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).