Quantifying Risk Factors of Violence in Maritime Piracy Incidents Using Categorical Association Measures ^†

Abstract

Maritime piracy remains a persistent security challenge across several global regions, with violent incidents posing the greatest threat to crew safety and vessel operations. This study investigates the relationship between violent escalation in piracy incidents and a set of contextual and operational variables using classical categorical data statistics. A dataset comprising reported maritime piracy and armed robbery events from 2015–2024 was compiled from IMB, OBP, and IMO sources and analysed through chi-square tests of independence, followed by Cramér’s V to quantify the strength of association. The results demonstrate that violence is not randomly distributed across incident characteristics. Geographic region exhibits the strongest measurable association with violent outcomes, reflecting the influence of regional security dynamics and the presence of organized criminal networks. Attack type and weapon type show additional, though weaker, associations, indicating that close-range engagement and the presence of firearms increase the likelihood of escalation. Vessel type, flag state, and seasonal timing display only marginal effects. Overall, the findings highlight that the probability of violence during piracy events is primarily shaped by spatial context and tactical execution. The study confirms that chi-square and Cramér’s V offer a transparent, interpretable framework for identifying key risk factors and can serve as a foundation for operational threat assessments and maritime security planning.

1. Introduction

Despite advances in navigational technologies, vessel monitoring systems, and international security cooperation, maritime piracy remains a persistent risk to global maritime transport [1]. The consequences of piracy attacks include direct harm to crew members, disruption of shipping routes, increased insurance costs, and destabilization of maritime supply chains [2]. Traditional preventive measures, such as naval patrols and convoying, have proven effective only in specific regions and time periods, while piracy organizations have demonstrated an ability to adapt to changing operational conditions [3].

One of the key challenges in countering maritime piracy is the ability to identify attack patterns and anticipate emerging risks in advance. This requires a systematic understanding of the features that most strongly influence the likelihood of an attack, including geographical, operational, and behavioural factors. Recent advances in artificial intelligence and machine learning offer new opportunities for analyzing large-scale heterogeneous maritime security datasets and generating predictive assessments in near real-time [4].

Although a growing number of studies describe regional piracy trends, relatively few works provide a systematic, quantitative assessment of the specific factors associated with violent escalation. Existing research often focuses either on descriptive statistics or on pre-dictive modelling, but the strength of categorical associations between operational variables and violence remains insufficiently explored [5]. This creates a methodological gap that limits the interpretability of current risk assessment tools.

Contrary to advanced machine learning models, classical categorical-data statistics offer full transparency, interpretability, and robustness when dealing with heterogeneous maritime piracy records [6]. Chi-square tests and Cramér’s V allow researchers to identify fundamental structural relationships in the data without making assumptions about model architecture or training procedures [6,7,8]. This makes such methods particularly suitable for exploratory risk factor screening in maritime security research [8].

This study focuses exclusively on statistical association techniques to identify the factors most closely related to the occurrence of violence in maritime piracy incidents. By applying chi-square tests of independence and Cramér’s V effect size, the analysis provides a transparent, interpretable evaluation of how operational and contextual variables contribute to escalation risks.

2. Literature Review

Maritime piracy is a multidisciplinary research domain spanning security studies, criminology, economics, human factors, and maritime logistics [9]. Over the past two decades, the field has evolved from descriptive reporting of incidents toward increasingly sophisticated analytical frameworks that combine spatial statistics, behavioural modelling, and risk evaluation [10,11]. The literature generally converges on the observation that piracy is neither random nor uniformly distributed across the maritime domain, but instead shaped by structural socio-economic conditions, regional governance capacity, and tactical behaviours of offenders [12].

Piracy is a highly diverse phenomenon geographically [13]. A large body of work emphasizes that piracy is profoundly shaped by regional conditions. Conditions conducive to piracy include limited state control of coastal waters, political and economic instability, high poverty levels in local communities, and easy access to weapons and motorboats [14]. The literature indicates the dominance of three high-risk areas: the Gulf of Aden and the Western Indian Ocean, the Gulf of Guinea, and the Southeast Asian region, particularly the Indonesian Straits and the Sulu Sea [13]. Classic studies of the Gulf of Aden, the Somali Basin and the Horn of Africa link high-intensity piracy to state fragility, weak maritime domain awareness, and the presence of organized criminal groups with access to weapons and logistics [15]. Research concerning the Gulf of Guinea shows a different operational signature: attacks are often short-range, highly violent, and oriented toward kidnapping for ransom rather than hijacking for resale, reflecting local insurgent group structures and political fragmentation [15]. In Southeast Asia, numerous studies describe piracy as an entrenched problem driven by dense traffic lanes, narrow straits, and permissive geography, where small-boat operations allow rapid boarding and escape [16]. Empirical works consistently support the view that piracy hotspots persist due to spatial opportunity structures such as chokepoints, anchorage zones, and coastal proximity [16].

The behavioural dimension of piracy—attack type, target selection, and modus operandi—has been explored by security analysts and criminologists [17]. Boarding-type attacks are commonly associated with sudden close-range confrontations, which raises the likelihood of violence, especially in anchorages or low-speed contexts. Studies of weapon use (guns vs. knives vs. blunt tools) show that firearms substantially increase the probability of coercion, hostage-taking, or crew injuries [18,19]. Research also highlights that certain vessel types attract different offender strategies. Tankers and bulk carriers are traditionally viewed as high-value targets, while fishing vessels and offshore supply ships are often attacked opportunistically because of predictability and limited defensive posture [20].

