Multilevel Statistical Models to Assess Factors Influencing Land Use: A Case Study in the Ecuadorian Amazon

William Pan, University of North Carolina at Chapel Hill
Richard Bilsborrow, University of North Carolina at Chapel Hill

Statistical models examining the interaction between population and land use (LU) have been evolving through the integration of survey and satellite data as well as the application of more sophisticated estimation techniques. However, very few approaches incorporate contextual factors influencing LU or directly test or control for spatial autocorrelation among model variables. The absence of context ignores much of the historical development of LU dynamics, while the presence of spatially correlated variables can provide results that are exaggerated due to abnormally small standard errors. Therefore, we have developed a multilevel model to simultaneously investigate contextual effects at the household and community levels. In addition, we explicitly test for the presence of spatial autocorrelation and control for spatial dependencies in the model. We will present the statistical formulations of multilevel models, spatial tests for autocorrelation, and results demonstrating the improved estimation when including community variables and controlling for autocorrelation.