A Multinomial Logistic Model of Global Tropical Deforestation

Researchers have failed to reach on a consensus about the impact of different factors on tropical deforestation. Such failure has its origins in the problems with: deforestation data, hypothesizing the one-way effect of explanatory variables on deforestation, and estimation techniques. In this paper, these problems are dealt with. The data problem is taken care of by classifying the data on rate of deforestation into three classes. Our hypothesis is that any explanatory variable may have both positive and negative effect on deforestation, and the net effect will vary in different circumstances. Multinomial logistic regression is used to address some of the estimation issues. Only underlying causes of deforestation from different sectors of an economy are used as the independent variables. Data set for 64 tropical countries spread over Africa, Asia, and Latin America for two periods, 1980-90 and 1990-95 is used to estimate the deforestation model. Results indicate that same explanatory variable has different impacts on deforestation across the countries of three classes – high, medium, and low deforestation. 1 Respectively, M.Sc. student (Email: [email protected]) and Asst Professor, Faculty of Forestry, University of Toronto, 33 Willcocks St. Toronto, M5S 3B3; Email: [email protected], Ph. (416) 978 – 6196. INTRODUCTION Tropical forests are valued for direct economic benefits as well as host of intangible benefits they provide to the society. They are crucial for the conservation of 70% of the world’s plants and animal species, for the livelihood of 150 million people, carbon and hydrological cycles, timber production (around US $ 100 billion/year), variety of non timber forest products etc. (Roper and Roberts, 1999). Despite these benefits, tropical deforestation has been quite high during recent decades. According to Food and Agricultural Organization (FAO) of United Nations, annual tropical natural forest loss was 14.63 and 12.91 million hectares during 1980-90 and 199095 respectively (FAO, 1997). People from all walks of life have voiced concern over the shrinking forest cover and the resulting consequences on mankind. Therefore, there is a need to identify the causes of tropical deforestation, so that effective policy measures can be taken for a better future. PREVIOUS STUDIES AND PROBLEMS Large number of studies has been carried out at micro level i.e. household studies, and macro level i.e. national and international studies to describe the process of deforestation. The macro level studies are of 4 types. They are: descriptive (WRI, 1994), theoretical (Amsberg, 1998), empirical (Kant and Redantz, 1997), and simulation (Saxena et al., 1997) models. In spite of all these studies, there is a lack of general agreement about the causes of deforestation. One of the main sources of the disagreement lies with the diverging results from the empirical models that are mostly linear regression equations. The variation in results of linear regression models is due to three different problems related to deforestation data, hypothesis, and estimation techniques. Deforestation data problem: Different authors have used different sources of forest area data that differ from each other and also there is inconsistency in each source. FAO Production yearbook reports data on forests and woodland annually that are based on response of member countries to annual questionnaire by FAO. National governments’ always have incentive to misreport the actual data (Shafik, 1994). Also, these figures are that of forest area rather than forest cover, since no country does forest inventory every year. The most comprehensive and widely used dataset on forest resources till date has been the Forest Resource Assessment (FRA) 1990 and it’s revised publication State of the World’s Forest report 1997. However, the deforestation data are based on a deforestation model that uses rural population density and some ecological parameters as independent variables. So, there is always a problem of circularity in linear regression models when population is used as an independent variable. To deal with the problem of deforestation data, researchers (Rudel and Roper, 1997) have used other sources such as local reports to construct a crude, dichotomized measure of tropical deforestation as either high or low in a country and binary logistic regression is used to estimate the probability of a country having high or low deforestation. However, it is too simplistic to divide countries into two categories (high/low, with 1 percent as cut-off point), while remote sensing data for 21 countries that carried out at least one survey shows that deforestation rates vary for a wide range i.e. 0.3% in Rwanda to 7.2% in Jamaica.