Empirically Predicting When States Comply with the I.C.J.

I. Introduction

The question of whether the International Court of Justice (I.C.J.) is effective has long been a contentious topic and the subject of multiple research papers. This paper is one of great importance as the I.C.J. has been the “principal judicial organ” of the United Nations since its inception in 1945.1 The I.C.J. plays a dual role in the international legal system by serving as the forum for the peaceful resolution of interstate disputes as well as the body that articulates and clarifies principles of the law.2 Yet despite its importance, its jurisdiction is entirely based on state consent. The precursor to the current I.C.J., the Permanent Court of International Justice, justified this decision by declaring that “[i]nternational law governs relations between independent States. The results of law binding upon States therefore emanate from their own free will . . . . Restrictions upon the independence of States cannot therefore be presumed.”3 This view espouses positivism, the concept of state consent and the basis for international law.4 The current I.C.J. continues to follow this guideline by having states voluntarily submit to the court’s jurisdiction either by treaty, special agreement, or a general declaration under Article 36(2) of the I.C.J. statute that is sometimes referred to as compulsory jurisdiction.5 The I.C.J.’s consent-based model emphasized the respect and primacy of state sovereignty. However, it has also drawn criticism for inadvertently producing uneven enforcement and participation.6 There is concern that if the rules in international law are not enforced by “an effective structure of coercion comparable to a national police force,” the court’s ability to compel compliance or consistently shape international law will be weakened.7 If a sovereign state believes that a ruling will go against their self-interest, there is a risk of noncompliance. To skeptics of the I.C.J., this structure risks the court’s judgments devolving into mere advice. “[E]ven old law cannot survive if enough states, or a few powerful and influential ones, reject it.”8

There is evidence that even those intimately involved with the I.C.J. and the creation of international law share these concerns. At the 2014 Annual Meeting of the American Society of International Law, then-I.C.J. Judge Joan E. Donoghue acknowledged this persistent tension in international law.9 She recounted moments when she would be asked whether states truly follow the Court’s judgment, a seemingly straightforward question. However, that simple yes-or-no question would turn into a “thicket” of complexity.10 In a moment of rare judicial frankness, Judge Donoghue admitted that at times what the I.C.J. “says sounds simply like truisms.”11

A large literature has continued to examine the question of whether I.C.J. enforcement is an illusory concept, with only a handful of this research based on empirical research. Those that do exist have not been recently revisited.12 This paper examines data on state compliance with rulings of the I.C.J. that specifically asked a state to comply either behaviorally or through remedial actions. Multiple independent variables were tested against these rulings to see whether they had statistical importance. By doing so, I test the claim that state compliance with I.C.J. decisions is largely unpredictable such as being shaped more by self-interest or geopolitical factors rather than by legal reasoning or institutional authority. The null hypothesis is that compliance is not systematically predictable. If this is found, then it would suggest that a state’s decision to comply or not is not reliably associated with identifiable legal, political, or structural factors. An ideal study would be able to determine whether state compliance in each case was lawful, however this would require assessing the normative strength of each judgement before observing whether states obey when they “should.” Clearly, this approach is impractical as the legal merits of many international disputes are complex. Additionally, states may fail to comply even with well-reasoned rulings for reasons unrelated to their legality or normative value.

Instead, this study focuses on testing whether compliance can be predicted based on observable variables such as economic status, institutional design, or domestic legal mobilizations. Using a combinatorial regression model on a dataset of contentious I.C.J. decisions, this study seeks to identify which features, if any, predict compliance. If such predictors emerge, it would suggest that I.C.J. compliance is not random but conditional and structured. The findings speak not only to the I.C.J.’s institutional effectiveness, but to the broader question of whether international adjudication operates as law or merely legalized diplomacy.

This comment proceeds in three parts. Part I will outline the theoretical and institutional background of the I.C.J. as well as the persistent debates about its effectiveness. Part II will describe the methodology of the case selection and data classification process. Part III will detail the modeling approach and test the hypothesis.13

II. Background

The I.C.J.’s jurisdiction is laid down in both the Charter of the U.N. as well as in the statute of the court.14 The I.C.J. can hear a dispute in three types of cases: (1) cases authorized by a treaty that provides that future disputes arising from under the treaty will be adjudicated by the I.C.J;15 (2) “special agreements” where the parties agree to submit their case to the court regarding a specific dispute; and 3) cases between states that have declared themselves subject to the “compulsory jurisdiction” of the court, meaning that the State has opted into I.C.J. jurisdiction over any international legal dispute involving the state.16 As of now, there are 75 states that have accepted I.C.J. compulsory jurisdiction.17

When exercising its judicial functions, the I.C.J. distinguishes between two primary types of cases: contentious cases in which the states are parties to a legal dispute and the Court’s ruling is binding, and advisory opinions, which are non-binding legal interpretations issued at the request of the U.N. General Assembly, Security Council, or other authorized international bodies.18 While advisory opinions carry significant normative and political weight, they do not impose legal obligations on specific states.19 As such, they serve a different function than contentious rulings which seek to result in direct findings of legal responsibility and instructions for future state behavior such as demands for reparations, cessation of conduct, or other forms of compliance. This study will focus primarily on the contentious rulings.

Since its inception, the I.C.J. has rendered decisions in a little over 170 total cases.20 Of these, 51 constitute contentious cases in which at least one state is called upon to take or refrain from specific action under the court’s judgment.21 These cases form the empirical basis for this study. Unlike advisory opinions, contentious cases offer the opportunity to test whether legal adjudication translates into behavioral outcomes.

A. Defining Noncompliance

Despite the formal avenues that the I.C.J. provides for adjudication in contentious cases, states will sometimes attach reservations or withdraw from jurisdiction altogether, as the United States and France have done in notable cases.22 International law’s grounding in consent impacts the court’s jurisdiction at the outset of a dispute, but it can also play a significant role during proceedings. In the rare instance when a state refuses to comply with the court’s ruling, there is no mechanism to compel obedience.  Cases where a state chooses to withdraw from the court’s authority during or after proceedings raise deeper concerns about what happens after a judgment is issued.23 As a purposeful choice, in these instances, neither the statute of the court nor its rules provide a mechanism for enforcing judgments.24 Therefore, though final judgments are generally followed, in instances where a state refuses to adhere to the I.C.J.’s ruling, the court lacks an enforcement arm. Even if the I.C.J. had an enforcement arm, enforcing adherence to rulings would be further complicated by the fact that compliance can often be a matter of degrees with states partially implementing decisions, delaying actions, or contesting interpretations.25 In this study, compliance is measured by a state’s measurable actions and whether it aligned with the I.C.J.’s ordered remedy. A few examples follow below.