Several studies emphasize the socio-economic determinants of maritime piracy, indicating that poverty, weak law enforcement, corruption, and high youth unemployment function as structural enablers of maritime crime [21,22,23,24,25]. Economic downturns and political instability are consistently associated with increased incident rates, particularly in regions where coastal communities depend heavily on maritime resources. Governance capacity—including the effectiveness of coastal state policing and the presence of international naval forces—shapes the extent to which piracy is suppressed, displaced, or allowed to persist.

Although early piracy studies relied heavily on narrative descriptions and case studies, the last decade observed a rise in quantitative, data-driven approaches [26,27,28]. Spatial risk models using GIS, kernel density estimation, or spatial autocorrelation have been widely utilized to visualize hotspots and predict incident clustering [26,29,30]. Some research focuses on machine learning algorithms—random forests, neural networks, or gradient boosting—to forecast incident likelihood, although these methods often lack interpretability and depend on highly curated input features [26,27,31]. In contrast, classical categorical-data statistics (chi-square tests, contingency tables, Cramér’s V) remain underutilized despite their transparency and suitability for large, heterogeneous datasets [27,28]. These methods allow researchers to quantify association strength between operational variables and violent outcomes without complex model assumptions. Works in security analysis and transportation safety repeatedly emphasize the importance of interpretable effect sizes for incident management, making such approaches particularly relevant for policy-oriented maritime security research.

Previous research has focused on descriptive statistical analysis and incident classification. Advanced predictive models have been relatively rarely used. The literature indicates that ML methods can improve incident prediction accuracy, but they require precise selection of input variables and model validation [32,33,34,35,36,37].

Despite the growing body of literature on maritime piracy, several methodological and conceptual limitations remain evident. A substantial proportion of existing studies relies on descriptive statistics or regional trend analysis, focusing primarily on incident frequency rather than on the mechanisms underlying violent escalation. While such approaches provide valuable situational awareness, they offer limited insight into the relative importance of specific operational and contextual factors associated with violence. These methodological and data-related limitations, frequently acknowledged in prior studies, collectively constrain the interpretability, comparability, and operational relevance of existing findings on maritime piracy violence.

In addition, many empirical studies are constrained by the quality and consistency of available piracy data. Underreporting of incidents, regional differences in reporting practices, and incomplete documentation of attack characteristics—particularly regarding the use of violence and weapons—introduce bias and reduce the comparability of results across regions and time periods. These data limitations complicate cross-study synthesis and weaken the generalizability of findings.

In recent years, machine learning and artificial intelligence methods have been increasingly applied to piracy risk prediction. Although these models often achieve higher predictive accuracy, they are frequently characterized by limited interpretability, strong dependence on feature engineering, and sensitivity to data completeness and quality. As a result, their applicability for operational decision-making and policy formulation remains constrained, particularly in maritime security contexts where transparency and explainability are essential.

Only a limited number of studies explicitly focus on violent escalation as a distinct analytical outcome rather than treating violence as a secondary or implicit attribute of piracy incidents. Moreover, relatively few studies systematically compare the strength and relative importance of associations between multiple categorical variables—such as geographic region, attack type, weapon use, vessel characteristics, flag state, or temporal factors—and the occurrence of violence during piracy incidents. Existing research often examines these variables in isolation or reports their influence using descriptive frequencies, without formally testing whether observed differences are statistically significant or substantively meaningful. In many cases, statistical significance is reported without accompanying effect-size measures, making it difficult to assess whether identified relationships are weak, moderate, or practically relevant. Without effect-size-based comparisons, there is a risk that security responses may be guided by statistically significant yet operationally marginal factors, while more influential drivers of violence remain underemphasized. This limitation is particularly problematic in large datasets, where statistically significant results may emerge even for marginal associations, potentially leading to misleading interpretations. The frequent absence of standardized and interpretable effect-size metrics—such as Cramér’s V—restricts the ability to rank risk factors according to their explanatory strength and to distinguish dominant drivers of violent escalation from secondary or contextual influences. Furthermore, inconsistencies in variable definitions, categorization schemes, and analytical frameworks across studies hinder cross-study comparability and the accumulation of coherent empirical evidence. As a result, maritime security practitioners and policymakers are often left without clear guidance on which operational or contextual factors should be prioritized when designing preventive measures, allocating security resources, or developing risk-based routing strategies. This lack of systematic, comparative assessment of categorical risk factors represents a significant methodological shortcoming in the piracy literature. Addressing this gap requires transparent and reproducible statistical approaches that not only test for dependence but also quantify the magnitude of associations in a manner that is both interpretable and operationally actionable.

Consequently, a methodological gap persists in the literature regarding transparent, reproducible statistical frameworks capable of quantifying the strength of association between piracy incident characteristics and violent outcomes. This study addresses this gap by applying chi-square tests of independence combined with Cramér’s V effect size measures to a multi-year global piracy dataset, providing an interpretable and operationally relevant assessment of key risk factors.

