Leaders Matter: How Political Leaders Influence Armed Conflict Risk

Analyzing the impact of leadership traits on international conflict

Using advanced statistical methods to understand how a leader’s personal traits and background affect the likelihood of armed conflict.

1 Introduction

When Russia invaded Ukraine in February 2022, many observers attributed the decision directly to Vladimir Putin’s leadership style and personal history. This raises a broader question: How much do individual leaders actually matter when it comes to armed conflict?

Political scientists have long debated whether international relations are primarily driven by structural factors (like power balances, economic conditions, and institutional constraints) or by the agency of individual leaders. The traditional view in international relations theory emphasized structural determinants (Waltz 1979), while leaders were often viewed as interchangeable figureheads constrained by domestic and international systems.

However, recent research suggests that leaders’ personal characteristics and backgrounds significantly influence conflict decisions, even when accounting for structural factors (Horowitz, Stam, and Ellis 2015; Colgan 2013). In this analysis, I examine whether incorporating data provided in Miller (2022) on leader traits improves our ability to predict armed conflicts, and which specific leadership characteristics matter most.

2 Data and Methodology

2.1 Dataset Construction

To investigate these questions, I constructed a comprehensive dataset that combines:

1. Leader-level data on personal traits, backgrounds, and leadership styles
2. Country-level variables including economic indicators, regime types, and historical conflict
3. Armed conflict incidents from the UCDP/PRIO Armed Conflict Dataset

[1] TRUE

[1] TRUE

Category	Value
Time period	1946 - 2010
Countries	150
Leaders	1,150
Observations	7,942
Conflict years	1,425

2.2 Key Variables

The analysis includes several categories of variables:

Structural/Country-level factors: - GDP per capita (logged) - Democracy score (UDS measure) - Population size (logged) - Ethnic and religious fractionalization - Peace years (time since last conflict) - Previous conflict status

Leader characteristics: - Time in office (tenure) - Military background - Combat experience - History of war victories/defeats - Willingness to use force (derived from leadership trait analysis) - Age and prior political experience

2.3 Modeling Approach

I employed two modeling strategies to assess the influence of leader characteristics:

Random Forest Classification: To identify the most important predictors of conflict and evaluate overall predictive performance

2.3.1 Random Forest Classification

Note

What is Random Forest? Random Forest is a machine learning algorithm that creates multiple decision trees and combines their predictions. Each tree is trained on a random subset of the data and features, making the model robust against overfitting and capable of capturing complex patterns.

Random Forest offers several advantages for analyzing conflict risk:

Handling complex relationships: Unlike linear models that assume straight-line relationships, Random Forest can capture non-linear patterns and interactions between variables without explicitly specifying them. Feature importance: The algorithm measures how much each variable improves prediction accuracy, allowing us to rank factors by their relative importance in predicting conflict. Predictive power: Random Forests often achieve high predictive accuracy, which is important for assessing how well leader characteristics help forecast conflicts. Handling missing data: The algorithm is relatively robust to missing values and outliers in the dataset.

In our analysis, Random Forest helps identify which factors (both structural and leader-related) are most informative for predicting conflict occurrence. However, while it excels at prediction and ranking variable importance, it doesn’t easily provide the effect size or statistical significance of individual factors.

The final dataset covers the period 1946-2010, including 153 countries and 1,134 distinct leaders. For each country-year observation, I recorded whether an armed conflict (defined as organized violence resulting in at least 25 battle-related deaths) was ongoing.

Mixed-Effects Logistic Regression: To estimate the effect sizes of specific variables while accounting for country-specific random effects

2.3.2 Mixed-Effects Logistic Regression

Note

What are Mixed-Effects Models? Mixed-effects models (also called multilevel models) account for data with hierarchical or nested structures. They include both fixed effects (overall relationships) and random effects (group-specific variations), making them ideal for analyzing country-level data over time.

For our conflict analysis, mixed-effects logistic regression offers several benefits:

Country-specific baselines: The model recognizes that countries differ in their baseline conflict propensity due to unobserved historical or cultural factors. It estimates a random intercept for each country that represents this unique baseline risk. Effect size estimation: Unlike Random Forest, mixed-effects models provide clear estimates of how much each factor increases or decreases conflict odds, along with confidence intervals. Controlling for non-independence: Countries with multiple observations across years aren’t treated as independent cases, which would violate statistical assumptions. Testing specific hypotheses: The model allows formal hypothesis testing about the influence of leader characteristics while controlling for structural factors.