Corfu Channel (1949).26 This was the first contentious case that the I.C.J. ever adjudicated. This case involved Albania’s failure to warn British warships about naval mines in Albanian territorial waters which led to loss of human life and significant damages.27 The I.C.J. found Albania responsible for failing to notify ships of the danger and ordered a sum in monetary damages to the United Kingdom.28 Albania refused to pay the sum for the next half century.29 Though Albania eventually complied, due to the significant time between the ruling and actual compliance, this case is better classified as an example of lesser compliance or noncompliance.

Military and Paramilitary Activities in and Against Nicaragua (1986).30 This case involved Nicaragua’s claim that the U.S. had violated international law by arming and supporting “contra” rebel forces, mining Nicaraguan ports, and attacking ports, oil installations, and a naval base.31 The I.C.J. ruled in favor of Nicaragua and ordered reparations, finding that the U.S. had breached both Nicaragua’s sovereignty as well as obligations under international law.32 Before the ruling had been issued, the U.S., attempted to withdraw from the proceedings and rejected the court’s jurisdiction.33 After the ruling, the U.S. refused to comply with the judgment, never paid reparations, and continued to challenge the court’s authority.34 This case would become one of the most infamous cases of a state refusing to respect the I.C.J.’s decision due to the U.S.’s significant presence in international law, and the overt nature of the snub. This case is classified as an example of noncompliance.

Territorial Dispute (Libya/Chad) (1994).35 This case concerned a long-standing conflict over the “Aouzou Strip,” a mineral-rich region of about 44,000 square miles along the Libya-Chad border.36 Libya argued that there was no existing boundary between the countries whereas Chad argued that there was a preexisting border.37 The I.C.J. ruled in favor of Chad, upholding a 1955 treaty that recognized Chad’s sovereignty over the disputed area.38 Libya initially rejected the verdict and began reinforcing troops in the area.39 However, after a month of negotiations, Col. Muammar el-Qaddafi signaled that he would comply with the I.C.J.40 Although Libya ultimately complied with the I.C.J.’s ruling, its initial resort to military escalation places the episode in the category of partial compliance or lesser compliance.

Avena and Other Mexican Nationals (Mexico v. United States) (2004).41 This case concerned the U.S.’s failure to inform 52 Mexican nationals on death row of their right to consular assistance under the Vienna Convention.42 The I.C.J. ruled that the U.S. had violated its international obligations and ordered it to review the convictions and sentences of the affected individuals.43 While the U.S. acknowledged the judgment of the court, it did not enforce comply within the relevant U.S. states resulting in several of the individuals in the case still being executed.44 Therefore, this case is classified as partially noncompliant due to the U.S.’s failure to implement the required judicial review and stay the executions.

III. Hypothesis

Although they do not make up the majority of contentious I.C.J. cases, there are numerous instances in which states challenge the I.C.J.’s jurisdiction or refuse to comply altogether. Scholars have proposed a range of motives for why states may choose to skirt the lines of I.C.J. enforcement, or at least why states feel they can do so. One proposed explanation is that economically advantaged states may believe that they can refuse compliance with the I.C.J. without incurring meaningful economic consequences.45 This feeling may become stronger if the state they are opposing in litigation is far weaker than they are in economic strength. This has been speculated by some scholars as part of the U.S.’s justification for withdrawing from participation in Nicaragua v. United States: a world power’s refusal to “accept defeat gracefully.”46 If the legal dispute is about a military conflict and there is a substantial difference in military strength, this, too, may lead to a feeling that the stronger state need not comply, especially if the judgment goes against the stronger state’s perceived self-interest. Noncompliance may also stem from perceived conflicts with national security or strategic interests. For example, in Territorial Dispute (Libya/Chad), both countries in the dispute were concerned with the military tensions, with Libya being influenced enough to initially defy the ruling.47

Another reason that states may not comply is that they do not agree with the I.C.J. having jurisdiction over the proceedings. States can appeal the I.C.J.’s jurisdiction, but these appeals are not always successful, and it is possible that the I.C.J. may proceed if it deems it does have jurisdiction. This was another central feature in Nicaragua v. United States and part of the U.S.’s reasoning when they withdrew from the proceedings.48 The U.S. State Department took the view that the conflict had to be resolved through political and diplomatic means rather than judicial resolution.49 Additionally, the U.S. believed that the court had ignored the original intentions of the U.N. charter making jurisdiction inappropriate.50

In 1987, scholar Jonathan Charney empirically studied state compliance with final decisions.51 He found that states generally comply with I.C.J. decisions. He also found that cases that were submitted by special agreement—situations where states agree to submit their case to the court regarding a specific dispute—retained a high degree of compliance. However, Charney also found that if a party showed initial reluctance to submit themselves to the Court’s judgment, they were far less likely to comply with the ruling.52

However, it is not the purpose of this paper to decide which of these explanations, if any, are true.53 Instead, this paper seeks to determine whether compliance is systematically related to observable state and case-level characteristics. If no variables consistently correlate with compliance, this will support the theory that the I.C.J. is largely dependent on state whim, making its functions symbolic. If specific features such as remedial clarity or economic interdependence reliably predict whether a state complies, that would suggest that the I.C.J.’s authority functions symbolically or states will view the rulings from an opportunistic angle. The null hypothesis is that no such variables will significantly predict compliance outcomes, suggesting that judgments are treated as non-binding or subject to ad hoc political calculations. Another hypothesis is that states will comply with I.C.J. judgments when doing so aligns with their strategic interests and institutional commitments. Under this hypothesis, states are more likely to comply when they have voluntarily accepted the court’s jurisdiction, particularly under special agreements.

These factors will be tested, but many other factors beyond this study may also align with the strategic interests of the states.

Population size. I predict that larger states may be less likely to comply with I.C.J. decisions because their relative power may reduce external enforcement pressure. Conversely, smaller states may comply more readily as they may feel greater pressure to maintain a favorable standing in international institutions by avoiding reputational damage.

