Causal and Predictive Data Analysis for Conservation: Simulation Based Comparisons and a Case Study for Detecting...

Estimating the causal effect of a variable is an important task for applied ecology. While several methods have been applied to empirical, observational studies, there have been a few attempts to employ the causal inference approach based on the propensity score methods in ecology and evolutionary biology despite its widespread usage in other scientific fields. This paper applies the overlapping-weighted estimator to the Eurasian red squirrel Sciurus vulgaris to evaluate human activity on behavioral tolerance to humans as a model case. This statistical method is one of the common propensity score methods in the statistical community to better evaluate the causal effect of particular variables on a target variable. We focused on the effect of artificial feeding on tolerance to humans because feeding has been suggested to be a main driver of habituation to humans, while the causal effect has not been statistically tested. We performed an estimation of causal effects and compared results with the analysis that employed commonly used methods including AIC and LASSO. The results showed that the effect of artificial feeding is larger than previously known and that AIC and LASSO yielded biased results by dismissing confounding variables. Our results indicate that propensity score methods can be useful for wildlife management by offering a more accurate evaluation of causal effects.