The Evidence

Regression results from 1.9 million observations across 30 years

Regression Results

The analysis uses a panel of 64,818 grid cells (55 x 55 km each) covering the entire globe from 1989 to 2018.

Table 1: Proximity and Conflict (Baseline)

Areas closer to maritime choke points are associated with significantly higher conflict risk — across all types of violence.

TipWhat do the column headers mean?

• Any Event: Any type of armed conflict
• State-based: Fighting involving a government (e.g., civil wars, interstate wars)
• Non-state: Fighting between non-government armed groups (e.g., militias, rebel factions)
• One-sided: Violence deliberately targeting civilians (by a government or armed group)
• ln(deaths+1): Death count on a logarithmic scale — prevents a single massive battle from dominating the results, capturing overall intensity

	Any Event	State-based	Non-state	One-sided	ln(deaths+1)
Proximity to choke point	0.0031*** (0.0004)	0.0022*** (0.0002)	0.0002*** (0.0001)	0.0007*** (0.0002)	0.0013 (0.0012)
Latitude FE	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes
Avg. conflict rate	0.015	0.007	0.002	0.006	0.035
Observations	1,944,540

NoteReading the table

Each column uses a different measure of conflict. The coefficient in the highlighted row tells you: for each unit increase in proximity (one SD = 1,100 km closer), how much does conflict risk change? Standard errors (in parentheses) measure precision — smaller is better. *** means the result is statistically significant at the 1% level. “FE” stands for fixed effects — statistical controls that account for differences across latitudes, countries, or years.

Table 2: The Interaction — Proximity x Trade Openness

The key result: proximity is associated with more conflict when trade is low, but the interaction term flips the sign when trade is high.

	Any Event	State-based	Non-state	One-sided	ln(deaths+1)
Proximity to choke point	0.0148*** (0.0010)	0.0052*** (0.0006)	0.0017*** (0.0003)	0.0079*** (0.0006)	0.0338*** (0.0032)
Proximity x Trade Openness	-0.0277*** (0.008)	-0.0087*** (0.0011)	-0.0034*** (0.0006)	-0.0156*** (0.0010)	-0.0711*** (0.0053)
Latitude FE	Yes	Yes	Yes	Yes	Yes
Country FE	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes
Avg. conflict rate	0.015	0.007	0.002	0.006	0.035
Observations	1,944,540

In plain English: Column 1 shows that being near a choke point is associated with higher conflict risk (positive coefficient), but this estimated effect shrinks — and eventually reverses — as global trade grows (negative interaction). This pattern holds across all conflict types, with all interaction terms significant at the 1% level.

The “Two Worlds” Comparison

{
  const beta1 = 0.0148;
  const beta2 = -0.0277;
  const baseline = 0.015;

  const meLow = beta1 + beta2 * 0.40;
  const meHigh = beta1 + beta2 * 0.60;
  const pctLow = (meLow / baseline * 100).toFixed(1);
  const pctHigh = (meHigh / baseline * 100).toFixed(1);

  return htl.html`
  <div style="display: flex; gap: 2rem; flex-wrap: wrap; margin: 2rem 0;">
    <div style="flex: 1; min-width: 280px; background: #ffebee; border-radius: 8px; padding: 1.5rem;">
      <div style="font-weight: 700; color: #c62828; font-size: 1.1rem; margin-bottom: 0.5rem;">
        World 1: Low Globalization
      </div>
      <div style="color: #555; font-size: 0.95rem;">
        Trade openness = 0.40 (early 1990s)
      </div>
      <div style="font-family: 'JetBrains Mono', monospace; font-size: 1.3rem; color: #c62828; margin: 1rem 0;">
        ME = ${meLow > 0 ? '+' : ''}${meLow.toFixed(4)}
      </div>
      <div style="font-size: 0.95rem;">
        Being 1,100 km closer to a strait is associated with <strong style="color: #c62828;">${pctLow}% higher</strong> conflict risk (relative to baseline)
      </div>
    </div>
    <div style="flex: 1; min-width: 280px; background: #e8f5e9; border-radius: 8px; padding: 1.5rem;">
      <div style="font-weight: 700; color: #2e7d32; font-size: 1.1rem; margin-bottom: 0.5rem;">
        World 2: High Globalization
      </div>
      <div style="color: #555; font-size: 0.95rem;">
        Trade openness = 0.60 (mid-2000s peak)
      </div>
      <div style="font-family: 'JetBrains Mono', monospace; font-size: 1.3rem; color: #2e7d32; margin: 1rem 0;">
        ME = ${meHigh.toFixed(4)}
      </div>
      <div style="font-size: 0.95rem;">
        Being 1,100 km closer to a strait is associated with <strong style="color: #2e7d32;">${Math.abs(pctHigh)}% lower</strong> conflict risk (relative to baseline)
      </div>
    </div>
  </div>`;
}

Robustness

The paper includes extensive robustness checks documented in the SI Appendix: cell fixed effects (controlling for all time-invariant local characteristics), alternative choke point definitions (excluding man-made canals), nonlinear specifications (tercile and quartile regressions), alternative conflict data sources (GDELT), and alternative standard error clustering.

NoteA note on causality

The paper uses latitude, country, and year fixed effects, clustered standard errors, and extensive robustness checks (cell FE, alternative definitions, alternative data sources) to address confounding. The theoretical model provides a game-theoretic foundation for the empirical patterns. However, as with any observational study, these estimates reflect strong and robust associations — not definitive proof of causality. The results are consistent with the proposed mechanism, but alternative explanations cannot be fully ruled out.

What do these findings mean for the real world — and for policy? Next: Policy Implications