Electoral democracy index. It has been speculated in the past that democracies may be more likely to comply due to institutional norms of the rule of law and greater public scrutiny in populist systems. After all, democratic governments may face greater electoral costs for openly defying international judgments than other forms of governments such as autocracies.54 However, as shown in a 2004 study by academics Tom Ginsburg and Richard McAdams, democracy had no measurable effect on I.C.J. compliance.55

Differences in wealth between countries. I predict that extremely wealthy states might resist compliance if they perceive I.C.J. judgments as infringing on their sovereign discretion or geopolitical autonomy. Alternatively, however, it may be found that wealthier states are more capable of absorbing the material costs of compliance. They may also be more concerned with maintaining international credibility.

Presence of a representative judge on the panel. I predict that having a judge [of shared national origin/nationality] on the panel will correlate with increased compliance, as states may feel that they were given a fairer chance of adjudication. States may appreciate having someone on the court who better understood their perspective, even if the result is ultimately not the one that they were hoping for.

Military strength. I predict that countries with higher military spending will skew toward being more non-compliant. States with a strong military may feel less constrained by international legal obligations, believing that they can withstand the diplomatic consequences of noncompliance.

Jurisdictional challenge. I predict that if a state has issued a jurisdictional challenge, there would be less compliance. A state that challenges the court’s jurisdiction may be revealing a deeper resistance to the court’s authority or a firmer belief that the court is overstepping its jurisdictional bounds. These tendencies could correlate with a lower likelihood of compliance.56

Jurisdiction type. I predict that special agreements will have a higher rate of compliance because in these instances, states have decided to submit themselves to I.C.J. jurisdiction for that specific issue, signaling that they sought out the court’s judgment. I predict lower compliance for compulsory and treaty-based jurisdiction as these forms are often more generic and less situation-based. These more general agreements may leave states feeling as though they submitted to far more jurisdiction than they originally intended.

IV. Data

A. The Approach

Despite the I.C.J. having rendered decisions in over 170 cases, only 51 of these were contentious cases in which at least one state is called upon to take or refrain from specific action under the Court’s judgment. In cases where the I.C.J. asked both states to comply, such as in territorial boundary disputes like Maritime Delimitation in the Caribbean Sea and the Pacific Ocean (Costa Rica v. Nicaragua),57 I examined both states resulting in two data points. This creates the risk of non-independence of observations, as the outcomes for both states arise from a single case and may be influenced by shared facts, geopolitical context, or reciprocal compliance behavior. However, I retained both observations because the analysis seeks to capture state-specific features that condition compliance, and the inclusion of both parties provides a richer, more granular dataset. This resulted in 62 data points.

Because of the high-profile nature of the I.C.J. cases and the international attention they receive, there is a substantial body of reporting and commentary on whether states comply with the Court’s rulings.58 Scholarly articles, news coverage, and official government communications often provide information on the aftermath of each judgment, which made it possible to assess general patterns of compliance. Again, it is worth noting that many of these cases have a range of noncompliance. Some cases involve outright defiance, while others reflect delayed, incomplete, or ambiguous forms of implementation. In an ideal study, we would have been able to code this range. However, due to the low number of datapoints, we ultimately chose to construct the dependent variable as a binary variable.

Let Logit(Compliance) = 0 for full compliance and Logit(Compliance) = 1 for partial compliance or noncompliance.59

Logit(C) = θj - (β1 ⋅ [population_size] +β2 ⋅ [democracy_index_score] + β3 ⋅ [Log_GDP_billions] + β4 ⋅ [judge_from_country] + β5 ⋅ [military_strength] + β6 ⋅ [juris_challenge] + β7 ⋅ [juris_type])

When determining what cases are noncompliant, sources such as Paulson’s 2004 article on I.C.J. compliance provided valuable guidance.60 However, it did not cover more recent cases. A variety of sources were utilized to determine whether states had been compliant to I.C.J. rulings, including journal articles, UN records, and media coverage, to assign each decision to a compliance category as consistently as possible. After the data was gathered, a regression model was run with all the independent variables to determine the statistical significance of each. It must be noted that statistical significance does not necessarily indicate that a variable is substantively important in predicting the outcome. To address this, I analyzed each independent variable individually to assess its predictive strength. After identifying the variables that demonstrated both statistical and substantive importance, I isolated them and ran a second regression model using only that refined set of predictors.61

Log_population_size is the first variable and it measures the respondent’s state logarithmic population at the time of the I.C.J. decision in millions.

Democracy_index was measured using a continuous index ranging from 0 (authoritarian) to 1 (full electoral democracy). This was sourced from Our World in Data, the data portal produced by the University of Oxford.62

Log_GDP_billions represents the difference between the economic capacities of the states in conflict. The dollar amount was inflation-adjusted to account for comparability across time and jurisdiction.63 Because of the huge variation in amounts, I log-transformed the data before finding the difference.64 The final variable represents the difference between the logged GDP of the state being asked to comply and the logged GDP of its opposing party.

Judge_from_country is a binary variable that is equal to 1 if a sitting or ad hoc judge on the panel is from the state being asked to comply; 0, otherwise. Ad hoc judges are judges who are chosen by a state party to a case if they do not have a judge of their nationality already on the Bench.65

Military_strength captures the state’s military expenditures in billion USD as a proxy for its coercive capacity. Though military spending is not a perfect parallel for military strength and capacity, it aligned well enough for this study and was an easy way to quantify the value.

Juris_challenge is a binary variable equal to 1 if the state formally contested the Court’s jurisdiction; otherwise, 0..

Juris_type codes the legal basis for I.C.J. jurisdiction in the case: whether it arose from the optional clause of Article 36(2) giving compulsory jurisdiction, special agreement, or a treaty clause. 0 is compulsory jurisdiction, 1 is jurisdiction by treaty, and 2 is jurisdiction by special agreement. I ordered them from the least tailored to the most tailored. As compulsory jurisdiction is often accepted generically, it reflects the least tailored and the least situational consent. Treaty-based jurisdiction is tied to a specific treaty regime, showing a more issue-focused willingness to adjudicate. And finally, special agreements are case-specific, fully voluntary submissions to the court, and show the strongest form of consent.

While the model includes controls for jurisdictional structure and the presence of a national judge, it does not incorporate case-level fixed effects or variables for the presence of interveners or multiple parties. These omitted controls would ideally be included as they can capture structure or political complexity that may also influence compliance behavior. However, given the relatively small number of I.C.J. cases that resulted in a ruling and limited sample size, I elected to prioritize key institutional and geopolitical variables that could be measured consistently across cases.

