Interpretation
Early warning signals are potentially powerful tools for managers. However, their interpretation can be complex and requires nuance. This is particularly true for rolling window approaches and EWSNet as it remains unclear what constitutes a ‘strong’ correlation or prediction probability. We however believe there are three approaches to defining an appropriate warning using EWSs. Firstly, a user may refer to a reference period for a baseline correlation, or track change in the strength of a signal through time (as in the expanding window approach above), where deviations from the general trend are informative. The second requires the user to define how conservative an assessment they require. For example, if the negative consequence of a transition is significantly larger than the consequence of acting upon a false positive, then a lower confidence warning may be appropriate (i.e. a low Kendall’s Tau coefficient/EWSNet prediction probability). And finally, the third requires comparing the observed signal to a distribution of signals generated via permutation of the original time series. If the observed signal is in the top x-th quantile of the distribution (the 95th quantile is commonly used) then a warning may be identified. Alternatively, a fourth option is applicable for EWSNet following the original authors’ suggestions, where a probability larger than 0.33 (the chance that all outcomes are equally likely) is indicative of an approaching transition (Deb et al. 2022).