Prosecution of complex criminal networks: a multilevel ERGMs approach to CICIG's judicial cases

1. Problem Statement

The paper addresses a critical gap in criminology and legal studies: the lack of empirical understanding regarding how prosecutors strategically select charges and structure imputation strategies to dismantle complex criminal networks. While traditional deterrence theory focuses on individual offenders (certainty, severity, and swiftness of punishment), it often fails to account for the systemic dynamics of organized crime, where networks are sustained by functional ties between public and private actors.

The authors argue that simplistic prosecution strategies (e.g., targeting isolated incidents or single offenders) often lead to counterproductive outcomes, such as over-criminalization or the rapid reorganization of networks. Using the International Commission Against Impunity in Guatemala (CICIG) and the Special Prosecutor's Office Against Impunity (FECI) as a case study, the paper seeks to determine whether prosecutorial decisions were driven by a strategic, network-disrupting logic or merely by case-by-case legal necessities. Specifically, it investigates how prosecutors navigated the intersection of criminal networks, legal frameworks, and imputation ties to achieve systemic deterrence.

2. Methodology

The study employs a Multilevel Exponential Random Graph Model (ERGM), a sophisticated inferential network analysis technique adapted from socio-ecological systems research. This approach allows for the simultaneous modeling of relationships within and between different network layers.

Data Structure

The dataset is conceptualized as a multilevel network comprising three interconnected layers:

1. Criminal Network (Level A): A one-mode network of 189 individual actors (87 public, 102 private) connected by 365 ties representing illicit interactions (communications, bribes, contracts).
2. Legal Framework Network (Level B): A one-mode network of 21 distinct criminal offenses linked by the 6 laws governing them. Offenses are categorized by type (corruption vs. non-corruption) and severity (high vs. low).
3. Prosecution Network (Level X): A bipartite (two-mode) network connecting the 189 actors to the 21 offenses via 250 imputation ties (charges filed).

Analytical Framework

The authors utilize Multilevel ERGMs to test four specific hypotheses regarding prosecutorial strategies, modeled through "building blocks" (network motifs):

• H1 (Co-offenders): Do prosecutors jointly charge connected co-offenders with the same or different offenses?
• H2 (Multi-offenders): Do prosecutors charge individuals involved in multiple crimes under a single law or across multiple statutes?
• H3 (Severity): Is there a strategic preference for imputing high-severity offenses (statutory penalty > 6-8 years)?
• H4 (Offense Type): Do prosecutors use "hybrid" strategies, combining corruption and non-corruption charges for the same actors?

The models control for endogenous network effects (e.g., tie density, triadic closure) and actor attributes (public vs. private sector).

3. Key Contributions

• Theoretical Integration: The paper bridges deterrence theory and network science, proposing that prosecutorial effectiveness should be measured by the capacity to disrupt the relational ties and organizational dynamics of criminal networks, rather than just individual convictions.
• Methodological Innovation: It is the first application of Multilevel ERGMs to real-world prosecutorial data. This moves beyond linear regression (HLM) to model the conditional dependence of ties across criminal, legal, and prosecution layers.
• Strategic Typology: The study defines specific "prosecution building blocks" (e.g., Interaction Triangles, Cross-level 2-stars) that serve as metrics for evaluating the complexity and strategic intent of imputation strategies.

4. Key Results

The ERGM analysis yielded mixed but significant findings regarding the strategic nature of CICIG-FECI prosecutions:

• H1 (Co-offenders): Supported (Partially). The model found a significant positive effect for the Interaction Triangle configuration. This indicates that prosecutors systematically charged directly connected co-offenders with the same offense. However, more complex configurations (e.g., charging co-offenders with different offenses under different laws) were not statistically significant, suggesting a reliance on simpler, shared-evidence strategies rather than highly differentiated imputation.
• H2 (Multi-offenders): Strongly Supported. The Cross-level 2-star term was positive and significant. This confirms that multi-offenders were frequently charged with offenses falling under different legal statutes rather than a single law. This reflects a deliberate strategy to expand legal exposure and capture the multifaceted nature of illicit political-economic networks.
• H3 (Severity): Supported (Partially). There was a significant positive tendency to impute high-severity offenses overall. However, the model did not find evidence that prosecutors systematically combined multiple high-severity charges for the same individual (the "Similar severity 2-star" was not significant). Severity was used as a general lever, but not necessarily stacked in a specific multi-offense pattern.
• H4 (Offense Type): Not Supported. Contrary to the hypothesis, there was no statistically significant evidence that prosecutors systematically combined corruption and non-corruption offenses for multi-offenders. While corruption charges dominated, the "hybrid" strategy of mixing offense types was not a consistent, systematic pattern in the data.
• Control Variables:
  • Actor Popularity: Central actors in the criminal network were significantly more likely to be charged with more offenses.
  • Public Sector Bias: Prosecution ties were significantly more likely to target public actors (state officials) than private actors, aligning with the mandate to dismantle state-captured networks.

5. Significance and Implications

• Systemic Deterrence: The study demonstrates that effective prosecution in high-corruption environments requires moving beyond a "case-by-case" model to a multi-case, multi-offender framework. The successful use of cross-statute charging (H2) highlights a strategy that leverages the legal system's full breadth to disrupt network resilience.
• Limitations of Current Practice: The lack of systematic hybrid charging (H4) and complex co-offender differentiation (H1) suggests that while prosecutors targeted central nodes and used severe penalties, they did not fully exploit the potential of "robust imputation structures" that could theoretically prevent network reorganization.
• Policy Recommendations:
  • Prosecutorial strategies must evolve to integrate intelligence and data-driven analysis to design strategic charge combinations (mixing offense types and statutes) that specifically target the functional ties sustaining criminal networks.
  • Future interventions should aim to disrupt both public and private brokers simultaneously, as focusing solely on state officials may leave the private sector infrastructure intact.
• Broader Impact: The paper provides a benchmark for evaluating prosecutorial effectiveness not just by conviction rates, but by the structural disruption of illicit networks. It offers a novel socio-legal perspective for analyzing how legal decisions reverberate through complex political-economic ecosystems.

In conclusion, the paper validates that while CICIG-FECI employed a sophisticated, network-aware approach (particularly in targeting central public actors and using multiple statutes), there remains room for more complex, systematic imputation strategies to fully dismantle the regenerative capacity of organized crime networks.