Before turning to the data, we will discuss potential areas for selection effects. One potential effect arises from the late-stage nature of the cases that are brought before the I.C.J. The court is often viewed by states as a forum of last resort, reserved only for disputes involving significant international implications such as military incursions or treaty violations with high-profile states. These types of cases may be more inherently prone to noncompliance because the costs are high enough that states that are significantly more powerful might feel less legal obligation to listen to the court. This could skew the dataset toward more difficult cases and make compliance appear rarer than it would be across all international disputes. However, this will not undermine the analysis in a significant way. Even if the circumstances result in the I.C.J. disproportionately hearing high-stakes disputes, this analysis includes circumstances where states did not have control over whether they were brought before the court. A notable number of cases arise from compulsory jurisdiction declarations or are initiated by other states. If anything, the possibility that the seriousness of the underlying dispute may impact compliance from the states reinforces the value of investigating which conditions or factors correlate with successful enforcement. If it is found, for example, that the difference in GDP value between countries has predictive value, then these concerns would be worth revisiting.

A second concern is the chance of selection bias. States that are more compliant or norm-following may be overrepresented in the I.C.J.’s caseload, especially as the court’s jurisdiction requires consent. For example, China, Russia, Brazil, and South Korea are notably some states that have never submitted themselves to the jurisdiction of the court.66 This raises the risk of circularity, where the results of this model will not identify the causes of compliance, but instead the descriptive factors of states that are already predisposed to litigate in international forums and submit themselves to the jurisdiction of the I.C.J. Moreover, states that are frequent litigants may learn from past interactions with the court which could make them better litigants to attain terms that they are more willing to comply with. Being repeat players may also influence states to comply more often than states who rarely appear before the court. However, these concerns are mitigated by two factors. First, many I.C.J. cases are initiated by a different state than the one ultimately asked to comply, and consent to jurisdiction may occur years before the dispute arises. Second, the purpose of my analysis is not to assume that all states are equally likely to comply, but to ask what factors help explain variations in compliance given that a state is given an adverse judgment by the court. Even if certain states are more likely to enter litigation, the variation in compliance across similarly situated respondents allows meaningful analysis.

B. Description of the Data

Averages by Compliance Level

Compliance level	Log_Population_size	Democracy Index	Log GDP Difference	Judge from country	Military spending	Jurisdictional challenge	Jurisdiction type
Full Compliance	2.40	0.545	10.546	0.894	10.780	0.447	1.191
Non-Compliance	3.25	0.560	10.951	0.600	60.061	0.733	0.533

Figure 1: Average values for each independent variable by compliance level.

 

Before turning to regression models, the descriptive averages already suggest patterns across compliance levels. States that fully complied with I.C.J. rulings had substantially smaller populations and lower military spending compared to noncompliant states, indicating that smaller or less powerful states may face stronger incentives to comply. For instance, the average logarithmic population size for full compliance was 2.40 (11.02 million people), while for non-compliance it was 3.25 (25.79 million people), and average military spending was USD 10.780 billion for full compliance versus USD 60.061 billion for non-compliance

Conversely, there seems to be no difference in democracy index, indicating that a state's level of democracy does not significantly predict or correlate with its compliance behavior. Economic asymmetry, measured by the log GDP difference, increased sharply among noncompliant states, suggesting that economically dominant states may feel less constrained by legal rulings.

The presence of a judge from the state being asked to comply correlated positively with compliance, potentially reflecting greater perceived legitimacy or procedural fairness. Furthermore, jurisdictional challenges were more frequent among noncompliant states, reinforcing the idea that early resistance to the court’s authority predicts later defiance. Finally, states that submitted to the court under more specific or consensual jurisdictional bases, such as special agreements, were more likely to comply than those under compulsory jurisdiction. These initial findings offer strong support for the idea that both legal process features and structural power disparities shape compliance behavior.

C. Limitations and Methodological Considerations

While the empirical approach employed in this study offers insights into patterns of state compliance with I.C.J. judgments, there are several limitations in the data that warrant critical examination. First, the dataset is inherently constrained by size. With only 51 contentious cases forming the base at the time of writing, and a handful of those involving dual compliance obligations, the sample is not large enough to allow for extensive disaggregation or more complex modeling techniques.

Second, although considerable effort was made to ensure consistency in coding compliance outcomes, the binary classification of compliance versus noncompliance obscures important gradations. Cases such as Territorial Dispute (Libya/Chad) or Avena were not simple to classify as either clear compliance or noncompliance due to ambiguous timelines and partial implementation. These ambiguities raised questions about how to classify cases with delays or partial fulfillment. Initially, an ordinal model was used where 0 = full compliance, 1 = partial compliance, and 2 = noncompliance. However, the model suffered from overfitting.67 This was avoided by using a binary model of 0 = full compliance, 1 = partial or full compliance. However, by collapsing compliance into a binary outcome, the analysis trades off interpretive nuance for statistical tractability. While this choice was justified due to the small sample size and to avoid class imbalance, future work would benefit from a more fine-grained model.

Finally, there is the potential for confounding variables that were omitted or could not be reliably coded. Some examples of confounding variables include domestic legal capacity, regime stability, international reputation concerns, or the involvement of regional organizations influencing the compliance behavior of states. The presence of ongoing diplomatic negotiations between the states, enforcement pressure from third-party states, or third parties such as Security Council involvement may also affect outcomes independently of the I.C.J.'s procedural structure. These omitted variables may correlate with both jurisdictional posture and compliance likelihood, introducing endogeneity into the model.

Despite these limitations, the analysis provides a meaningful starting point for understanding compliance not as a stochastic or purely power-driven phenomenon, but one shaped by institutional and procedural context. Future empirical work should aim to expand the dataset, refine outcome measures, and explore interactive effects—for example, whether jurisdictional challenges matter more when the state has no judge on the bench, or whether compliance patterns vary over time as the court’s legitimacy evolves.

D. Results

The previous section went over the raw data; however, raw data is merely suggestive and of limited value. While they suggested potential patterns, regression analysis will be more useful in assessing which variables remain predictive while controlling for others. Here we will report the results of several logistic regression models. The first will be a binary regression model that goes over every single independent variable initially tested. Following that, I will run a model-building report of the cumulative pseudo R2, incremental predictive impact of each variable, and whether any of the variables are collinear. That analysis will help me determine if any factors are unimportant to my regression model. If there are, I will then exclude them and create another regression model with only the important variables.

