Global literature review - Methodology

The research on which this paper is based began with a hypothesis derived from earlier research in Bangladesh by the authors of this paper (insert Belton, Thilsted, Hossain refs) that the dried fish literature diverged from the broader fisheries literature in three ways. First, the dried fish literature puts relatively greater weight on downstream segments of the value chain. Second, the dried fish literature has a greater 'technical' orientation where natural science and engineering methods of investigation are relatively predominant. Third, unlike the broader fisheries literature, the dried fish sector has suffered from a relative paucity of attention to social science, governance, and ecological matters.

Dried Fish Matters' project design has sought to build systematically on the observations of Belton and colleagues. It is deliberately positioned to address the gaps we hypothesized to exist in research on dried fish by bringing careful social science attention to the subject. The project retains an applied interest in the dominant food safety, productivity, and nutritional applied concerns that inspire the existing dried fish literature, but seeks to situate them in relation to broader human wellbeing considerations, including attention to environmental impacts, consumption disruptions, and the social relational shortcomings that are exhibited in the sector. Dried Fish Matters (DFM) has characterized this effort to broaden the scope of attention to dried fish as moving towards a social economy of the subject where value chains serve as a tool for spatially organizing interdisciplinary analysis of the dried fish sector.

The literature review was pursued using two primary tools: Google Scholar and Zotero. Google Scholar was our search tool and Zotero our reference manager. Both tools have strengths and limitations. Google Scholar is a proprietary web-based software that permits complex searches of all reference materials available through the World Wide Web. It has an enormous reach and is a powerful tool for collating digitally available materials that can be ranked by their impact measured by number of citations. Its limitations include the obvious inability to locate materials that exist only in hard copy form, but also its opaque algorithm and the very volume of results it generates. These result in often laborious parsing of results to weed out irrelevant items. Zotero provides a collaborative platform for collating, organizing, annotating, and tagging search results for further analysis. Zotero was invaluable for coordinating shared work tasks and iteratively refining categorizations of the literature. While its core functions are not difficult to learn, Zotero is nonetheless sufficiently complex that it poses a barrier for some potential users.

Our literature search proceeded following an iterative learning strategy (references here) where we developed search terms, conducted searches, analyzed results, refined search terms, and conducted further searches to address insights and gaps in the previous search round. Initially, we limited our search to the six focus of countries of the Dried Fish Matters project: India, Sri Lanka, Bangladesh, Myanmar, Thailand, and Cambodia. We systematically conducted searches using country search terms in conjunction with a first list of variants of the term dried fish and a preliminary list of thematic search terms of interest to the DFM project (Appendix X)[DJ1] [ET2] . At this stage we saved references to Zotero if they were either very clearly focused on dried fish (particularly with some relation to the value chain) regardless of regional focus (i.e. not the region used as the keyword/not part of the focus country group), or provided some discussion of dried fish in one of the focus regions. All sources were tagged in Zotero with the search terms used to find them[ET3] . Search term Zotero tags were marked off from other tags using a particular notation so as not to confuse search and analysis stages of the work. Notably, many of the most relevant sources were a few (sometimes 10 or more) pages in on Google Scholar[DJ4] . “Value chains” proved to be often an especially effective search term, yielding results from Africa, Asia, and Bangladesh in particular. Search terms of academic disciplines tended to overly narrow the search and generate few highly specific results.

The literature search had two further major stages. The first was an expansion of dried fish related terms to include “fermented fish” (N=382), “smoked fish” (N=383), and “cured fish” (N=328). This added search term set produced some new sources, but not as many as “dried fish” (N=657) and “dry fish[DJ5] [ET6] ” (N=449). [DJ7] [ET8] [DJ9] [ET10]

The second, and larger, expansion of the literature search was the decision to expand the literature search to the global level. We chose to take this step, as it had become apparent that our analysis would be of greatest interest if we could speak about the contours of the entire dried fish literature, not just that of a sub-set of Asian countries. We therefore added search terms for every country in the world using the UN list of global countries (add URL). We also added search terms for all continents except Antarctica, sub-continental geographical areas, and some provinces and states in order cast our search net as widely as possible. Our final list of spatial search terms is listed in Appendix X. We then conducted searches in Google Scholar that combined each new geographical search term with all existing thematic and dried fish product search terms.