3. Research Objective

The objective of this research is to identify and evaluate the features most strongly associated with the occurrence and severity of maritime piracy attacks. The selected features serve as key inputs to predictive models designed to support decision-making processes in maritime security management, including route planning, threat detection, and preventive resource allocation.

Initial feature relevance was assessed using the chi-square independence test and Cramer’s V coefficient. Features with non-significant dependencies or effect size below 0.12 were excluded from further analysis.

4. Data and Methodology

4.1. Data Sources

The methodological framework of this study follows a structured, multi-stage analytical procedure. First, piracy incident data were compiled and standardized from multiple international sources. Next, categorical variables describing incident characteristics were defined and preprocessed. Finally, statistical association analysis was conducted using chi-square tests of independence and Cramér’s V to identify and compare factors associated with violent escalation.

The empirical basis for this study is a multi-year dataset covering reported maritime piracy and armed robbery incidents that occurred between 2015 and 2024. The core source of incident-level information is the ICC International Maritime Bureau (IMB), which maintains a global reporting system for maritime crime. The IMB Piracy and Armed Robbery Annual Reports and quarterly updates provide structured records that include incident dates, locations, vessel details, attack types, and outcomes of each event. These reports were supplemented with analytical summaries published by Oceans Beyond Piracy (OBP) within the State of Maritime Piracy series, providing qualitative and economic context regarding ransom practices, regional security dynamics, and crew impact.

Additional verification and classification support was obtained from International Maritime Organization (IMO) Maritime Safety Committee circulars, which supply standardized definitions of piracy and armed robbery under UNCLOS and SOLAS frameworks. This ensured conceptual consistency across regions and reporting bodies.

In

Table 1. Definition and measurement scale of analyzed variables. Author’s own elaboration based on IMB data.

Variable	Description	Type
Region	Geographic area where incident occurred	Nominal
Attack type	Boarding, attempted boarding, fired upon, hijacked	Nominal
Weapon type	Guns, knives, tools, or unarmed	Nominal
Vessel type	Tanker, bulk carrier, container ship, fishing, offshore, etc.	Nominal
Flag state	Ship registry nationality	Nominal
Month	Incident month (proxy for environmental seasonality)	Ordinal
Violence (target)	Binary outcome: violence vs. no violence	Dichotomous

the dataset includes variables describing

• Geographical region of the incident, harmonized across IMB, OBP, and IMO taxonomies;
• Type of attack, distinguishing boarding, hijacking, attempted attacks, and incidents involving firearms;
• Outcome of the attack, including whether the attack was successful or repelled,
• Vessel attributes, such as vessel class and operational status at the time of the incident;
• Forms of violence used against crews, including hostage-taking, kidnapping, threats, and injuries;
• Operational and environmental conditions, such as vessel speed, proximity to shore, anchorage status, and navigational corridor characteristics.

These variables collectively form the basis for quantitative risk modelling and the subsequent development of predictive indicators.

4.2. Data Preprocessing

Data preprocessing was conducted in several stages to standardize, clean, and structure the dataset for machine learning analysis. First, categorical variables were normalized to ensure consistency in spelling and naming conventions across sources, particularly for regional identifiers and vessel categories. Where geographical descriptions were originally recorded as port names or coastal zones, they were converted into standardized maritime regions to enable spatial comparability.

Records containing missing or ambiguous critical information (e.g., unspecified attack outcomes or unknown location) were removed to reduce noise and prevent model distortion. To examine fundamental patterns in the data, exploratory data analysis (EDA) was carried out, focusing on

• Temporal trends in piracy intensity across the 2015–2024 period;
• Spatial clustering patterns within high-risk maritime corridors;
• Distributions of attack methods and violence severity levels.

Violence was operationalized as a binary outcome variable, coded as 1 when incidents involved physical assault, hostage-taking, kidnapping, or explicit violent threats against crew members, and 0 otherwise. This classification follows standardized definitions used in IMB and IMO reporting frameworks.

This step enabled the identification of stable, recurring patterns in pirate behavior and regional operational risk exposure.

Table 2. Annual distribution of maritime piracy incidents by region (2015–2024). Author’s own elaboration based IMB data.

Region	2015	2016	2017	2018	2019	2020	2021	2022	2023	2024	Total
Africa	35	62	57	87	71	88	37	21	26	26	510
SE Asia	147	68	76	60	53	62	56	58	67	70	717
Indian Sub-Cont	24	17	15	18	4	10	2	10	5	16	121
Americas	8	27	24	29	29	30	36	24	19	2	228
East Asia	31	16	4	7	5	4	1	2	3	2	75
Rest of World	1	1	4	0	0	1	0	0	0	0	7
Total	246	191	180	201	162	195	132	115	120	116	1658

presents the annual distribution of reported maritime piracy incidents across six major geographic regions between 2015 and 2024. The data summarize the total number of attacks recorded each year and highlight clear spatial patterns in global piracy activity over the decade.

The Southeast Asia region consistently reports the highest number of incidents, accounting for 717 cases and maintaining its position as the dominant hotspot throughout most of the study period. Africa follows with 510 incidents, reflecting the persistent concentration of violent and organized piracy groups in areas such as the Gulf of Guinea and East African waters. Other regions—namely the Americas, the Indian Sub-Continent, and East Asia—show significantly lower but fluctuating incident counts, indicating episodic increases in opportunistic or region-specific piracy. The “Rest of World” category contributes only marginally, demonstrating that piracy remains highly geographically concentrated.