I.C.J. Binary Compliance Model

( 0 = full compliance, 1 = partial or no compliance)

Figure 2: Results of the regression model of I.C.J. compliance.68

The variable juris_type has a coefficient of -1.207 (p = 0.008), suggesting that as the value of this variable increases (that is, as we move from compulsory to special agreement to treaty-based jurisdiction), the likelihood of compliance increases. Since 0 corresponds to compliance and 1 to noncompliance, a negative coefficient indicates movement toward compliance. In substantive terms, this result implies that treaty-based and especially special agreement-based jurisdictions are associated with greater compliance compared to compulsory jurisdiction. This finding reinforces the idea that when states consent more directly or explicitly to I.C.J. jurisdiction in each case, as with a special agreement or by treaty, they are more likely to view the court's ruling as legitimate and binding and are therefore more likely to comply. In contrast, compulsory jurisdiction may be viewed as more adversarial or externally imposed, increasing resistance to compliance.

The variable of judge_from_country has a negative coefficient (-1.723) that is also statistically significant (p = 0.015), suggesting that when a judge on the I.C.J. is a national of the respondent state, that state is more likely to comply with the judgment. In the context of a binary dependent variable where 0 is compliance, this negative coefficient indicates a movement toward compliance. This finding may reflect the legitimacy-enhancing effect of representation; when states perceive that their perspectives are included through a national judge (either as a permanent or ad hoc member), they may be more inclined to view the ruling as fair or procedurally just. Alternatively, it may reflect behind-the-scenes signaling or internal pressure exerted by the judge to ensure compliance and protect the state’s international reputation.

The coefficient for juris_challenge (1.225) is positive and almost statistically significant (p = 0.061), indicating that when a state challenges the jurisdiction of the I.C.J., it is more likely to be noncompliant with the court’s judgment. With the dependent variable is coded such that 0 represents full compliance and 1 represents noncompliance, a positive coefficient increases the likelihood of moving toward noncompliance. This aligns with the hypothesis for this value that jurisdictional objections signal a state’s reluctance to accept the legitimacy or authority of the court, making subsequent compliance less likely. Therefore, a jurisdictional challenge may be viewed as more than a mere procedural step, but as an early indicator of a state’s substantive unwillingness to abide by the court’s ruling.

The other factors, such as population_size, military_strength, log_GDP_billons, and democracy_index, were not statistically significant, which means that there is insufficient evidence to conclude that they have a reliable effect on a state’s likelihood of complying with I.C.J. rulings. Though these variables might plausibly influence a state’s legal behavior, their p-values exceed conventional thresholds for significance suggesting that any observed associations could be due to random variation rather than a systematic relationship. In practical terms, this implies that structural indicators of state power or regime type do not consistently predict compliance behavior in this dataset, at least not when controlling for legal and procedural variables such as jurisdictional challenge or judge nationality. However, a low p-value is not enough on its own to determine the importance of an independent variable.69 Pseudo R2 values can be a start, but better than that is looking at how it increases when each variable is added to a model that already contains all the other variables.

E. The Model Building Report for State Compliance

This portion of the paper will do further analysis of the variables to determine which are statistically important.

 

The full equation is currently:

 

Logit(C) = -0.406 - (0.005 ⋅ [log_population_size] + (0.181 ⋅ [democracy_index_score]) + (0.353 ⋅ [log_GDP_billions]) + (-1.723 ⋅ [judge_from_country]) + (0.007 ⋅ [military_strength]) + (1.2250 ⋅ [juris_challenge]) + (-1.207 ⋅ [juris_type])

 

 

Figure 3: Stepwise model building by cumulative pseudo R2.

 

 

 

Figure 4: The graph above shows incremental pseudo R2 with each variable alone. This is the standalone predictive power of each variable.

 

Figure 5: The graph above shows a collinearity check which is how well each independent variable can be explained by other variables.

 

Figure 4 shows the incremental predictive power, which measures the independent strength of each predictor. This reveals that judge_from_country has the highest standalone power at nearly 14% of the variation in compliance when used in isolation. juris_type followed closely behind, reinforcing its central role in the full model. Interestingly, though juris_challenge was not statistically significant, Figure 4 supports it having significant standalone power. Therefore, I decided to add it to the model.

The other variables, such as democracy index, logarithmic population size, logarithmic GDP, and military strength (with military spending as a proxy), displayed minimal individual explanatory value, suggesting they may not provide meaningful insight on their own. The collinearity graph in Figure 5 shows that juris_type, juris_challenge, and judge_from_country have low R2 values, confirming their informational independence and supporting their inclusion in a reduced model.70

 

The new equation is therefore:

 

Logit(C) = θj - (β1 ⋅ [is there a judge representing the country] + β2 ⋅ [jurisdictional challenge raised] + β3 ⋅ [type of jurisdiction].

I.C.J. Compliance Model With Only Statistically Important Variables
(0 = full compliance, 1 = partial or no compliance)

 

Figure 6: Results of the regression model of I.C.J. compliance.

 

In this updated model with the three variables, there are a few things to note. First, the coefficient and p-values of the model have not changed. Therefore the same trends we saw before apply where (1) as we move from compulsory to treaty-based jurisdictions to special agreement, the likelihood of compliance increases; (2) when a state challenges the jurisdiction of the I.C.J., it is more likely to be noncompliant; and (3) when a judge on the I.C.J. is a national of the state being asked to comply, compliance is more likely.

Substantively, the findings from this pared-down model underscore that compliance is not significantly dependent on a state’s structure power such as GDP, population size, or military strength. It instead seems more deeply tied to the legitimacy of the legal process with jurisdictional consent and procedural voice appear to meaningfully shift how a state engages with I.C.J. judgments. This is supported by these features remaining highly predictive even when controlling for each other. Therefore, to states, the perception of fairness and procedural justice may be just as important to international legal compliance as material incentives or enforcement mechanisms. The implications for international law are significant as despite being a field long criticized for its lack of enforcement power, this model illustrates that design choices within the I.C.J.’s procedural architecture can significantly shape state behavior. Going forward, these factors may provide for a potential roadmap for reform and strengthening the Court’s institutional legitimacy.

