BETA

CrisisDB Explorer

Exploring power transitions and elite dynamics using the Crisis Database

3,447Transitions
264Polities
38%Violent

What is this?

This is an interactive exploration of the CrisisDB Power Transitions dataset, testing predictions from Peter Turchin's Structural Demographic Theory (SDT).

SDT predicts that institutional complexity leads to elite overproduction — more administrative levels mean more elite positions, more competition, and more intra-elite conflict during power transitions.

Data: 3,447 power transitions from 264 polities, merged with Seshat complexity metrics. This is a subset of the full CrisisDB.

Explore the Patterns

What do polities at different complexity levels typically experience? Configure parameters and explore historical patterns from 3,447 observed transitions.

Configure Polity Profile

SimpleComplex

Using observed Markov transition rates

Compare to Real Polity

Historical Patterns at Complexity 5

Based on 25 polities, 352 transitions

Conflict Rate
30%
Median Tenure
8 yrs
Equilibrium Violence
36%
Next Transition
22% violent

Simulation

Click "Simulate" to see one possible trajectory based on historical rates

Detailed Findings

1. Elite Overproduction

Does administrative complexity predict intra-elite conflict?

Elite Overproduction: Complexity → Conflict

Administrative levels vs intra-elite conflict rate

n = 87 polities
Administrative Levels (Seshat)2468Intra-Elite Conflict Rate0%25%50%75%100%r = 0.362p < 0.001+5.6 pp / level
Finding: Positive correlation (r = 0.362, p < 0.001). Each additional administrative level associates with +5.6 percentage points higher intra-elite conflict rate.

2. Violence Contagion

Does a violent transition increase the probability of subsequent violence?

Violence Contagion

Markov transition dynamics: violence is "sticky"

Transition Matrix

→ Peaceful→ Violent
Peaceful78%22%
Violent40%60%
Key insight: After a violent transition, the next transition is 2.7x more likely to also be violent (60% vs 22% baseline).

Convergence to Equilibrium

Step 0/20
Peaceful
100.0%
Violent
0.0%
Stationary distribution: 64% peaceful, 36% violent
The system spends ~36% of time in violent states at equilibrium.
Finding: P(violent | previous violent) = 60% vs P(violent | previous peaceful) = 22%. Stationary distribution: 36% violent.

3. Temporal Dynamics

If the system were stationary, we'd expect constant rates across centuries. It isn't.

0%20%40%60%80%1200 BCE400 BCE0 CE400 CE800 CE1200 CE1600 CE2000 CECrisis of 3rd c.CenturyRate
Assassination
Military Revolt
Intra-Elite Conflict
Finding: Assassination dominates in antiquity (peaking at 45% in the 3rd century crisis), while intra-elite conflict shows distinct waves. The overall rate of political violence has declined since ~500 CE, but the composition shifts — different mechanisms dominate in different eras.

4. Polity Trajectories

Aggregate statistics mask divergent paths. Some polities maintain stability for centuries; others spiral into violence.

Select a polity above to see its transition trajectory
Compare: Venice maintains near-zero violence across 41 transitions (800–1797 CE), while Byzantine III shows 100% conflict in its final phase — distinct regimesthat polities can inhabit, with transitions between them.

5. Ruler Tenure

Do violent usurpers reign shorter?

Ruler Tenure: Survival by Accession Type

Does violent accession predict shorter reigns?

Violent Accession8 yrsmedian reign (n=1,303)
Peaceful Accession10 yrsmedian reign (n=1,586)
Difference-2 yrsp < 0.0001 (Mann-Whitney)
50%8yr10yr01020304050Years Since Accession0%25%50%75%100%PeacefulViolent

Which Violence Types Shorten Reigns Most?

Military Revolt
10yr
6yr
-4yr
Intra Elite
10yr
7yr
-3yr
Contested
10yr
7yr
-3yr
Predecessor Assassination
9yr
8yr
-1yr

Military revolts have the strongest effect: usurpers who seize power via coup reign 4 years shorter on average.

Violence Begets Violence

Peaceful Exit
Violent Exit
Peaceful Entry
91%
9%
Violent Entry
77.5%
22.5%

Rulers who seized power violently are 2.5x more likely to be removed violently. Chi-square p < 0.0001.

Finding: Violent accession → 2 years shorter median reign (8 vs 10 years, p < 0.0001). Military coups show the strongest effect (-4 years). Usurpers are 2.5x more likely to be removed violently themselves.

6. Transitions Over Time

When and where do power transitions cluster? Explore the temporal distribution.

