Quantitative evaluation of regulatory policies for reducing deforestation using the bent-cable regression model

Abstract: Reducing and redressing the effects of deforestation is a complex public policy challenge, and evaluating the efficacy of such policy efforts is crucial for policy learning and adaptation. Deforestation in high-income nations can contribute substantially to global forest loss, despite the presence of strong institutions and high policy capacity. In Queensland, Australia, over 5 million hectares of native forest has been lost since 1988. Successive regulatory policies have aimed to reduce deforestation in Queensland, though debate exists over their effect given the influence of other drivers of forest loss. Using a hierarchical Bayesian statistical framework, we combine satellite imagery of forest loss with macroeconomic, land tenure, biophysical and climatic variables to collectively model deforestation for 50 local government areas (LGAs) across Queensland. We apply the spatially explicit bent-cable regression model to detect trend change that may signal a regulatory policy effect. We find that annual % growth in GDP was the only clear driver of LGA-specific deforestation after adjusting for other covariate effects. Our model shows strong evidence of spatial contagion in deforestation across Queensland, and this effect is influenced by the dominant land tenure type within each LGA. We find our model exhibits a "bend" mostly between 2000 and 2007, consistent with expectations, but the signal is not particularly strong due extreme variation in deforestation trends between and within LGAs. Our results demonstrate that the bent-cable model is a promising technique for detecting system changes in response to policy interventions, but future work should be conducted at a national scale to provide more data points, and incorporate more LGA-specific data to improve model goodness-of-fit.