V. Conclusion

Though further research should be done in the future, the findings of this paper suggest that compliance with the I.C.J. is not purely based on state discretion. Instead, three legal and procedural factors, jurisdictional challenges, the method by which the Court asserts jurisdiction, and whether there is a judge on the bench from the state being asked to comply, emerge as significant predictors of compliance. Cases in which jurisdiction is actively contested are more likely to result in noncompliance, suggesting that resistance to the Court’s authority at the threshold level reverberates through to the enforcement stage. The presence of a judge from the respondent state on the bench is associated with increased compliance indicating that representation on the bench may enhance the perceived legitimacy of the decision and encourage states to comply with the I.C.J. ruling, even if it costs them. Likewise, cases heard under compulsory jurisdiction are associated with lower compliance compared to those initiated by special agreement or treaty. This may be due to the higher baseline of consent in compulsory jurisdiction and due to states being unaware of which cases could and will be brought before the I.C.J. Structural factors such as population size, military strength, and economic dominance do not significantly explain why powerful states may resist compliance, even in the face of adverse rulings. Democracy continues to not show a consistent relationship with legal obedience.

Together, these findings suggest that procedural legitimacy in the eyes of states can temper resistance and enhance perceptions of fairness. This challenges the view that I.C.J. decisions are dictated only by self-interest, strategic power, or external enforcement pressure. If compliance were truly random, or if states simply ignored judgments inconsistent with their geopolitical aims, we would not expect to find any significant predictors in the data. Yet, the regression models show that states behave in systematically different ways depending on how legal authority is constructed and conveyed.

Ultimately, these findings provide reason for cautious optimism. That compliance is empirically predictable at all is itself a meaningful result because it suggests that international adjudication is not mere theater, but a form of governance whose effectiveness can be studied, refined, and improved. Rather than confirming the skepticism that international law is “mere advice,” this study suggests a more complex and textured reality. It seems that legal and procedural legitimacy are key factors capable of mediating the hard edges of state sovereignty, even if they do not manage to fully constrain it. Therefore, though international law may lack the teeth of domestic enforcement, it is not entirely toothless. We suggest that international law focus on these public and observable variables to improve the state compliance to a ruling from the I.C.J.

VI. Appendix One

I.C.J. Ordinal Compliance Model

( 0 = full compliance, 1 = partial compliance, 2 = no compliance)71

 

VII. Appendix Two

I.C.J. Contentious Cases Used in the Model

Case Name	Decision Year	Applicant	Respondent	Country Asked to Comply	Compliance
Corfu Chanel	1949	United Kingdom	Albania	Albania	2
Right of nations of U.S. in Morocco	1952	France	United States	France	0
Minquiers and Ecrehos	1953	France  United Kingdom		France	0
Sovereignty over Certain Frontier Land	1959	Belgium Netherlands		Netherlands	0
Arbitral Award Made by the King of Spain on 23 Dec. 1906	1960	Honduras	Nicaragua	Nicaragua	1
Temple of Preah Vihear	1962	Cambodia	Thailand	Thailand	1
North Sea Continental Shelf	1969	Netherlands West Germany		Germany	0
North Sea Continental Shelf	1969	Denmark West Germany		Germany	0
Appeal Relating to the Jurisdiction of the ICAO Council	1972	India	Pakistan	India	0
Fisheries Jurisdiction	1974	West Germany	Iceland	Iceland	2
Fisheries Jurisdiction	1974	United Kingdom	Iceland	Iceland	2
United States Diplomatic and Consular Staff in Tehran	1981	United States	Iran	Iran	1
Continental Shift	1982	Libya Malta		Libya	0
Delimitation of the Maritime Boundary in the Gulf of Maine Area	1984	Canada United States		Canada &  United States	0; 0
Continental Shelf	1985	Libya Malta		Libya and Malta	0; 0
Frontier Dispute	1986	Mali Upper Volta		Burkina Faso and Mali	0; 0
Military and Paramilitary Activities in and against Nicaragua	1991	Nicaragua	United States	United States	2
Land, Island and Maritime Frontier Dispute	1992	El Salvador   Honduras	Nicaragua	Honduras and El Salvador	0; 0
Maritime Delimitation in the Area between Greenland and Jan Mayen	1993	Denmark	Norway	Norway	0
Territorial Dispute	1994	Chad	Libya	Libya	0
Kasikili/Sedudu Island	1999	Botswana Namibia		Namibia	0
Maritime Delimitation and Territorial Questions between Qatar and Bahrain	2001	Qatar	Bahrain	Qatar and Bahrain	0
LaGrand	2001	Germany	United States	United States	1
Sovereignty over Pulau Ligitan and Pulau Sipadan	2002	Indonesia  Malaysia		Indonesia	0
Land and Maritime Boundary between Cameroon and Nigeria	2002	Cameroon  Nigeria		Nigeria	1
Arrest Warrant of 11 Apr. 2000	2002	Democratic Republic of the Congo	Belgium	Belgium	0
Avena and Other Mexican Nations	2004	Mexico	United States	United States	1
Frontier Dispute	2005	Benin Niger		Benin and Niger	0; 0
Territorial and Maritime Dispute between Nicaragua and Honduras in the Caribbean Sea	2007	Nicaragua	Honduras	Honduras and Nicaragua	0
Application of the Convention on the Prevention and Punishment of the Crime of Genocide	2007	Yugoslavia	Bosnia and Herzegovina	Serbia and Montenegro	0
Certain Questions of Mutual Assistance in Criminal matters	2008	Djibouti	France	France	0
Sovereignty over Pedra Branca/Pulau Batu Puteh, Middle Rocks and South Ledge	2008	Malaysia Singapore		Malaysia and Singapore	0
Dispute regarding Navigational and Related Rights	2009	Costa Rica	Nicaragua	Nicaragua	0
Pulp Mills on the River Uruguay	2010	Argentina	Uruguay	Uruguay	0
Application of the Interim Accord of 13 September 1995	2011	FYROM	Greece	Greece	0
Questions relating to the Obligation to Prosecute or Extradite	2012	Belgium	Senegal	Senegal	0
Territorial and Maritime Dispute	2012	Nicaragua	Colombia	Colombia	2
Ahmadou Sadio Diallo	2012	Guinea	Democratic Republic of the Congo	Democratic Republic of the Congo	0
Jurisdictional Immunities of the State	2012	Germany	Italy	Italy	1
Frontier Dispute	2013	Burkina Faso Niger		Burkina Faso and Niger	0; 0
Request for Interpretation of the Judgment of 15 June 1962 in the Case Concerning the Temple of Preah Vihear	2013	Cambodia	Thailand	Thailand	0
Maritime Dispute	2014	Peru	Chile	Peru and Chile	0; 0
Whaling in the Antarctic	2014	Australia	Japan	Japan	1
Construction of a Road in Costa Rica along the San Juan River	2015	Nicaragua	Costa Rica	Costa Rica	0
Land Boundary in the Northern Part of Isla Portillos	2018	Costa Rica	Nicaragua	Nicaragua	0
Maritime Delimitation in the Caribbean Sea and the Pacific Ocean	2018	Costa Rica	Nicaragua	Costa Rica and Nicaragua	0; 0
Certain Activities Carried Out by Nicaragua in the Border Area	2018	Costa Rica	Nicaragua	Nicaragua	0
Jadhav	2019	India	Pakistan	Pakistan	1
Maritime Delimitation in the Indian Ocean	2021	Somalia	Kenya	Somalia and Kenya	0; 1
Armed Activities on the Territory of the Congo	2022	Democratic Republic of the Congo	Uganda	Uganda	0
Alleged Violations of Sovereign Rights and Maritime Spaces in the Caribbean Sea	2022	Nicaragua	Colombia	Colombia	0

