Uncertainty Of Statistical Models Associated With The Levels Of Aggregation Of Spatial Information

The modeling of phenomena such as land use / cover changes is based on the evaluation of the relationship between change and explanatory variables using statistical methods as regression models. The explanatory variables describe the physical and socioeconomic conditions of the territory. The available information is often presented in a spatially aggregated form based on political-administrative units such as municipalities. However, results of statistical analyses are not independent from the spatial configuration of the units used to aggregate the information. In this study, we analyze the effects of this effect, known as the modifiable areal unit problem (MAUP), on the evaluation of the factors of the distribution of forest cover in Mexico at different aggregation levels. We used population census variables along with topographic and accessibility variables aggregated using basic geostatistical areas, municipalities and states. The results show that the level of aggregation of the information affected the values of the correlation coefficient and the fitting of the regression models. The MAUP had a substantial effect on these models, in particular, when there is no strong relationship between the dependent variable and the explanatory variables. These results suggest that the relationships and inferences obtained using aggregated data with administrative units such as counties, provinces or municipalities, should be interpreted with caution.