In land use research regression techniques are a widely used approach to explore datasets and to test hypotheses between land use variables and socio-economic, institutional and environmental variables. Within land use science researchers have argued the importance of scale and levels. Nevertheless, the incorporation of multiple scales and levels and their interactions in one analysis is often lacking. Ignoring the hierarchical data structure originating from scale effects and levels, may lead to erroneous conclusions due to invalid specification of the regression model. The objective of this paper is to apply a multilevel analysis to construct a predictive statistical model for the occurrence of land use. Multilevel modelling is a statistically sound methodology for the analysis of hierarchically structured data with regression models that explicitly takes variability at different levels into account. For a land use study in the Philippines multilevel models are presented for two land use types that incorporate the field, household and village level. The value of multilevel modelling for land use studies and the implications of multilevel modelling for data collection will be discussed. The results show that explanatory variables can account for group level variability, but in most cases a multilevel approach is necessary to construct a sound regression model. Although land use studies often show clear hierarchical structures, it is not always possible to use a multilevel approach due to the structure of most land use datasets and due to data quality. Potentially, multilevel models can address many important land use issues involving scales and levels. Therefore, it is important in land use change research to formulate hypotheses that explicitly take scale and levels into account and then collect the appropriate data to answer these questions with approaches such as multilevel analysis.
|Publication status||Published - 2006|
- tropical deforestation
- multiscale analysis
- global change