 

 

• U.N. Charter art. 92, June 26, 1945, 59 Stat. 1031, 3 Bevans 1153; see also Robert Y. Jennings, The International Court of Justice after Fifty Years, 89 Am. J. Int’l L. 493, 493 (1995).
• The Court, Int’l Court of Justice, https://perma.cc/F8EY-LH9V (last visited May 2, 2026).
• Allen Weiner et al., International Law 36 (8th ed. 2023) (citing The Case of the S.S. “Lotus” (Fr. v. Turk.), 1927 P.C.I.J. (ser. A) No. 10, at 18).
• Id. 
• Statute of the International Court of Justice art. 36, ¶ 2, June 26, 1945; see also International Court of Justice, Northeastern University School of Law, https://perma.cc/4D2C-5BSQ; see also U.S. Dep’t of Just., Legal Disputes Under Article 36(2) of the Statute (From International Court of Justice at a Crossroad), 183–222 (1987), https://perma.cc/8YQ2-KYKC.  
• See Louis Henkin, How Nations Behave: Law and Foreign Policy, 92–95 (2d ed. 1979) (discussing how skeptics claim that the society of nations “has no effective law-making body or process”).
• Thomas M. Franck, Legitimacy in the International System, 82 Am. J. Int’l L. 705, 707 (1988).
• Henkin, supra note 6, at 95. This was not Henkin’s own perspective of the I.C.J. Rather, this was his description of how skeptics view the I.C.J. and its jurisdictional power.
• Joan E. Donoghue, The Effectiveness of the International Court of Justice, 108 Am. Soc’y Int’l L. 114, 115 (2014). Judge Donoghue stepped down from the I.C.J. in 2024. She had been a judge between 2010–14.
• Id.
• Id. 
• See generally, Tom Ginsburg & Richard H. McAdams, Adjudicating in Anarchy: An Expressive Theory of International Dispute Resolution, 45 Wm. & Mary L. Rev. 1229, 1311–17 (2004); see also generally Eric A. Posner & Miguel F.P. de Figueiredo, Is the International Court of Justice Biased?, Chi. Unbound 1, 8 (2004). One of the goals of this paper was incorporating new potential independent variables.
• See infra, Appendix One for data on the cases.
• U.N. Charter art. 92, supra note 1; see also Statute of the International Court of Justice, supra note 5, art. 36, ¶ 2.
• How the Court Works, I.C.J. , https://perma.cc/8THX-YEAP (last visited Feb. 21, 2026).
• Id.
• Declarations recognizing the jurisdiction of the Court as compulsory, I.C.J., https://perma.cc/72W5-N3MA (last visited Feb. 21, 2026).
• See Michael J. Matheson, The Opinions of the International Court of Justice on the Threat or Use of Nuclear Weapons, 91 Am. J. Int’l L. 417, 418 (1997) (The Court “may give an advisory opinion on any legal question at the request of whatever body may be authorized by or in accordance with the Charter of the United Nations to make such a request.”).
• See Edvard Hambro, The Authority of the Advisory Opinions of the International Court of Justice, 3 Int’l Compar. L.Q. 1, 5 (1954). Both legal scholars and the court itself have taken the position that advisory opinions are to give advice to countries rather than operate as legally binding law.
• List of All Cases, I.C.J., https://perma.cc/5ZHF-NA6D (last visited Feb. 21, 2026).
• See generallyContentious cases, I.C.J., https://perma.cc/R4ZN-3HLN (last visited Feb. 21, 2026).
• See generally Gary L. Scott & Craig L. Carr, The ICJ and Compulsory Jurisdiction: The Case for Closing the Clause, 81 Am. J. Int’l L. 57 (1987) (discussing the refusal of the United States to bind itself to the decision of Case Concerning Military and Paramilitary Activities in and against Nicaragua); see also Joseph Sinde Warioiba, Monitoring Compliance with and Enforcement of Binding Decisions of International Courts, 5 Max Plank Y.B. 41, 45 (2001) (“France withdrew from the jurisdiction of the ICJ after the Nuclear Test Cases.”).
• Colter Paulson, Compliance with Final Judgment of the International Court of Justice since 1987, 98 Am. J. Int’l L. 434, 435 (2004) (citing Jonathan I. Charney, Disputes Implicating the Institutional Credibility of the Court: Problems of Non-Appearance, Non-Participation, and Non-Performance, in The International Court of Justice at a Crossroads 288 (1988)).
• See Edgardo Sobenes Obregon, Non-Compliance of Judgments and the Inherent Jurisdiction of the ICJ, 7 J. Territorial & Mar. Stud. 53, 56 (2020).
• See generally id. 
• Corfu Channel (U.K. v. Alb.), Judgment, 1949 I.C.J. 4 (Apr. 9).
• Id. at 10–11.
• Corfu Channel Overview of the Case, I.C.J, https://perma.cc/SRY2-C5UM. The amount in reparations deemed owed to the U.K. by Albania was £844,000.
• See Posner & de Figueiredo, supra note 12, at 8.
• Military and Paramilitary Activities in and Against Nicaragua (Nicar. v. U.S.), Judgment, 1986 I.C.J. 14 (June 27).
• Id. ¶¶ 18–25.
• Id. ¶ 292.
• Id. ¶ 33. The U.S. had written a letter to the Court informing it that “the United States is constrained to conclude that the judgement of the Court was clearly and manifestly erroneous as to both fact and law. The United States remains firmly of the view, for the reasons given in its written and oral pleadings that the Court is without jurisdiction to entertain the dispute.”
• Charney, supra note 23, at 288.
• Territorial Dispute (Libya v. Chad), Judgment, 1994 I.C.J. 6 (Feb. 3).
• Gino J. Naldi, The Aouzou Strip Dispute—A Legal Analysis, 33 J. Afr. L. 72, 72 (1989).
• Libya v. Chad, supra note 35 ¶ 17.
• Id. ¶ 77.
• See Paulson, supra note 23, at 440–41.
• Id. (citing 1 Libya Reinforcing Aouzou Strip, Radio France Internationale, Feb. 13, 1994; Libya Rules out Withdrawal, Moroccan Radio, Feb. 19, 1994).
• Avena and Other Mexican Nationals (Mex. v. U.S.), Judgment, 2004 I.C.J. 12 (Mar. 31).
• Id. ¶¶ 1–12.
• Id. ¶ 153(11).
• See Paulson, supra note 23, at 445–46.