Overall, the table illustrates both temporal variability and strong regional clustering, with global totals ranging from 115 to 246 incidents per year. These patterns reinforce the importance of region-specific security dynamics and provide the contextual background for subsequent statistical association analysis using the chi-square test and Cramér’s V.

4.3. Feature Selection and Modelling

The analytical procedure was grounded entirely in classical categorical-data statistics. Each variable describing piracy incidents was evaluated using chi-square independence tests, followed by the computation of Cramér’s V to quantify effect size. This two-step approach enabled the detection of statistically significant associations and the ranking of their relative strengths, without relying on predictive modelling or machine learning assumptions.

The use of chi-square tests of independence is appropriate given the categorical nature of all analyzed variables, while Cramér’s V was selected to provide a standardized and comparable measure of association strength across variables with differing category counts.

The dataset covering maritime piracy incidents recorded between 2015 and 2024 was subjected to a structured analytical procedure aimed at assessing the relative strength of association between each explanatory variable and the target outcome variable, defined as the presence or absence of violence during the incident. For each categorical feature—specifically region, attack type, weapon type, vessel type, flag state, and month—the relationship with the violence variable was evaluated using the Chi-square test of independence. This allowed us to determine whether the distribution of violent vs. non-violent incidents differed significantly across the categories of each feature.

Initial screening using chi-square and Cramér’s V allowed the identification of variables exhibiting meaningful explanatory power. Variables below the interpretation threshold (V < 0.10–0.12) were considered to have negligible association and were excluded from further discussion.

4.3.1. Chi-Square Test of Independence

To determine whether each variable is statistically associated with the occurrence of violence, the chi-square test of independence was applied [38]:

$${\chi }^2=\sum {\displaystyle \frac{(O_{ij}-E_{ij}{)}^2}{E_{ij}}}$$

(1)

where $O_{ij}$ are observed frequencies and $E_{ij}$ expected frequencies.

Table 3. Chi-Square Test of Independence Between Year of Attack and Region (2015–2024). Author’s own elaboration.

Year	Count % Within Attacks by Years	Africa	SE ASIA	Indian Sub-Cont	Americas	East Asia	Rest of World	Total
2015	Count	35	147	24	8	31	1	246
	Expected Count	75.67	106.38	17.95	33.83	11.13	1.04	246.00
	% within attacks years	14.23%	59.76%	9.76%	3.25%	12.60%	0.41%	100.00%
	χ2	21.86	15.51	2.04	19.72	35.49	0.00	0.00
2016	Count	62	68	17	27	16	1	191
	Expected Count	58.75	82.60	13.94	26.27	26.27	0.81	191.00
	% within attacks years	32.46%	35.60%	8.90%	14.14%	8.38%	0.52%	100.00%
	χ2	0.18	2.58	0.67	0.02	4.01	0.05	0.00
2017	Count	57	76	15	24	4	4	180
	Expected Count	55.37	77.84	13.14	24.75	8.14	0.76	180.00
	% within attacks years	31.67%	42.22%	8.33%	13.33%	2.22%	2.22%	100.00%
	χ2	0.05	0.04	0.26	0.02	2.11	13.81	0.00
2018	Count	87	60	18	29	7	0	201
	Expected Count	61.83	86.92	14.67	27.64	9.09	0.85	201.00
	% within attacks years	43.28%	29.85%	8.96%	14.43%	3.48%	0.00%	100.00%
	χ2	10.25	8.34	0.76	0.07	0.48	0.85	0.00
2019	Count	71	53	4	29	5	0	162
	Expected Count	49.83	70.06	11.82	22.28	7.33	0.68	162.00
	% within attacks years	43.83%	32.72%	2.47%	17.90%	3.09%	0.00%	100.00%
	χ2	8.99	4.15	5.18	2.03	0.74	0.68	0.00
2020	Count	88	62	10	30	4	1	195
	Expected Count	59.98	84.33	14.23	26.82	8.82	0.82	195.00
	% within attack years	45.13%	31.79%	5.13%	15.38%	2.05%	0.51%	100.00%
	χ2	13.09	5.91	1.26	0.38	2.63	0.04	0.00
2021	Count	37	56	2	36	1	0	132
	Expected Count	40.60	57.08	9.63	18.15	5.97	0.56	132.00
	% within attacks years	28.03%	42.42%	1.52%	27.27%	0.76%	0.00%	100.00%
	χ2	0.32	0.02	6.05	17.55	4.14	0.56	0.00
2022	Count	21	58	10	24	2	0	115
	Expected Count	35.37	49.73	8.39	15.81	5.20	0.49	115.00
	% within attacks years	18.26%	50.43%	8.70%	20.87%	1.74%	0.00%	100.00%
	χ2	5.84	1.37	0.31	4.24	1.97	0.49	0.00
2023	Count	26	67	5	19	3	0	120
	Expected Count	36.91	51.89	8.76	16.50	5.43	0.51	120.00
	% within attacks years	21.67%	55.83%	4.17%	15.83%	2.50%	0.00%	100.00%
	χ2	3.23	4.40	1.61	0.38	1.09	0.51	0.00
2024	Count	26	70	16	2	2	0	116
	Expected Count	35.68	50.16	8.47	15.95	5.25	0.49	116.00
	% within attacks years	22.41%	60.34%	13.79%	1.72%	1.72%	0.00%	100.00%
	χ2	2.63	7.84	6.71	12.20	2.01	0.49	0.00

presents the results of the chi-square test conducted for the relationship between the year of attack and the region in which the incident occurred. The test confirmed that piracy incidents are unevenly distributed across regions over time, with statistically significant deviations between observed and expected frequencies (p < 0.05). This indicates that the geographical concentration of attacks changes across years rather than remaining constant, suggesting dynamic shifts in regional piracy hotspots.

