For better comparability of the results, the texts were divided into five main sections (Abstract, Introduction and Review of the Literature, Methods, Results and Discussion, Summary and Conclusions), which were almost always present in the analyzed texts in both the RA and the MA thesis corpora, though sometimes under different titles. In the majority of the RAs the Introduction and the Theoretical background (in theses: Review of the literature) were one and the same section, in the Figures below they will appear as one section. Having identified the section boundaries, new chains for each section were counted from the column of the first new presupposed item which appeared in it to the last one (the number of columns thus counted are the new chains), whereas chains whose presupposed items were found in a preceding section were counted as extended.

Figure 20 shows the normalized proportions of new and extended chains in parallel according to the main sections in the two corpora of 20 RAs and 20 MA theses. Apparently, RAs introduce a greater number of chains in the abstract compared to MA theses. This means that RA abstracts contain more presupposed items of what will become reference chains as the text proceeds. What makes this result less meaningful is that – as shown in further discussions in this section related to abstracts – this difference is probably due to the fact that MA abstracts are not connected by referential but by lexical cohesion to the subsequent introduction. That is, they tend to stand more apart from the rest of the paper than they do in RAs. Comparing the high and low rated corpus, the same difference is found; high rated theses contain twice as many new chains in the abstract than low rated ones.

The first pedagogically relevant difference to be observe here concerns the Methods section. Even though MA theses have double the space to explain the Methods of analysis, and that they had about five times as much space to establish a theoretical background, they do not rely on that background so much in describing the methods. The lack of extended chains (on average there are only 5 extended chains in a thesis) points to fewer links in between the methodology and the theoretical background. There may be many reasons behind this. One is that the theory cited in the paper is not directly relevant to the research methods of the paper. Another is that the theoretical foundations of the paper followed a chronological description of the development of a certain field, which needs a proportionately greater number of new chains, and extended links will mainly relate to the final sections before the methodology section. When a literature review centres around the main concepts directly relevant for understanding the discussion in the paper or cites theory closely related to the particular methodology applied in the paper (as is the case in RAs), there will be more extended chains. Interestingly, high rated thesis contained more extended chains (5.9 on average) than low rated ones (4.6).

The Findings section is approximately twice as long in theses as in RAs; nevertheless, Figure 20 shows that even the normalized data reflects a major difference between the two corpora here in terms of new chains. The comparatively high proportion of new chains in MA theses is quite unexpected and is largely due to the high number of new chains in the low rated corpus. This has two relatively straightforward explanations for studying the RCA tables in detail, both contributing to the final results. First, RAs introduce the main referents (with the first presupposed items of long cohesive chains) as early as in the Introduction, or in the Methodology section. These main referents include a detailed introduction of the participants of the research, the research methodology (instruments, key questions and concepts), the setting and the research procedures. As Jacoby et al. (1995) also write, expert writers link results to hypotheses and research questions posed in the introductions, they compare results with previously cited research and draw general conclusions from details discussed. The early introduction of these main referents and consistent reference to them establishes long chains of reference with a steadily increasing number of extended chains. MA theses, especially low rated theses lack a specific introduction of these referents; for example, if the participants are students, the reader does not find out the exact number, age, gender, class, groups, etc., of students up until the Findings section. As a result, the specific presupposed items for participants surface first only in the discussion of the findings. Secondly, it is observable that the Findings sections of RAs have a plain, distilled structure: they consist of a logically related list of several conclusive statements, each supported by results from the research process. In contrast, MA theses have a more extensive discussion section preceding each of the findings. These discussions involve an abundance of cohesive ties that do not encompass text segments longer than three sentences, which is the other explanation for the high density of new cohesive ties.

The last section examined in the corpuses is the Summary (or Conclusions) section. Here, the MA theses contain fewer chains whether extended or new. In research articles a steady increase in the number of extended chains can be observed towards the end of the article, and as we have already observed (in 6.6 in this book), the greatest proportion of extended chains is found in the Summary or Conclusions section. This is not surprising; the Conclusion must be grounded in the previous sections; new chains usually contain reference to future directions of research or reference to practical implications. Indeed, the scarcity of new chains in MA theses as compared to RAs is the result of much briefer discussion of such information in theses.

Overall, RAs contain a steadily increasing proportion of extended chains section by section, building up a firm structure by introducing referents with fully detailed presupposed items as early as possible. MA theses display a slightly different pattern, which is due to a number of factors. As mentioned before, one difference is the introduction of presupposed items with some delay. In addition, there are more short ties in theses (in RAs, 39% of the total number of chains are long, while – despite the greater length of MA theses, long chains make up 34.5% of the total). Arguably, the arbitrary distinction between long and short chains influences these results; if the minimum criterion for a long chain was shifted to 4 or 5 ties, it would produce slightly different results. Nevertheless, the tendencies observed in the frequencies of new and extended chains already suggest that it is the Methods and Results sections that need a more in-depth investigation.