Data issues

• Data sources and definitions
• Data requirements and predicted effects

Data sources and definitions

We combine high resolution digital maps of land use and soils with municipio level Census information from the 1990 Population Census and the 1991 Agricultural Census for the states of Chiapas and Oaxaca. The information from digital maps has been used to determine properties of a sample of approximately 160,000 points obtained by applying a 1x1 km grid to the states in question. Municipio-level variables, by contrast, are available as muncipio aggregates for each of the 684 municipios in the two states studied.

Digital data on actual land use, derived from detailed satellite imagery and, based on extensive ground-truthing,[1] interpreted by a team from UNAM (Universidad Autonoma de Mexico), allow to distinguish cultivable land, forest (classified into several distinct tree species), and agricultural land -again subdivided into rainfed irrigated, pasture, and plantations.[2] Comparison of these data, most of which refer to the early 1990s, to the forest cover in 1980 does, furthermore, facilitate identification of areas that had been deforested during this decade.

Information on soil quality and precipitation is derived from a digital soil map based on the FAO classification which was converted into soil fertility limitations using the Fertility Classification System (Sanchez, Cuoto and Buol 1981). Physio-geographic data on elevation and slope were similarly constructed from a 30’ digital elevation map of Mexico available from the US Government.

Distance to infrastructure was calculated as the straight line distance to the nearest road for each point under consideration. To avoid the deficiencies associated with data on infrastructure available from the Digital Chart of the World, we digitized the road network for the two states from local road maps (1:800,000 scale).

While use of physio-geographic variables has become standard in the empirical literature on deforestation, socio-economic variables could, according to the model above, have an independent effect on decisions concerning natural resource use. To explore this possibility, we use Census data to provide information on population density, land tenure (percentage of area under ejido-tenure), poverty (the number of people with monetary incomes of one minimum wage or less), the percentage of the population that is indigenous, the endowment with productivity-enhancing infrastructure (percentage of area irrigated), and the availability of government-sponsored services such as extension and the number of (subsidized) credit transactions for agricultural production.

While socio-economic data are available only at the municipio-level[3], they allow not only to evaluate controversial hypotheses regarding the effect of socio-economic variables on deforestation but also to test whether similar results would have been obtained by using only subsets of the independent variables, an issue that is of particular importance as many empirical studies implicitly or explicitly omit the plot-specific physio-geographic information.

Data requirements and predicted effects

Following the rent model described above, land use at any point will be a function of productivity shifters and local prices. The variables used as proxies for factors affecting productivity, and the costs and benefits of different types of land use (i.e. agriculture or forestry) are summarized in Table 1. Plus- or minus signs in the last column indicate the expected effect on finding forest on any given plot.

Table 1: List of variables included in the estimation of forest cover in Mexico

Our data contain three classes of plot-specific variables. First, physio-geographic characteristics such as slope, elevation, soil fertility limitations, rainfall and squared rainfall are expected to be important productivity-shifters that would have a negative (positive) sign insofar as they limit (increase) agricultural productivity on any given plot. Second, input and output prices are strongly affected by proximity to infrastructure, a variable captured in our data by the straight line distance to the closest road,[5] expected to be negatively related to the probability of agricultural use. Finally, we expect the location of a point within a protected area to decrease the expected benefits from agricultural cultivation, as location within a protected area will make it very difficult to establish rights for continued agricultural use of a given plot. The coefficient on protected areas would also allow us to make inferences regarding the effectiveness of protection, a topic to which we will return below.

The main socio-economic variables included in our analysis are poverty or the unskilled wage rate, population density, availability of subsidized credit, the presence of communal tenure systems, access to extension, and availability of irrigation. The unskilled wage rate, a variable that is almost synonymous to the level of poverty, is likely to be an important determinant of the form of cultivation. Given that agriculture (even extensive cattle ranching) is more labor intensive than forest use, one would expect higher wages to be associated with higher levels of forest cover and less agricultural land use. The reason is that benefits from agricultural production would no longer be sufficient to cover the cost of labor, in addition to a potential increase in the appreciation for the non-timber uses of forest that comes with higher levels of income. By contrast, where wages, i.e. the opportunity cost of labor, are low, one would expect lower levels of forest cover because poor people would be willing to utilize even relatively marginal areas to make a living (Sandler 1992, Amelung and Diehl 1992, Reardon and Vosti 1995, Angelsen 1995).[6] As we do not have data on unskilled wage rates, we use instrumented "poverty", i.e. the percentage of economically active individuals earning less than one minimum wage, as a proxy variable.[7]

While population density has figured importantly in the literature, it is easy to confound the effect of this variable with other parameters (e.g. access to infrastructure, soil fertility, and availability of physical capital such as irrigation) with which it is normally highly correlated. To avoid such misinterpretation, we are careful to include whatever municipio-level variables are available and instrument population density, using access to social services and education at the municipio level as identifying instrument.

Another variable that would have an important impact on forest cover is the relative cost of capital in agriculture. Agricultural credit subsidies, equivalent to a direct reduction of the cost of agricultural capital that is at least intended to be non-fungible, would be expected to increase the optimal amount of land that is used for agricultural activities, thus being associated with lower forest cover. The variable used to approximate this parameter is the number of farmers in any given municipio who received, in the reference period for the 1991 Agricultural Census, (subsidized) credit from the state bank BANRURAL.

There is a common belief that traditional forms of tenure would be associated with a "tragedy of the commons" whereby the absence of well-defined property rights would reduce the benefits to be obtained from continued agricultural cultivation (R*) and therefore induce individuals to deforest in order to reap a "one-off" gain even in cases where continued forest use would be economically optimal. Availability of irrigation would shift productivity on existing agricultural areas upwards and, to the degree that this can substitute for deforestation, be associated with higher levels of forest cover.[8] A similar (but somewhat weaker) argument can be made with respect to agricultural extension which -if it does not induce encroachment on forested areas- would be associated with higher levels of forest cover.

Before discussing empirical results, we briefly provide descriptive statistics emerging from our data-set (Table 2). Compared to Mexico as a whole, the study area is characterized by below average population density (38 persons per km² of cultivable area, compared to 68 at the national average), a high rate of deforestation (3.6% annually) and low levels of plantation forest (0.07% as compared to 0.56% at the national level). The share of the population making its living from agriculture is, with 56% compared to 22% at the national average, high, despite a very unfavorable physical environment. Harsh physical conditions are illustrated by the prevalence of sloped and steep land (48% and 3% versus 26% and 1% nation-wide, respectively), high incidence of soil fertility limitations, and a low share of irrigated land (only slightly above 10% compared to a national average of 31%). The share of households who do not have access to water, drainage and electricity is more than double the national average, as is the percentage of illiteracy. All of these factors are reflected in low average levels of income (0.68 minimum salaries compared to 1.31 minimum salaries at the national average) and a high incidence of poverty, with about 59% of the economically active population living on less than one minimum salary.