Estimation of damage to human health due to forest burning in the Amazon

Abstract

Forest burning stands as a highly used practice in the Brazilian Amazon Forest. Burning trees is the cheapest way to expand agricultural frontiers in the Amazon region. Nevertheless, the smoke generated in this process can produce undesirable negative effects, in particular health-induced problems by the polluted air. This study aims to investigate whether these effects on the health of the region’s population are statistically significant, using as variables mortality and morbidity rates associated with respiratory diseases. The estimated morbidity model was used to estimate the health damage to those people who became ill only by the fire.

Appendix

1.1 Appendix 1 Descriptive statistics of the variables of the model

Listed below are the basic statistics for the variables used in the model. As the database was structured as panel data, a slightly more complex presentation was needed for the statistic description of the variables. Had the variables been described in a trivial manner, either by a time series or cross-section, information related to the data sample would be contained in the very first line, where the designation “overall” appears, for which statistics are calculated for the entire data set, and for which differences among groups are not considered. In this case, the moments are described by the means $\overline{\overline x}$ and $\overline{\overline y} ,$ in both i and t, for the dependent and independent variables and the total sum of the squares is given by:

$$S^{o}_{{xx}} = {\sum\limits_i {{\sum\limits_t {{\left( {x_{{it}} - \overline{\overline x} } \right)}{\left( {x_{{it}} - \overline{\overline x} } \right)}\prime \quad {\text{and}}\quad S^{o}_{{xy}} = } }} }{\sum\limits_i {{\sum\limits_t {{\left( {x_{{it}} - \overline{\overline x} } \right)}{\left( {y_{{it}} - \overline{\overline y} } \right)}} }} }$$

As the data structure of the model is presented in panel data form, further analysis is required. The term “within” describes the situation in which the means are broken down by group with regard to the entire sample. Thus, instead of $\overline{\overline x}$ and $\overline{\overline y} ,$ we have $\overline{x} _{i}$ and $\overline{y} _{i}$ considered only in t for each group i; and the sum of the squares and of the cross products are given by the following expressions:

$$S^{w}_{{xx}} = {\sum\limits_i {{\sum\limits_t {{\left( {x_{{it}} - \overline{x} } \right)}{\left( {x_{{it}} - \overline{x} _{{_{i} }} } \right)}\prime \quad {\text{and}}\quad S^{w}_{{xy}} = } }} }{\sum\limits_i {{\sum\limits_t {{\left( {x_{{it}} - \overline{x} _{{_{i} }} } \right)}{\left( {y_{{it}} - \overline{y} _{i} } \right)}} }} }$$

Lastly, definitions may be made considering the differences among groups. In this case, the term “between” is used. Here, the sum of the squares and the cross products may be expressed as follows:

$$S^{b}_{{xx}} = {\sum\limits_i {T{\left( {\overline{x} _{i} - \overline{\overline x} } \right)}{\left( {\overline{x} _{i} - \overline{\overline x} } \right)}\prime \quad {\text{and}}\quad } }S^{b}_{{xy}} {\sum\limits_i {T{\left( {\overline{x} _{i} - \overline{\overline x} } \right)}{\left( {\overline{y} _{i} - \overline{\overline y} } \right)}} }$$

Table 5 lists the basic statistics of the variables, taking into consideration what was outlined above. Naturally, the blank spaces for the means, with regard to the sample groups “within” and “between,” are there to save space. As there are 261 groups (districts), we would need the same number for these measurements.

Table 5 Descriptive statistics of the variables

1.2 Appendix 2. Database

Regarding the health data, in this study, we have used annual hospital treatment values, average in-patient treatment value, and the average in-patient treatment period related to respiratory illnesses (CID-10 X). The sample covered all districts in the Amazon region. The data also refer to the public health services, extracted from DATASUS, the Ministry of Health’s information service.

The estimation of the area burned used in this study was obtained from the so-called hot pixels. These hot pixels represent satellite-observed fire outbreak points. This information was processed by the Wood Hole Research Center.

With regard to the population data used in this study, a series for this variable had to be constructed. This was due to the fact that there were no annual data available for the districts, as this information is only available for those periods in which the Brazilian Institute of Geography and Statistics (IBGE) census was conducted. The procedure used in this case was to take the district population data for the 1991–1997 period and determine the remaining periods by interpolation based on a linear growth rate, determined by the district population values for the 2 years indicated above.

As information regarding district income was also not available for the period, the real lumber production value (logs), and the real agricultural production value (temporary planting) were tested as proxies for this variable. Another possible proxy for municipal income is the use of a cattle-raising activity index, which is directly related to the growth in the region’s cattle herd. All these data were constructed from information obtained from the Sistema IBGE de Recuperacão Automatica (SIDRA) system of the Agricultural Census, maintained by the IBGE. The municipal income data for all municipalities were only for those years for which a demographic census was conducted.

Data regarding unpaved roads were generated from a joint study conducted by Institute of Applied Economic Research (IPEA) and IBGE (IBGE 1987). Sawed lumber production was obtained from the Institute of Environmental Research for Amazônia (IPAM) database.