Increasing the robustness of natural resource management research by using multiple methods

Abstract

Research on natural resource management like fisheries, irrigation systems or forestry traditionally uses case studies, providing us with a rich, in-depth perspective on many single systems. This comes with a disadvantage - lacking comparability. One reason for that is a multitude of methods used. If, for example, the drivers for ecological success in irrigation systems are to be identified, disagreements between studies may be due to differences in the variables examined, their operationalization or, precisely the methods used. To my knowledge, there is no study in the natural resource management and social-ecological systems literature that analyses the influence of method choice on the results. Therefore, we use a high-quality data set, the Nepal Irrigation Institutions and Systems database (NIIS), developed at the Ostrom Workshop, of 263 cases with each record having information on around 600 variables. With that data set we test the influence of methods on the results, i.e. the relevance of well-known concepts for success in natural resource management, like participation of users, relations with other groups, etc. We use multivariate linear regressions (MLR), random forests (RF), gradient boosting (GBM), shallow neural networks (SNN) and deep neural networks (DNN). The results indicate that although some agreements exist across methods, there are substantial differences as well. We see this research as a step towards increasing the robustness of results and improving the generalisability and reproducibility of natural resource management research.