The use of a multilevel statistical model to analyze factors influencing land use: a study of the Ecuadorian Amazon

This research introduces a novel application of a constrained multivariate response multilevel model of land use. Methodological objectives are (1) to examine the conceptual specification of a multilevel model of land use and (2) to incorporate constraints in multilevel model estimation to reflect the ipsative nature of land use as a response. The substantive objective is to demonstrate and report multilevel results from a representative sample of farm plots (fincas) in 1999 nested in 59 communities throughout the Northern Ecuadorian Amazon. Results indicate that population pressure, labor, and accessibility and proximity between households and communities are strong predictors of land use. Communities are seen to significantly influence land use at the finca level through unobserved and observed structural effects. We conclude with four recommendations to apply multilevel models: (1) consider alternative types of multilevel models to better match land use data distributions; (2) test and control for spatial autocorrelation; (3) collect the best data possible at higher levels; and (4) consider further integration of remotely sensed and household survey data using a multilevel framework.