The same chi-square and Cramér’s V procedure was applied analogously to all other categorical variables included in the study—attack type, weapon type, vessel type, flag state, and month—to ensure a consistent and comparable assessment of association strength across the full set of piracy-related features.

4.3.2. Effect Size: Cramér’s V

Because chi-square significance alone does not indicate strength of association, Cramér’s V was computed [39]:

$$V=\sqrt{{\displaystyle \frac{{\chi }^2}{n(k-1)}}}$$

(2)

All statistical analyses were conducted using standard statistical software. Expected cell frequencies were examined to ensure the validity of chi-square assumptions, and records with missing critical variables were excluded from the analysis to prevent distortion of association measures.

A pre-selection threshold of V ≈ 0.12 was applied, informed by common practice in maritime criminal risk modelling and supported by exploratory dependency structure inspection.

This analytical approach ensured that feature selection was empirically justified, reducing the influence of noise and preventing the inclusion of variables lacking predictive relevance [40]. The retained variables then proceeded to the model-based feature importance evaluation and predictive modelling stages. To interpretative framework provides standardized thresholds for evaluating the magnitude of associations identified through Cramér’s V. Establishing these thresholds enables consistent classification of each variable’s influence on the occurrence of violence across the dataset (

Table 4. Interpretation Thresholds for Cramér’s V Effect Size. Author’s own elaboration.

V Value	Interpretation
0.00–0.10	negligible association
0.10–0.20	weak association
0.20–0.30	moderate association
>0.30	strong association

).

Following the Chi-square test, the Cramér’s V coefficient was calculated in order to quantify the effect size of each association and to allow for direct comparison across variables with different numbers of categories.

Table 4 presents the interpretation categories for Cramér’s V effect size, ranging from negligible to strong associations. These thresholds allow for a systematic assessment of how strongly each independent variable is related to violent outcomes in piracy incidents. Values below 0.10 indicate minimal explanatory power, whereas coefficients exceeding 0.20 denote moderate to meaningful associations, and those above 0.30 reflect strong contextual influence. Applying this classification framework ensures that subsequent analytical decisions—such as variable retention or exclusion—are grounded in clear, replicable criteria.

This metric served as the basis for the feature pre-selection stage, where variables demonstrating a non-negligible association with the target (typically Cramér’s V ≥ 0.12) were retained for further modelling.

According to Cramér’s V interpretation framework presented in Table 4, each categorical variable included in the analysis was assigned an association strength category. Based on the computed V values, the piracy-related features demonstrated weak to moderate relationships with the occurrence of violence. Variables with V values close to or above the threshold of 0.12 were classified as exhibiting a moderate association, whereas those below this level were interpreted as having a weak association. This classification enabled a structured comparison of the relative influence of each attribute—such as region, attack type, weapon type, vessel type, flag state, and month—on the likelihood of violent outcomes in piracy incidents

Table 5. Chi-Square and Cramér’s V Results for Year–Region Association. Author’s own elaboration.

Variable	Value	df	Asymp. Sig. (2-Sided)
Person Chi-Square	270.18	45	0.00
Likelihood Ratio	273.1	45	0.00
Linear by Linear relation.	3.5	1	0.06
Cramer’s V	0.181	-	0.00
Number of Valid Cases	1658	-	-

summarizes the results of the chi-square test of independence conducted for the relationship between the year of attack and the region in which the incident occurred. The Pearson chi-square value indicates a statistically significant association, meaning that the geographical distribution of piracy incidents changed over the 2015–2024 period. The Cramér’s V coefficient (V = 0.181) suggests a weak-to-moderate strength of association, reflecting measurable but not dominant shifts in the geographical patterns of piracy activity across the examined decade.

This effect size implies that while temporal fluctuations exist—particularly in high-risk regions such as Southeast Asia and Africa—the spatial concentration of incidents remains partly shaped by enduring regional risk dynamics. Table 5 therefore provides a quantitative foundation for interpreting long-term spatial trends in maritime piracy and supports subsequent comparisons of association strength across other variables evaluated in the study.

Table 6. Ranking of Variable Influence. Author’s own elaboration.

Variable	Cramér’s V	Strength of Association
Violence type	0.252	Moderate
Region	0.181	Weak–Moderate
Attack type	0.127	Weak
Weapon type	0.126	Weak
Flag state	0.121	Weak
Vessel type	0.120	Weak
Month	0.102	Negligible–Weak

presents the chi-square statistics and corresponding Cramér’s V coefficients for all operational and contextual variables included in the study. The results confirm that each variable shows a statistically significant association with the occurrence of violence in maritime piracy incidents (p < 0.05). However, the strength of these associations varies across features. The screening conducted on the 2015–2024 dataset yielded the following effect sizes (Table 6). The resulting screening revealed that violence type and region exhibited the strongest measurable associations with violent outcomes, while attack type, weapon type, vessel type, and flag displayed moderate but operationally meaningful effect sizes. Month showed a comparatively weaker association, suggesting that seasonal variation plays only a limited role in determining the probability of violence.