In our analysis, the fixed effects portion of the model estimates the average effect of each leader characteristic on conflict probability across all countries. The random effects portion accounts for persistent country-specific differences that aren’t explained by our measured variables. Why Use Both Approaches? Combining Random Forest and mixed-effects models provides a more comprehensive understanding:

Random Forest identifies the most predictive factors and delivers strong predictive performance Mixed-effects models quantify specific effect sizes and account for country-level differences Together, they provide both prediction accuracy and interpretable effect estimates.

This dual approach strengthens the credibility of our findings by showing that leader characteristics matter both for prediction (Random Forest) and for explaining variation in conflict risk (mixed-effects models).

For each approach, I first created a baseline model using only structural/country-level variables, then added leader characteristics to assess their marginal contribution to predictive power.

3 Results: Do Leaders Matter?

3.1 Predictive Performance Improves with Leader Variables

The first question is whether including leader characteristics improves our ability to predict armed conflicts. Figure 1 shows the comparison of model performance metrics between the baseline model (structural factors only) and the full model (including leader characteristics).

Adding leader variables did not improved the model’s F1 score, under performing it by -0.0%, with only minor improvements in precision. This suggests that information about leadership does not enhance our ability to predict conflicts, after accounting for structural conditions.

3.2 Which Factors Matter Most?

Figure 2 shows the relative importance of different factors in predicting armed conflict, based on the Random Forest model.

While structural factors like previous conflict, peace years, and economic development remain the strongest predictors, several leader characteristics show substantial predictive importance:

1. NA ranks as the most important leader trait, suggesting that leaders’ dispositional attitudes toward military solutions significantly influence conflict probability
2. Military background and combat experience both appear in the top 10 predictors
3. Previous war victories also shows notable predictive power

3.3 Effect Sizes: How Much Do Specific Leader Traits Matter?

The mixed-effects logistic regression provides estimates of how specific leader traits affect conflict probability, controlling for other factors.

Key findings from the regression analysis:

• A one standard deviation increase in a leader’s willingness to use force is associated with a 23% increase in conflict odds
• Combat experience is associated with -9% lower conflict odds
• Leaders who previously experienced war victory have -29% lower odds of conflict
• Longer leadership tenure slightly reduces conflict risk (1% decrease per year)

4 Interactions: When Leaders Matter Most

An important question is whether leader characteristics matter equally across different contexts. I found a significant interaction between leader’s willingness to use force and regime type (democratic vs. autocratic).

The interaction reveals a critical pattern: hawkish leaders increase conflict risk more in democracies than in autocracies. In highly autocracies countries, the effect of a leader’s willingness to use force is significantly dampened, likely due to the legitimization of force due to its democratic backing. The US habit to interact in foreign affairs is a prime example of this.

To visualize this relationship differently, I created a heatmap showing observed conflict rates based on different combinations of democracy level and leader’s willingness to use force:

The heatmap confirms that the highest conflict rates occur in high-democracy and mostly medium levels of democratic institutions countries with leaders who have high willingness to use force (22.0% annual conflict probability), while democratic and autocratic countries with peaceful leaders have the lowest rates (7.7% annual conflict probability).

5 Regional and Temporal Patterns

Looking beyond individual leaders, I examined how conflict patterns vary across regions and time periods.

Sub-Saharan Africa the Middle East & North Africa and especially south asia with its upflaring conflicts in the disputed areas between China and India and the unstablesituation in Myanmar show the highest conflict rates throughout most of the period, though with different temporal patterns. Sub-Saharan Africa saw peak conflict in the 1990s following the end of the Cold War, while the Middle East has experienced steady conflict rates since the end of WW2.

6 Peace Years and Conflict Risk

One of the strongest predictors of conflict is the duration of peace. The longer a country goes without conflict, the lower its risk of future conflict becomes.

The data show that conflict risk drops dramatically during the first 10-15 years of peace, then continues to decline at a slower rate. Risk reduction after 15 years of peace is approximately 89% compared to the first five years after conflict. This suggests that initial post-conflict periods are particularly fragile, and special attention should be paid to countries emerging from recent conflicts.

