Abstract: Statistical causal inference methods have become widespread in epidemiology and the social sciences. Some, such as multiple regression analysis and generalised linear models, complement the methods traditionally used in ecology and have gained popularity in that field. Others remain largely unfamiliar to ecologists. Among these, this paper describes the propensity score method, disjunctive regression design, and the instrumental variable method, all widely used in epidemiology and the social sciences. The propensity score method estimates causal effects without bias by constructing a single synthetic variable that expresses the probability, or “propensity”, of a unit of interest being assigned to a given treatment based on multiple background factors. In conservation ecology, there are various situations where propensity scores may be appropriate, such as when estimating the causal effects of a binary conservation intervention. Where the objectives and circumstances fit, the propensity score method can be a powerful way to adjust for background factors in a single step. The regression discontinuity design, by contrast, estimates causal effects by estimating discontinuous changes in the regression line at the boundaries of the treatment change. Similarly, the instrumental variable method estimates causal effects using so-called instrumental variables that bring about changes from outside the system. There may be relatively few situations in ecology where the latter two methods can be effectively applied. However, awareness of these methods increases the range of options available when developing survey designs and statistical strategies. If and when one of these methods is applied, it will serve as a pioneering example in ecology.

Abstract: Distinguishing correlation from causation when evaluating factors threatening wild populations of organisms is crucial for the implementation of effective conservation measures. However, few studies in the field of conservation ecology have applied a causal inference framework to evaluate wild populations. In this study, we applied an integrated causal inference approach to explore the relationship between insecticide use and population decline, using the dragonfly Sympetrum frequens (Selys), which is among the most common dragonflies in Japanese paddy fields, as an example. In the late 1990s, S. frequens populations declined sharply in many regions. The main cause of these declines is suspected to have been the use of systemic insecticides such as the neonicotinoid imidacloprid and the phenylpyrazole fipronil, which were widely used in rice seedling nursery boxes. These insecticides were introduced immediately before the S. frequens population decline began, and subsequent laboratory- and field-based analyses have shown them to be highly toxic to dragonfly nymphs and other invertebrates. However, a causal relationship between insecticides and the decline of S. frequens has not been systematically determined, mainly due to the limited availability of quantitative data on species population size and habitat characteristics over the period of decline. Given these limitations, we applied five different approaches to investigate the relationship between insecticide use and S. frequens populations, as follows. First, we conducted a review of evidence based on currently available information, followed by the application of a statistical causal inference model using available insecticide usage and dragonfly monitoring data. Next, we conducted a field experiment to assess the effects of a novel insecticide on S. frequens and evaluated the effects of climate warming as a potential alternative explanation for the decline of S. frequens. Finally, we performed a mechanism-based evaluation of the effects of each factor using a population model. Our results suggest that the sharp population declines of S. frequens in the late 1990s were caused by the combined effects of highly toxic insecticides and habitat degradation due to its conversion to well-drained paddy fields.

Abstract:　In ecological studies, methods used in other fields to elucidate causal relationships, such as randomized controlled experiments and large-scale computer simulations, may not be feasible or effective. However, the availability of large-scale data generated in ecosystem monitoring is dramatically increasing. Accompanying this recent trend are growing expectations for the development of methods to elucidate causal relationships in a data-driven manner. CCM, proposed by George Sugihara et al. in 2012, has led ecologists to focus on causal estimation by time series. On the other hand, Granger causality and information theoretic approaches have also developed as important methods for dealing with causality in dynamical systems. Time series based causal analysis is based on a wide range of fields, and it is necessary to properly understand the advantages and disadvantages of each method in order to use them appropriately. In this paper, we propose that by introducing a unified perspective through information theory, Granger causality deals with momentary information imbalances, whereas CCM deals with those in the larger temporal structures represented by attractors. From this perspective, technical complementarity is recognized, while a new challenge of the multifaceted nature of causality emerges for the study of ecological dynamics. Thus, the elucidation of causality in ecological systems will progress by developing new approaches from a broad perspective that transcends domain boundaries and confronts real-world complexity.