[1] The data used here are more solid than earlier Forest Inventories, the deficiencies of which are discussed at length in Sorani and Alvarez (1995). Two main advantages are the higher resolution of the underlying satellite images and the extensive ground-truthing (with a total of 22,000 sample plots -each 100*100 m in size) undertaken prior to the interpretation of these images.
[2] Our aggregate figures may differ from those published in the Inventario Nacional Forestal (1994) because (i) these figures are based on a higher resolution (1:250 000 rather than 1:500 000 as used here); (ii) several revisions have been undertaken since the publication of this volume.
[3] While such aggregation undoubtedly introduces some bias, the large number of municipios (681) in Chiapas and Oaxaca suggests that the associated problems may not be too severe.
[4] As explained in the text, this column indicates whether the variable is available at the plot level (i.e. derived from digital maps) or at the municipio level (i.e. derived from Census information).
[5] Use of friction-weighted distances, i.e. distances that were adjusted for expected length of travel time based on characteristics (steepness and vegetation cover) of the terrain as in Chomitz and Gray (1996) yielded very similar results and is therefore not reported separately.
[6] To capture possible interactions between poverty and distance to infrastructure, we add an interaction term between these two variables.
[7] Instruments are the same as for the population density equation plus the percentage of the population without social services (water, drainage, electricity), percentage of the population without primary education, and the percentage of area under irrigation. Instrumental equations are discussed in more detail in Deininger and Minten (1996).
[8] At the plot-level, irrigation would, of course, be associated with lower probability of forest cover but this is irrelevant as no irrigated forest exists.
[9] We do not report results from the instrumental regressions (for population density and poverty) but note that these have generally very high predictive power (R2s between 0.6 and 0.8). To make sure that the use of poverty as a proxy for the wage rate does not bias the results, we used the average level of income in the municipio as an alternative variable. Results were very similar and are therefore not reported here.
[10] As the poor are located predominantly in marginal areas where unfavorable physio-geographic conditions severely limit the benefits from deforestation, failure to include physical endowments would imply that the regression captures a spurious (negative) correlation between poverty and levels of deforestation.
[11] Careful examination of the data available from the Digital Chart of the World, which are used in much of the existing plot-level literature, led us to the conclusion that, at least in the case of Mexico, the errors associated with these data are too large to derive any substantive conclusions.
[12] While this seems complicated in abstract, it seems that, at least for Mexico, such a procedure could be implemented without major difficulties. Parametrization of a biodiversity function could rely on the more than 20,000 observations on variety of species, complexity, and degree of degradation of the ecosystem that have been assembled in the process of ground-truthing the satellite images on land use utilized in our regressions. Data on foregone profits from agricultural production could be generated based on farm-budget data that have been collected, on a nation-wide scale, in the course of preparations for PROCAMPO. The computationally most difficult part is probably that protection cost, as well as the value for biodiversity, are likely to be influenced by the contiguity of protected areas, a consideration that could be introduced by introducing a parametric penalty if two protected points are not contiguous.
[13] The reason for choosing a 60 km step-width was that -as illustrated in table 6- in each of the intervals thus created (<60 km, 60-120 km, 120-180 km, 180-240 km, and > 240 km) the number of points that were actually protected is about equal, thus minimizing the danger of drawing spurious conclusions.
[14] Note that, with more than 600 municipios our data are still more specific than province-level data that have been used in a number of applications (Panayotou and Sungsuwan 1994, Kummer and Sham 1994, Osgood 1994).