A practical method to control spatiotemporal confounding in environmental impact studies

Separating natural spatiotemporal variation from the impact of human activities has long been a challenge in environmental impact studies. To overcome this problem, a causal modelling method based on spatiotemporal data, integrated with existing statistical methods such as multivariate redundancy analysis, multiple regression and, ordination was used for inferring causal effects of wastewater on biotic ecosystems. The causal modelling techniques were structural equation modelling (SEM) and Bayesian Networks (BNs); SEM, with the help of statistical analysis, was used for building deterministic models while the composite hypothesis underlying the models was checked based on the principle of BNs. Both spatial and temporal variations were considered in the design of the study so that spatiotemporal confounding could be controlled by adjusting for 'time' and 'distance' in the models. This improved the external validity of the models, so they could be used for predicting the effect of interventions, e.g. manipulating the discharge loads. This could be possible where time-varying variables such as quantity of discharge effluent were included in the models. Models can be used for prediction the effect of an intervention in situations understood as causal. Thus, the causal structure of composite hypotheses of the study was tested using both local and global tests.