• See generally Daniel Behn et al., Poor States or Poor Governance? Explaining Outcomes in Investment Treaty Arbitration, 38 Nw. J. Int’l L. & Bus. 333, 343 (2018) (explaining how asymmetries of power and wealth can lead to unbalanced international investment treaty arbitration).
• B.S. Chimni, The International Court and the Maintenance of Peace and Security: The Nicaragua Decision and the United States Response, 35 Int’l Compar. L.Q. 960, 966 (1986).
• See Territorial Dispute (Libya v. Chad), supra note 35.
• Chimni, supra note 46, at 964.
• Id. 
• Id. 
• See Charney, supra note 23.
• See id. 
• The dataset of contentious I.C.J. cases is insufficiently voluminous, especially considering there are so many independent factors in political decisions to test.
• See Philip Pettit, Democracy, National and International, 89 Monist 301, 309 (2006) (speculating that in democratic governments, there are “formal channels . . . whereby people can react to government by appeal to the courts, or to various tribunals covering administrative matters, or to commissions governing issues like human rights”).
• See Ginsburg & McAdams, supra note 12, at 1311–14 (noting the evidence did not support that liberal democracies are more willing to comply). However, I am still interested in testing the impact of democratic value in case there has been any change since the study was done.
• See Charney, supra note 23.
• Maritime Delimitation in the Caribbean Sea and the Pacific Ocean (Costa Rica v. Nicar.), Judgment, 2018 I.C.J. 139 (Feb. 2). In this case, the I.C.J. adjusted the line near the mouth of the San Juan River which required both states to comply, which they did.
• See generally Paulson, supra note 23; Weiner et al., supra note 3, at 36–65 (providing various commentaries on the binding nature of and compliance with international law).
• See Emily C. Zabor et al., Logistic Regression in Clinical Studies, 112 Stat. for People 271, 273 (2022) (noting a logistic regression model is used when the outcome of interest is binary or ordinal).
• See generally Paulson, supra note 23.
• See generally Jim Frost, Identifying the Most Important Independent Variables in Regression Models, Statistics by Jim, https://perma.cc/6H4A-77BN.
• Our World in Data, Harv. U., https://perma.cc/5W9U-W63M.
• Because raw GDP figures can be extremely large and introduce scale imbalance into regression models, I applied a logarithmic transformation to compress the range of values.
• Because raw GDP figures can be extremely large and introduce scale imbalance into regression models, I applied a logarithmic transformation to compress the range of values. – to avoid confusion stemming from identical footnote, maybe add a clarifying word in this footnote or above that you ran two different log transformation models: one for inflation, one for international difference. Unless there’s only one, then two footnotes might be redundant.
• Statute of the International Court of Justice, supra note 5, art. 31, ¶¶ 2–3.
• See Posner & de Figueiredo, supra note 12, at 614 (“Two-thirds of states have never appeared before the ICJ, including China, Japan, Russia/USSR, South Korea, Brazil, Argentina, and Poland.”); see also Ahmed Sheikh, The Peoples’ Republic of China and International Law, 7 Int’l L. 770, 779 (1973). China will at times go far enough to publicly ridicule or criticize the writings of international law scholars on the rulings from the I.C.J.
• See infra Appendix 1.
• Appendix 1 shows the results of this analysis when I used an ordinal scale for the dependent variable. Where 0 = compliance, 1 = partial compliance, and 2 = non-compliance. The results were ultimately the same, further supporting the accuracy of this dataset.
• See also Ken Kelly & Scott E. Maxwell, Sample Size for Multiple Regression: Obtaining Regression Coefficients That are Accurate, Not Simply Significant, 8 Psych. Methods 305, 309–11 (2003) (generally discussing the importance of a large sample size and supporting that I will have to look beyond statistical significance to boost the overall fit of my model); see generally Zabor, supra note 59, at 275 (discussing how regression models should not include a large number of independent variables). It is important to figure out the important predictors and limit the number of variables. This was especially important for my research due the small number of cases and datapoints that I was working with.
• It is worth noting that Figure 3 shows that Log_GDP_billons has notable predictive power with the model jumping to around 23% after it is added. However, I ultimately decided against adding it in my final model. This decision was made due to Log_GDP_billons’s extremely high collinearity in Figure 5 and the minimal individual explanatory power shown in Figure 4. These factors suggested that it was not an important variable.
• As we can see from the data input with an ordinal dependent variable, the same two values are statistically significant (though judge_from_country and juris_type have switched positioning ranking from lowest P-value), and juris_challenge is still almost statistically significant. Despite similar results, when I ran the ordinal regression model with only the important variables (judge_from_country, juris_type, and juris_challenge), the “% of observations of compliance correctly predicted” came out as 100%. This indicated that the regression model was overfitted. This was likely due to the low number of cases, which represent full noncompliance (values 2). The ordinal model was used to avoid splitting the limited number of partially noncompliant and noncompliant cases.