This overview enables a direct comparison of explanatory relevance across variables and highlights those factors that most strongly contribute to the likelihood of violent outcomes. The table therefore serves as a central reference point for interpreting the relative importance of each feature within the broader risk environment of maritime piracy.

5. Results

The chi-square test of independence confirmed that the occurrence of violence in maritime piracy incidents is not randomly distributed across the examined categorical variables. For each feature, contingency tables comparing violent vs. non-violent outcomes were constructed, and the chi-square statistic indicated statistically significant associations at p < 0.05. To assess the strength of these associations, Cramér’s V was calculated for each variable.

Table 7. Strength of association between selected variables and occurrence of violence (Cramér’s V, 2015–2024). Author’s own elaboration.

Variable	Pre-Selection Decision
Violence type	Retained as key determinant
Region	Retained
Attack type	Retained due to operational relevance
Weapon type	Retained for interaction effects
Flag state	Retained for contextual influence
Vessel type	Retained conditionally
Month	Included for testing seasonality effects

presents the strength of association between selected variables and the occurrence of violence, measured using Cramér’s V, and summarizes the pre-selection decisions for variables retained in the analysis for the period 2015–2024.

The geographical region demonstrated the strongest association with violent outcomes, with Southeast Asia and the Gulf of Guinea showing the highest proportion of attacks involving physical coercion, hostage-taking, or kidnapping for ransom. This supports findings from previous maritime security studies indicating that localized socio-political instability and the presence of organized piracy networks shape operational risk environments.

Attack type also exhibited a meaningful association with violence. Incidents involving direct boarding were more likely to escalate into physical confrontation, suggesting that close-range engagement is a critical determinant of crew safety outcomes.

Although weapon type, vessel type, and flag state displayed weaker levels of association, their effects are operationally relevant. For example, incidents involving firearms were associated with significantly higher levels of intimidation and hostage-taking, while large, high-value cargo vessels were more frequently targeted by groups engaging in planned, profit-driven attacks.

Finally, seasonality (month) showed the weakest association, indicating that operational opportunity, rather than environmental factors, drives violent escalation.

6. Discussion

6.1. Interpretation of Findings in Relation to Existing Literature

The findings of this study provide empirical support for the view that violent escalation in maritime piracy incidents is shaped primarily by spatial and tactical factors rather than by vessel-specific or seasonal characteristics. By systematically comparing the strength of association between multiple categorical variables and violent outcomes, this research contributes to a more nuanced understanding of piracy-related violence within the existing body of literature.

The results indicate that geographic region exhibits the strongest association with the occurrence of violence during piracy incidents. This finding is consistent with previous studies emphasizing the regional concentration of piracy and the role of localized socio-political and security conditions in shaping attack dynamics. Prior research identifies the Gulf of Guinea, Southeast Asia, and the Western Indian Ocean as persistent high-risk areas characterized by differing operational patterns and levels of violence. In particular, studies focusing on the Gulf of Guinea report a high prevalence of armed attacks, kidnapping for ransom, and organized criminal activity, which aligns with the elevated violence levels observed in this study. The moderate effect size identified for region suggests that while spatial context is not the sole determinant of violence, it constitutes a dominant structural factor influencing escalation risk. This supports earlier qualitative and spatial analyses that link regional governance capacity, enforcement effectiveness, and criminal organization structures to piracy severity, while providing a quantitative measure of their relative importance.

Attack type also demonstrated a meaningful association with violent outcomes, particularly in cases involving direct boarding. This result corroborates prior situational and behavioural analyses indicating that boarding operations inherently increase the likelihood of physical confrontation due to close proximity between attackers and crew.

Earlier research has highlighted that low vessel speed, anchorage status, and reduced situational awareness facilitate such encounters, often resulting in intimidation or physical harm. By quantifying this relationship using Cramér’s V, the present study strengthens existing descriptive findings and confirms that attack modality plays a measurable role in shaping escalation dynamics, although its influence remains secondary to regional context.

The association observed between weapon type and violent outcomes in this study is consistent with a substantial body of prior research indicating that the presence of firearms significantly increases the severity and coercive potential of maritime piracy incidents. Previous studies have shown that firearms are most often employed as instruments of intimidation rather than as means of direct lethal violence, serving primarily to assert control over crews and to facilitate compliance during boarding or kidnapping operations. Empirical evidence from regions such as the Gulf of Guinea and Southeast Asia demonstrates that even when physical injuries are limited, the display or use of firearms markedly elevates the likelihood of hostage-taking, prolonged detention, and psychological trauma among seafarers. The relatively weak yet statistically significant Cramér’s V coefficient identified in this study suggests that weapon type alone does not constitute a dominant driver of violent escalation. Instead, weapon use appears to function as an escalation amplifier that interacts with other contextual and operational factors, particularly geographic region and attack type. This finding aligns with earlier qualitative and region-specific studies that emphasize the importance of organized group structures, tactical objectives, and local security environments in shaping how and when weapons are deployed during piracy incidents. Importantly, the results refine previous conclusions by quantitatively situating weapon use within a broader hierarchy of risk factors. While earlier research often implicitly assumes a strong causal relationship between firearms and violence, the present analysis demonstrates that the explanatory power of weapon type is conditional and comparatively limited when assessed alongside spatial and tactical variables. This distinction is operationally relevant, as it suggests that the mere presence of weapons should not be interpreted in isolation when assessing escalation risk. Rather, effective threat assessment requires simultaneous consideration of regional context, offender modus operandi, and engagement proximity.