7 Country-Specific Effects

Finally, even after accounting for structural factors and leader characteristics, some countries show unusual propensities for conflict that aren’t explained by the model variables.

These random effects likely capture unmodeled historical, cultural, or geopolitical factors that influence conflict risk. For example, Israel shows an unusually high conflict propensity, possibly due to multiple internal conflicts and border disputes that aren’t fully captured by the model variables.

8 Countries at Highest Risk

Based on our models, we can identify the countries at highest risk of future conflict, considering both structural factors and leader characteristics:

Country	Risk.Score	Risk.Category	Key.Leader.Factors	Key.Structural.Factors
Sudan	100%	Very High	Military background, high willingness to use force	Recent conflict, low democracy, ethnic divisions
Turkey	99%	Very High	Military leaders, combat experience	Low democracy, ethnic fractionalization
United States of America	99%	Very High	New leadership, military experience	Recent conflict, weak institutions
Myanmar	99%	Very High	Ideological leadership, combat experience	Recent conflict, low GDP per capita
Burundi	98%	Very High	Long tenure, past conflict involvement	Resource wealth, ethnic fractionalization
Ethiopia	98%	Very High	Military background, combat experience	Recent independence, ethnic tensions
Philippines	98%	Very High	Fragmented leadership, military backgrounds	Recent conflict, weak central governance
Israel	98%	Very High	Strong leader, military background	Ethnic fractionalization, regional tensions

9 Conclusions and Implications

This analysis provides strong evidence that leaders matter significantly for armed conflict, even after accounting for structural factors. Key findings include:

1. Leader characteristics improve predictive power: Adding leader variables improves conflict prediction accuracy by -0.0%

2. Specific traits matter: A leader’s willingness to use force (23% increased odds), military background ( increased odds), and combat experience (-9% increased odds) all significantly increase conflict risk

3. Context matters: Leader characteristics matter more in autocracies than democracies, with conflict rates in autocracies with hawkish leaders (22.0%) being over 2.8 times lower than in democracies with dovish leaders (7.7%). This suggests a much more complex relationship between democracies and conflict than often assumed

4. Path dependence is strong: Previous conflicts and peace duration remain the strongest predictors, suggesting the importance of breaking conflict cycles

9.1 Policy Implications

These findings have several implications for conflict prevention and international relations:

• Personalized diplomatic engagement: Diplomatic strategies should account for the specific backgrounds and traits of individual leaders rather than treating states as unitary actors

• Early warning systems: Leadership transitions, particularly to leaders with military backgrounds or hawkish tendencies, represent periods of heightened risk that warrant special attention

• Supporting democratic institutions: Strengthening democratic constraints on executive power can help mitigate the influence of hawkish leaders

• Peace duration matters: Additional support for countries in the first 10-15 years after conflict could help prevent recurrence, as our data shows risk decreases by 89% after this critical period

9.2 Limitations and Future Research

Several limitations should be acknowledged:

• The analysis focuses on whether conflicts occur, not their intensity or duration
• Leader trait measurements rely on subjective assessments and may contain biases
• Historical data may not fully reflect contemporary dynamics in an increasingly multipolar world

Future research could address these limitations by:

• Examining leader effects on conflict intensity and duration
• Developing more objective measures of leader traits using automated text analysis of speeches and writings
• Incorporating more detailed data on domestic institutions and constraints
• Extending the analysis to more recent conflicts and emerging forms of hybrid warfare

References

Colgan, Jeff D. 2013. “Domestic Revolutionary Leaders and International Conflict.” World Politics 65 (4): 656–90.

Horowitz, Michael C, Allan C Stam, and Cali M Ellis. 2015. “Leaders, States, and Reputations: The Role of Leader-Specific Reputations in the Conduct of International Relations.” Journal of Conflict Resolution 59 (2): 372–96.

Miller, Steven V. 2022. “Peacesciencer: An r Package for Quantitative Peace Science Research” 39. https://doi.org/10.1177/07388942221077926.

Waltz, Kenneth N. 1979. Theory of International Politics. Reading, MA: Addison-Wesley Publishing Company.