As we moved through the search stages of the literature review, we simultaneously developed and refined a logically organized a list of tags for coding the literature review references in Zotero. We used a range of prefixes to further distinguish these tags from the search terms. There is broad resemblance between the search terms and the tags, but the ability to clearly distinguish them is important due to their different functions of (1) search and retrieval from the public repositories and (2) analysis. Tagging takes the rough categories of the search and further refines them through careful readings of each reference. Through tagging, we developed more precise analytical distinctions and also discovered inconsistencies and mistakes in the automatically applied search term tags.

This searching/tagging distinction can be illustrated with reference to geography. Often Google Scholar searches generated mistaken geographical identifications for thematically relevant references. In the tagging phase of analysis, we first confirmed the accuracy of all geographical identifications by adding a new tag in the form of *(place). Often we observed locational inaccuracies (sometimes wildly so) in the search term in which case the newly added tag served as correction. Sometimes, also, in tagging we found that a reference also concerned one or more other places or geographical scales (such as, e.g., *Southeast Asia) that were not picked up by the search, which we added.

Through multiple rounds of review and discussion, our tag list was refined into five broad dimensions: (1) direct versus indirect, (2) geography (3) theme, (4) value chain segment, and (5) product. Our tags, by dimension, are listed in Appendix X. Definitions of the most important tags for our analysis are given in Appendix X. For each of these tag categories except (1) direct versus indirect, multiple tags from the same category may be applied to a single reference. Tags for categories (2) through (5), therefore, are not mutually exclusive. This point is important substantively for segment and product. Economic actors or organizations may be involved in more than one value chain segment and dried fish products often involve several different processing techniques, such as salting and fermenting.

After the process of removing references that were either incomplete, turned out to have no substantive connection to dried fish, or had no relevance to understanding the broader context of dried fish social economies, our literature generated a total of 3105 items. Of these, 1405 were direct references on dried fish and 1703 were references of indirect relevance to dried fish[ET11] .

[DJ1]Can we, should we, organize the search term list in the appendix by search round?

[ET2]These can be separate online documents. We can generate a zip folder, for example, containing the lists of search terms, the tags and commands used in generating queries, etc. This doesn't really need to be in print.

[ET3]There does need to be an explicit statement that we used a shared Zotero library, and an indication of the rationale for this choice. We can also address the strengths and limitations of Zotero in the discussion/conclusions as we describe what might happen with this database in future.

[DJ4]I think this is an important methodological observation about the buried nature of some of the most important references, but I am not totally sure if I have the meaning right. Jonah, please see if you can clarify this sentence.

[DJ5]Is there any way to roughly quantify the scale of difference retroactively? “Some” here is vaguer than I would prefer.

[ET6]We still have the tags. Added the numbers here.

[DJ7]We did not search on fish paste, fish powder, pickled fish, which means that these references may be more under-represented than they otherwise would be.

[ET8]ALSO: Let's decide what to do with the patents. We have included some and discarded others. I would vote to keep them in the Zotero library but not include them in the literature review. They amount to 10% of the general fisheries sample, which skews the results -- but that reflects Google Scholar's inclusion of a patent search rather than their importance within scholarly discourse.

[DJ9]A note on why we switched would be good here.

[ET10]Mention the limitations of Google Scholar. The search results ranking algorithm is proprietary and results are not always accurate. We retained the keywords as tags in the Zotero database, however they often concerned irrelevant resources (which were deleted) or related to dried fish but surfaced from a keyword search on an unrelated topic or country (e.g., “Myanmar” appears incidentally in the text of a document on fish in Cambodia, so the search term is [“dried fish” Myanmar]). This is relevant to our decision to use prefixed codes in the database.

[ET11]UPDATE numbers