In contrast to regional and tactical variables, vessel type, flag state, and month of occurrence exhibited weak associations with violent outcomes. While previous studies often identify certain vessel classes as preferred targets, the present findings suggest that these characteristics exert limited influence on whether an attack becomes violent once it occurs. Similarly, the minimal role of seasonality supports earlier observations that piracy violence is driven more by operational opportunity and criminal intent than by environmental or temporal conditions. These results challenge assumptions that structural vessel attributes alone can serve as reliable predictors of violent escalation and underscore the importance of contextual and behavioural factors.

Beyond substantive findings, this study offers a methodological contribution to maritime piracy research by demonstrating the utility of classical categorical-data statistics for violence-focused risk assessment. Unlike many machine learning-based approaches, the chi-square and Cramér’s V framework provides transparent and interpretable measures of association strength, facilitating direct comparison between risk factors. This addresses a recurring limitation in the literature, where statistical significance is often reported without adequate consideration of practical relevance.

6.2. Practical Implications for Maritime Security and Piracy Prevention

The results of this study yield important practical implications for maritime piracy prevention, risk assessment, and operational security planning. By empirically identifying the contextual and operational factors most strongly associated with violent escalation, the analysis offers an evidence-based foundation for prioritizing preventive strategies and optimizing the allocation of limited maritime security resources.

First, the strong association between geographic region and violent outcomes highlights the necessity of region-specific security strategies. The results confirm that violence is concentrated in particular maritime areas characterized by organized criminal networks and weak governance structures, such as the Gulf of Guinea and parts of Southeast Asia. This suggests that uniform, globally applied security measures are unlikely to be equally effective across all regions. Instead, preventive efforts should be tailored to regional risk profiles, incorporating localized threat assessments, intelligence sharing, and cooperation with coastal states. Shipping companies and maritime authorities may use region-specific risk classifications derived from statistical association measures to inform route planning, insurance assessments, and crew preparedness protocols.

Second, the significant association between attack type—particularly boarding incidents—and violent escalation has clear operational relevance for maritime security planning. Boarding operations inherently involve close physical proximity between attackers and crew, substantially increasing the likelihood of confrontation, intimidation, and coercive violence. Successful boarding represents a critical escalation threshold, as it creates immediate control asymmetries and severely limits the crew’s ability to disengage. The findings empirically support earlier situational analyses indicating that preventing boarding is more effective for reducing violence risk than focusing solely on post-incident response. Preventive strategies should therefore prioritize early disruption of boarding attempts through navigational, procedural, and technical measures. In practical terms, this includes maintaining safe transit speeds where feasible, implementing physical barriers such as reinforced access points and controlled ladder zones, and strengthening lookout procedures through dedicated watches and surveillance systems. The use of non-lethal deterrence measures—such as water cannons, acoustic devices, and evasive maneuvering—can further reduce the likelihood of close contact. Equally important is crew preparedness. Training programs should emphasize early detection, threat recognition, and avoidance strategies tailored to boarding scenarios, particularly in low-speed environments, port approaches, and anchorage areas. By focusing on interrupting boarding attempts before physical contact occurs, maritime operators can significantly reduce the probability of violent escalation and enhance crew safety.

Third, although weapon type shows a weaker association than region and attack modality, the presence of firearms functions as an important escalation amplifier. This indicates that risk assessments should account not only for weapon presence but also for its interaction with regional context and tactical intent. In high-risk regions, the elevated likelihood of firearm-based intimidation or kidnapping supports the use of enhanced protective measures, such as armed guards, citadels, and secure communication procedures. Importantly, this interaction-based perspective aligns with prior research emphasizing that weapons increase coercive potential primarily when embedded within organized and region-specific piracy strategies. At the same time, the findings caution against treating weapon presence as a standalone predictor of violence, underscoring the need for multi-factor risk evaluation.

Fourth, the weak associations for vessel type, flag state, and seasonality indicate that these factors play a secondary role in shaping violent outcomes once an attack occurs. This suggests that piracy prevention should prioritize situational awareness, tactical avoidance, and region-specific intelligence rather than assumptions linked to vessel class or flag. The findings also call into question the strong emphasis placed on vessel attributes in existing insurance and regulatory risk frameworks.

More broadly, the methodological framework employed in this study offers a practical tool for maritime security stakeholders. The use of chi-square tests and Cramér’s V provides transparent, interpretable metrics that can be integrated into decision-support systems without the complexity of machine learning models. Such metrics can support the development of risk matrices, escalation indicators, and early-warning systems that are understandable to ship operators, security officers, and policymakers alike. Finally, the findings underscore the importance of data-driven prioritization in piracy prevention. By quantifying the relative strength of associations rather than relying solely on statistical significance or anecdotal evidence, the study helps distinguish between dominant drivers of violence and factors of marginal operational relevance. This approach can reduce the risk of misallocating resources toward statistically significant but practically weak predictors, ensuring that preventive strategies target the most influential determinants of violent escalation.

