Making statistics biologically relevant in fragmented landscapes Robert M. Ewers, Charles J. Marsh and Oliver R. Wearn Imperial College London, Silwood Park Campus, Ascot SL5 7PY, UK The biological impacts of habitat fragmentation are routinely assessed using standard statistical modelling techniques that are used across many ecological disci- plines. However, to assess the biological relevance of fragmentation impacts, we must consider an extra, spa- tial dimension to the standard statistical model: the biological importance of a significant and well supported model with large effect sizes crucially depends on the configuration of habitat within the study area. We argue that mapping the outputs from statistical models across a study area generates biologically meaningful esti- mates of fragmentation impacts. Integrating traditional statistical approaches with geographic information sys- tems will facilitate rigorous comparisons of fragmenta- tion impacts between taxa, studies and ecosystems. Complexity of habitat fragmentation research Habitat loss and fragmentation are widely recognised as dominant threats to biodiversity around the world [1–3] and habitat fragmentation has become the single largest topic of research in conservation biology [4,5]. Yet despite intense research effort into fragmentation effects and a vast and diverse literature written on the subject, there is still no comprehensive meta-analysis of fragmentation impacts. This might seem surprising for such a large field of study, but it arises from the composite nature of frag- mentation [5–8]: it encompasses, for example, changes in habitat amount, patch size, patch isolation and exposure of patch edges to a novel matrix habitat. Currently, there is no way of comparing the outcomes of studies that investigate the many different aspects of fragmentation. For example, is the impact of patch size on blue-winged macaws Primolius maracana in Brazilian Atlantic Forest [9] greater or less than the impact of isolation on populations of the red-backed vole Clethrion- omys gapperi in the USA [10]? And how do these impacts compare to that of habitat edges on the beetle Xylechinus pilosus in Finland [11]? Although a seemingly incongruent set of studies, these all investigate the impacts of habitat fragmentation on individual species, and for studies of habitat fragmentation to form a coherent literature it is imperative that we find a way of directly comparing studies and results as diverse as these. Individually, the various components of fragmentation can be subjected to meta-analysis [12–15]. However, a shortcoming of meta-analyses is that they only include a highly selective subset of the various aspects of fragmen- tation. For example, conducting a meta-analysis on patch size and isolation [13] tells us nothing about how these impacts compare to those of habitat edges on species [14]. In fact, studies of edge effects cannot currently be incorpo- rated into a meta-analysis of area effects, even though it has been shown that edge effects generate many area effects observed in fragmented regions [16–18]. This issue is complicated by the facts that some taxa are negatively affected by habitat fragmentation whereas others benefit [6], and that different taxa can simultaneously respond to more than one component of fragmentation [18]. Combining the data gathered in thousands of fragmen- tation studies to form a defensible estimate of the impact that habitat fragmentation is having on biodiversity is a daunting task, but we argue that it is an achievable one. To achieve this will require researchers to make the leap from reporting statistical models, P-values and effect sizes to assessing the biological relevance of their statistical mod- els. Here, we show how this can be done by amalgamating statistics with geographic information systems (GIS), thereby making integrated assessments of the impacts of habitat loss and fragmentation that directly incorporate real-world patterns of habitat configuration. Quantifica- tion of the biological relevance of statistical models will generate metrics that are comparable among species and studies, paving the way to a comprehensive meta-analysis of fragmentation impacts. Biological relevance of statistics in fragmentation studies To ask whether existing statistical approaches are appro- priate for assessing fragmentation impacts, we must first be clear about what we want to know. In many fragmenta- tion studies the goal is to determine if fragmentation has a significant effect on some ecological parameter. For exam- ple, does population density change with patch size? In this situation, standard hypothesis testing is appropriate and will tell the researcher if the patterns in the observed data deviate from a null expectation [19]. However, the value of such a simplistic question is limited and should be chal- lenged. Indeed, in many cases it is self-evident that some effect of fragmentation is inevitable, leaving a null hypoth- esis as a ‘‘straw man’’ [20]. Simply knowing the statistical significance of the relationship between density and patch size does not help us to gauge the biological importance of that effect [21]. Rather, the magnitude of the effect is far more valuable information [20]. A significant, but small, effect should not influence a management plan in the same Opinion Corresponding author: Ewers, R.M. (r.ewers@imperial.ac.uk) 0169-5347/$ – see front matter ß 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.tree.2010.09.008 Trends in Ecology and Evolution, December 2010, Vol. 25, No. 12 699