Early-Warning Signals of Critical Transitions

Description

Detects rising autocorrelation, variance and slower recovery in a time series as signs that a system is approaching a tipping point, such as collapse of engagement.

Ecologists found that systems near a bifurcation recover more slowly from small perturbations, a phenomenon called critical slowing down. This appears in time series as rising lag-1 autocorrelation, rising variance and sometimes skewness or flickering, before a regime shift such as a lake turning turbid or a fishery collapsing. Rolling-window analysis with trend tests (Kendall's tau) and surrogate data provides statistical evidence. Applied at the level of individual users or communities, these indicators may warn that engagement, community health or a habit is losing resilience before it collapses, giving a theory-grounded alternative to opaque churn prediction.

Attributes

Loss type
Unmigrated
Stage
Post-launch
Effort
Weeks

Steps

  1. Select time series that represent system state (e.g. daily activity of a community, a user's weekly usage).
  2. Detrend or filter the series (e.g. Gaussian kernel) to remove long-term trends.
  3. Compute rolling-window indicators: lag-1 autocorrelation, variance, skewness.
  4. Test for rising trends with Kendall's tau and compare with surrogate series to assess significance.
  5. Validate on historical cases: do indicators rise before known collapses (churn, community death)?
  6. Use validated indicators for monitoring and target resilience-building interventions.

Sources

  • Scheffer et al., Early-warning signals for critical transitions (2009), Nature
  • Dakos et al., Methods for detecting early warnings of critical transitions in time series (2012), PLoS ONE
  • Scheffer, Critical Transitions in Nature and Society (2009)
  • van der Maas, H., Complex Systems Research in Psychology (2024)

Also known as

  • Critical slowing down
  • Early warning indicators
  • Resilience indicators

Categories

  • Strategy & Collaboration
  • Systems & Complexity