Power Transitions Over Time

1,862 transitions from -500 to 1500

44.5% violent828 / 1,862
500 BCE0 BCE500 CE1000 CE1500 CETotalViolent
Pattern: Transition density peaks around 1050 CE (late medieval). Ottoman and Roman empires dominate the dataset with the most recorded transitions.

7. Notable Patterns

Outliers and trajectories that complicate the complexity-conflict relationship.

Complexity Without Conflict

Venice III (admin=5, conflict=5%, n=41), Egypt Classic Old Kingdom (admin=7, conflict=0%, n=20), and Northern Song (admin=7, conflict=11%, n=9) maintained complex bureaucracies with low intra-elite violence during transitions.

Succession institutions may buffer elite competition even at high administrative complexity.

Byzantine Degradation

Intra-elite conflict rates escalate across Byzantine phases: I (56%, n=25) → II (50%, n=22) → III (100%, n=10). Every power transition in the late Byzantine period involved intra-elite violence.

Declining territorial base with persistent elite expectations — consistent with SDT predictions.

Mamluk Escalation

Similar trajectory in Mamluk Egypt: I (68%, n=19) → II (73%, n=11) → III (80%, n=10).

Highest sustained conflict rates in the dataset. Military slave systems may carry structural instability in succession.

8. Toward Prediction: Violence as a Dynamical System

Political violence has structure. Can we move from description to prediction?

Equilibrium or Illusion?

The Markov chain converges to 36% violent at stationarity. But the temporal dynamics (Section 3) show the system is not stationary — rates shift across centuries. Are polities oscillating between basins of attraction (stable vs. crisis regimes), or is the drift driven by external factors?

Critical Transitions

Byzantine Empire: 56% → 50% → 100% violence. Mamluk: 68% → 73% → 80%. These trajectories resemble tipping points more than gradual drift. Complex systems theory predicts early warning signals before critical transitions (rising autocorrelation, critical slowing down). Can we detect them in the historical record?

Hidden States

A Hidden Markov Model could formalize this: a latent state (stable vs. crisis) generating observed transitions with different probabilities. The polity trajectories in Section 4 suggest at least two distinct regimes. What triggers the switch between them?

Open Questions

These require additional variables beyond power transitions:

  • Economic stress as a trigger: Does fiscal crisis or resource scarcity predict the transition from stable to crisis regimes? (Requires CrisisDB economic variables)
  • Population dynamics: Turchin's SDT predicts that population pressure drives elite overproduction. Can we measure this interaction directly? (Requires Seshat demographic data)
  • Institutional resilience: Venice maintained stability for a millennium. What institutional features predict resistance to violence contagion? (Requires Seshat institutional complexity variables)
  • Regional contagion: Does violence in one polity increase the probability of violence in neighboring polities? (Requires spatial analysis across the full CrisisDB)

9. Robustness Check

Do findings hold when excluding polities with fewer than 5 transitions?

All Polities

r = 0.362
3,447 transitions · 264 polities
P(V|V) = 60%
Violence rate: 37.5%

Excluding Sparse Polities (n < 5)

r = 0.362
3,246 transitions · 221 polities
P(V|V) = 62%
Violence rate: 38.3%
Result: Excluding 39 sparse polities slightly increases the violence rate (37.5% → 38.3%) and persistence (P(V|V): 60% → 62%). Core findings are robust to this exclusion. All 39 sparse polities had fewer than 5 transitions and were already excluded from the correlation analysis (which requires n >= 5).

Methodology & Limitations

Data Sources

  • CrisisDB: Power transitions with mechanism coding (P/IP/A/IA)
  • Seshat Equinox 2022: Administrative levels and complexity metrics
  • Merged on polity name (n = 87 with ≥5 transitions)

Limitations

  • Partial CrisisDB subset (power_transitions.csv only)
  • Correlation does not imply causation
  • Selection bias toward well-documented polities
  • Merging introduces data loss

Source Quality

Polities with fewer than 5 recorded transitions are flagged as sparse — too few data points for reliable violence rate estimates. Use the filter above to exclude these 39 polities and verify that findings hold.

Acknowledgments

This work builds on data and theory from Peter Turchin and the Complexity Science Hub Vienna. CrisisDB and Seshat are maintained by the Seshat: Global History Databank team.

Special thanks to Jenny Reddish, Jakob Zsambok, and Peter Turchin for feedback on source quality and methodology, and to Daniel Kondor for ongoing work on the power transitions data.