Sensitivity Analysis and Uncertainty Parameter Quantification in a Regression Model: The Case of Deforestation in Tanzania

Abstract

In this paper a multiple regression model for the economic factors and policy that influence the rate of deforestation in Tanzania is formulated. Sensitivity analysis for parameters of explanatory variables using one-at-a time and direct methods is carried out and the model is fitted by classical least square (LSQ) and Markov Chain Monte Carlo (MCMC) methods. Uncertainty quantification of parameters by adaptive Markov Chain Monte Carlo methods is performed. The coefficient of determination indicates that 87% of deforestation rate is explained by explanatory variables captured in the model. Household poverty rate is found to be the most sensitive factor to deforestation, while purchasing power is the least sensitive in both methods. Model validation indicates a good agreement between the collected data and the predicted data by the model and Markoc Chain Monte Carlo method yielded a good sample mix. Thus, the study recommends that since economic activities tend to increase the rate of deforestation, then policy and decision-making processes should link the country ' s desire for economic growth and environmental management.

Keywords: deforestation; economic factors; Markov Chain Monte Carlo methods; regression model; sensitivity.