We began the task of describing the methodological and substantive components of the literature on tropical deforestation by building an inclusive bibliography on the subject. Beginning with a list of 120 citations that we had collected during the early 1990s, we used three strategies to update and expand our bibliography. First, we mined the bibliographies of well known published works on tropical deforestation. Second, we searched on-line bibliographies in the natural and social sciences, using the key words ‘tropical deforestation’ and ‘deforestation’. Third, we exchanged bibliographies with other researchers engaged in meta-analyses of the tropical deforestation literature. The creation of the bibliography has been a rolling process. Reading newly discovered articles leads to the discovery of more, as yet unread articles. While the numbers of new discoveries has declined as we compiled the bibliography, we continued to make additions to the bibliography until several weeks ago. Our efforts at this stage of the work have been inclusive; we frequently included works whose value could only be determined at a later stage by actually reading the report. Over the six month period of work we accumulated more than 1250 references on tropical deforestation and listed them in ENDNOTE, a computerized bibliographic software system.
It is important to be aware of the limitations inherent in this effort. We have undoubtedly missed a number of ‘grey area’ reports on tropical deforestation. Authored by personnel in local non-governmental organizations or government agencies, these reports describe land use trends or deforestation in particular locations in the tropical biome. The reports never get published in internationally recognized journals, and they circulate among a largely local audience. Because we have relied on electronic bibliographies which only catalog articles published in scientific journals and on the bibliographies in these articles themselves, our search strategy misses most of these reports. When we do find references to these reports, they are difficult to obtain because libraries do not acquire them which makes it impossible to obtain the reports through interlibrary loan systems. Although we read articles in four languages, English, French, Spanish, and Portuguese, the overwhelming number of articles included in the bibliography are in English. The predominance of English language articles in the bibliography undoubtedly reflects to some degree the prevalence of English as the international language of science, but it may also reflect a lack of access to the relevant literatures, particularly having to do with Francophone Africa.
We adopted a particular protocol in reading the articles. A large number of articles concerned tropical rain forests, but only mentioned tropical deforestation in passing. These articles were removed from the list. Other citations turned out to be earlier, unpublished versions of articles that we already had in the bibliography. We dropped these articles from our list. Some articles looked germane, but they proved impossible to obtain, either from the Rutgers libraries or through interlibrary loan. On the premise that it made no sense to list publications or reports which no one could obtain, we dropped these articles from our list. Through these procedures we removed more than 300 articles from our original bibliography. An additional 100 articles have proved difficult to obtain, but they appear to be substantively useful contributions to our understanding of tropical deforestation, so we will continue to try and acquire these articles in the coming months. The remaining 825 articles which we have read and coded form the basis for the analyses reported below.
Once we decided to include an item in our list and obtained it, we coded it for its contents. Appendix A contains the coding scheme that we applied to each of these items. The scheme contains two entries for location. One, entitled ‘geographical’ uses the standard FAO codes to identify the country, subcontinent, and continent described in a reference. A second entry, entitled ‘geographical scale’, records the geographical scale of the deforestation processes described in the reference. Did it concern a single community, a region within a country, a country, a cluster of countries, continent or all of the tropical places in the world. We also coded the references on a historical dimension. In which decade do the events described in a reference occur? We coded this item so that events could occur in just one decade or in a series of decades. We also coded items for the types of information that they used to draw conclusions about deforestation. Our coding scheme distinguishes between six different sources of information: (1) remote sensing/aerial photography, (2) household surveys, (3) key informant interviewing in a deforesting place, (4) key informant interviewing elsewhere, (5) direct observation of deforestation, and (6) secondary sources. Our coding scheme also records whether or not a study offered quantitative estimates of forest cover change (7) or loss (8). If an article used several different types of information, we included all of them on our coding sheets.
Because the ‘secondary sources’ category included diverse types of information, it requires some additional discussion. Items which rely exclusively on other reports about tropical deforestation to draw their conclusions would be coded as relying on secondary sources. This report, for instance, would be coded as relying on only ‘secondary sources’ because it did not involve the collection of any primary data in or from a rain-forest region. An article which presents a new model of tropical deforestation and uses already collected census and forest cover data to test the model would also be coded as relying on ‘secondary sources’. In contrast an article based on the collection of original data through a household survey administered by the authors or through field observations by the authors would be coded in categories 2 and 5. Heretofore unanalyzed remote sensing images which the authors analyzed in order to calculate a deforestation rate or forest cover estimates for a region would be coded in category 1; an article which relies on deforestation rates calculated in another report to arrive at its conclusions would be categorized as based on secondary sources, category 6.
Finally, we categorized each reference by the causes of deforestation that it cites. Based on an extensive review of the literature we created an inclusive typology of deforestation’s causes with twenty different causes. They are listed in Appendix A. A reference could cite one or multiple causes; all of them would be listed. In fact the number of cited causes of deforestation ranged from one to fourteen. To insure that we accurately conveyed the substance of the authors’ arguments, a cause had to be described and an analysis developed around it in the body of an article before we would list it. A mere description of a cause in an abstract or in the introduction to an article would not suffice to get it listed. Despite these strictures, the choice of terminology by authors, rather than differences in the conditions they describe, undoubtedly influenced the ways in which we coded some items. One analyst might cite a colonization program as the primary cause of deforestation in a location and say little about how the construction of roads, as part of the program, accelerated land clearing in a region. A second author, looking at the same situation, might feature road construction as a primary cause for deforestation in this region. Under these circumstances readers might code the two items describing the same situation quite differently.
The potential for this sort of error increases when you have to use multiple coders as we have in this analysis. Six different individuals, all of them graduate or undergraduate students at Rutgers with an interest in human ecology, have read and coded the items in the bibliography. To reduce unreliability in the coding, we conducted short training sessions for each coder to insure that they understood how they were supposed to code the items that they read. To measure the magnitude of the remaining error, we carried out an exercise in inter-coder reliability in which two coders read the same items independently, and then we calculate the degree of agreement between them. Intercoder reliability on information sources was quite high, 90%. The comparable score for the many causes of deforestation was lower, 64%; this score makes the use of these data in small samples somewhat problematic but does not prohibit their use in large samples.
Another potential source of error in the analysis stems from using the same term to describe different phenomena. This problem is quite evident in the treatment of population in the literature. Studies of deforestation processes in South Asia and Africa frequently refer to population increase as a cause of deforestation, and, when they do so, they usually point to high rates of natural increase among already densely settled populations of smallholders. Analysts in Latin America sometimes cite population increase as a source of deforestation, but here they are referring to the population increase which occurs with the in-migration of relatively small populations of settlers into a sparsely settled frontier region. These sorts of differences in meaning of terms seems to be an almost inevitable analytic cost in highly aggregated analyses. The offsetting benefits from this scale of analysis come in the form of patterns of information use and argument that only become apparent when we analyze the literature on tropical deforestation at a global scale of analysis.
The disc which accompanied this report contains the list of references that can be sorted by country. The remainder of this report describes the global scale analysis of the literature on deforestation. To carry out this analysis, we created a data set out of the coded ENDNOTE information and analyzed it using SPSS (Statistical Package for the Social Sciences) software for cross-tabulations. These analyses revealed the patterns of information use and substantive conclusions which we report below.