7. Limitations and Future Research

Despite the contributions of this study, several limitations should be acknowledged. First, the analysis relies on secondary data obtained from international piracy reporting databases, including IMB, OBP, and IMO sources. While these datasets represent the most comprehensive publicly available records of maritime piracy incidents, they are subject to reporting bias and inconsistencies. Underreporting of piracy events—particularly in regions where reporting mechanisms are weak or where ship operators may avoid disclosure—remains a well-documented issue. Additionally, variations in reporting practices across regions and time periods may affect the completeness and comparability of incident records, especially with regard to detailed information on violence, weapon use, and attack outcomes.

Second, the study adopts a categorical data framework, which necessitates the aggregation of complex operational phenomena into discrete categories. Although this approach enhances transparency and interpretability, it may obscure within-category heterogeneity. For example, the binary classification of violence does not capture gradations in severity, such as distinctions between threats, physical injury, and prolonged hostage situations. Similarly, categorical representations of attack type or weapon use may oversimplify tactical variations that influence escalation dynamics at a finer operational level.

Third, the application of chi-square tests of independence and Cramér’s V effect size measures enables the identification and comparison of statistical associations between categorical variables but does not allow for causal inference. The observed relationships therefore reflect dependency structures within the data rather than direct cause–effect mechanisms driving violent escalation in piracy incidents. As a result, the findings should be interpreted as indicative of relative risk patterns and structural tendencies rather than as deterministic or predictive rules governing pirate behaviour. In particular, the analytical framework employed in this study does not capture the directionality or temporal sequencing of events that may influence escalation dynamics. For example, while significant associations are observed between region, attack type, weapon use, and violent outcomes, the applied methods cannot determine whether specific operational choices precede violence or emerge as a consequence of evolving encounter conditions. This limitation is inherent to cross-sectional categorical analyses and constrains the extent to which underlying behavioural mechanisms can be inferred. Furthermore, potential interaction effects between variables are not explicitly modelled within the current framework. Complex combinations of factors—such as the joint influence of geographic region, attacker modus operandi, and weapon availability—may exert a compounded effect on violence risk that exceeds the explanatory power of each variable considered in isolation. While Cramér’s V provides a useful comparative measure of association strength, it does not account for higher-order dependencies or non-linear relationships that may be present in real-world piracy dynamics. Consequently, the analytical results should be understood as a foundational screening of key risk factors rather than a comprehensive explanatory model of violent escalation. The findings are best suited for identifying structurally relevant variables and prioritizing areas for further investigation, rather than for generating causal claims or operational forecasts. Addressing these limitations would require the integration of multivariate modelling approaches, longitudinal designs, or mixed-method analyses capable of capturing interaction effects, temporal dependencies, and contextual decision-making processes among piracy actors.

These limitations suggest several avenues for future research. Subsequent studies could integrate severity-scaled violence indicators or ordinal outcome variables to capture escalation intensity more precisely. Combining categorical association analysis with multivariate or mixed-method approaches may also allow for the examination of interaction effects and conditional dependencies between risk factors. Furthermore, the integration of contextual data—such as naval presence, coastal governance indicators, or socio-economic metrics—could enhance the explanatory power of region-based findings.

Future research may also explore hybrid analytical frameworks that combine interpretable statistical methods with machine learning techniques. Such approaches could retain transparency while improving predictive capability, particularly for early-warning or decision-support systems. Finally, expanding longitudinal analyses to assess temporal shifts in association strength would provide valuable insight into how piracy violence evolves in response to policy interventions, security operations, and regional instability.

8. Conclusions

This study provides an evidence-based assessment of the factors associated with violent escalation in maritime piracy incidents reported between 2015 and 2024. By applying chi-square tests of independence alongside Cramér’s V effect size, the analysis demonstrates that violence is not randomly distributed across incident characteristics but is instead shaped by a combination of spatial, tactical, and contextual elements.

The results show that geographical region exerts the strongest influence, confirming that the likelihood of violent escalation is closely tied to regional security dynamics, governance capacity, and the operational sophistication of local piracy networks. Attack type and weapon use also contribute meaningfully to the risk of violence, highlighting that close-contact scenarios—particularly boarding events—substantially raise the probability of crew intimidation, hostage-taking, or physical harm. In contrast, variables such as vessel type, flag state, and seasonal timing exhibit significantly weaker associations, suggesting that structural vessel characteristics or environmental conditions play a secondary role compared to spatial and tactical determinants.

Overall, the findings reinforce the importance of region-specific threat assessments and targeted mitigation strategies. The analytical framework used in this study demonstrates the value of classical categorical statistics for maritime security research, offering transparent, interpretable insights that can inform risk evaluation even when machine learning models are not employed. The study’s results may support the development of adaptive routing strategies, enhance crew preparedness protocols, and contribute to regional maritime security planning. Future research may integrate these statistically validated features into predictive or early-warning systems, allowing for more proactive and data-driven approaches to piracy risk management.