- More than 100 agencies publish homicide statistics across Latin America and the Caribbean, and in some countries up to seven institutions report different numbers for the same phenomenon. In practice, distorted measurement leads to distorted policies.
- Even homicide counts, the indicator researchers and policymakers trust most, face problems, including disagreement among domestic agencies, inconsistent definitions across countries, and incidents that never enter any official record.
- The IDB is working on new initiatives under the Alliance for Security, Justice and Development, such as the Regional Observatory on Transnational Organized Crime and the Secure Electronic Exchange of Criminal Records, aimed at increasing institutional capacity and coordination between governments' institutions.
Crime statistics lie at the foundation of citizen security policy. Governments use them to decide where police are deployed, which neighborhoods receive crime prevention programs, or whether security strategies are working. Yet these numbers are not direct observations of criminal activity; they are institutional products, shaped by reporting behavior, administrative procedures, legal classifications, and data quality practices. When the measure is distorted, the policy built on it is distorted too, and resources flow away from the communities that need them most.
At the Inter-American Development Bank (IDB), understanding the true scale and cost of violence has long been central to our citizen security work. Our study The Costs of Crime and Violence documented not only how expensive crime is for the region, but how difficult it is to measure. Further, through initiatives like the Regional Observatory on Transnational Organized Crime and the Secure Electronic Exchange of Criminal Records System (ICARO), designed within the Alliance for Security, Justice, and Development, we are working on increasing coordination and collaboration between government agencies to enhance consistency in reporting and reduce these gaps.
In this post, we trace the problem from its origins in how crime is recorded, through its consequences for policy, to what better measurement looks like in practice, using Chile as a case study.
Unlike economic transactions or health outcomes, criminal activity does not generate administrative records as a byproduct of its occurrence. Because it happens outside formal institutions, crime leaves no statistical trace unless a victim or the state initiates contact with formal channels. In other words, every crime statistic can only undercount the true number of crimes, because crimes that no one reports simply do not exist in the data.
These statistics mostly come from administrative records, such as police files, prosecution records or court proceedings, which capture only events that survived a series of institutional filters:
- Reporting. In Latin America and the Caribbean, approximately nine out of ten non-lethal crimes are never reported to authorities.
- Registration. Even reported offenses may never enter official systems if officials fail to open files or decline to register complaints.
- Classification. Recording an incident involves discretionary decisions about which offense category it falls into.
- Coverage. Populations with limited contact with formal institutions, such as residents of remote areas, informal settlements, and migrant communities, are systematically underrepresented.
As an alternative, victimization surveys were designed to close part of this gap by asking people directly about their experiences, regardless of whether an incident was ever reported. Surveys can capture information that administrative data may not, for instance perceived insecurity, reasons for not reporting, or how people change their behavior in response to crime. But surveys carry their own limitations. For example, people misremember dates or are reluctant to disclose sensitive incidents. Surveys are also expensive to conduct, especially in areas most affected by crime, so they are not often updated. They complement administrative data, but they do not replace it.
Between the two, we are left with two imperfect and incomplete pictures of crime, administrative records that miss incidents that are never reported, and surveys that miss whatever people will not disclose or cannot be asked about often enough. Neither gives a full count on its own. That's why researchers and policymakers gravitate toward the one crime that is hardest to hide entirely: homicide.
Because homicide leaves a physical record, a body, researchers and policymakers have converged on the homicide rate as the standard indicator of violence. The United Nations Office on Drugs and Crime (UNODC) and the World Bank use it as their primary cross-national measure, as does academic research. Yet homicide statistics face three distinct problems:
- Agencies within the same country disagree. Most countries maintain both a criminal justice count (police or prosecutors) and an epidemiological count (ministries of health using the World Health Organization's International Classification of Diseases, ICD-10). Often there is more; women's ministries track femicides, and child-protection services track deaths of minors, each under its own definitions. The gap between these counts can be significant and multidirectional.
- Countries define homicide differently. Even when institutions agree on a number, they may not agree on what counts as a homicide. Some countries report femicide as a separate legal category, not included in its homicide total. Another source of variation comes from deaths caused during police interventions, where some countries count them towards homicides and others do not. Thus, a cross-country analysis of homicide rates is never a clean comparison of violence alone; it is also, unavoidably, a comparison of each country's institutional choices about what gets counted as a homicide in the first place.
- Some homicides are never recorded in any system. When a person is killed and the body is concealed, no death certificate is issued and no homicide is counted anywhere — not by police, not by health authorities. No one knows exactly how common this is, but these hidden killings appear to be concentrated in areas with a strong presence of organized crime.
This is not a hypothetical risk. If a government allocates security resources based on municipal homicide rates, and one municipality's homicides are systematically undercounted while another is measured accurately, the government will underinvest precisely where violence is highest. This is not random. Weaker institutions tend to produce weaker data, so the places with the least State presence often have the least reliable numbers. Citizens get penalized twice, first by weak institutions, then by the funding decisions based on those same institutions' flawed data.
The problem extends to policy evaluation as well. A drop in reported homicides might reflect a genuine improvement in public safety, or a change in how an office classifies deaths, or which of several competing counts got used that year. From the outside, the two are indistinguishable. Without reliable measurement, governments cannot know whether their crime-prevention policies are working.
No country has solved crime mismeasurement entirely, but Chile's experience shows how institutional coordination can help. In 2022, the country created the Centro para la Prevención de Homicidios y Delitos Violentos (Center for the Prevention of Homicides and Violent Crimes, CPHDV) at the Undersecretariat of Crime Prevention. Before the center existed, institutions published homicide data independently, with up to five different measures for the same period.
The center's model rests on three functions:
- Cross-validation of the homicide registers across eleven state agencies on a shared platform with victim-level traceability, records are matched using age, date, as well as the place of aggression and death, and counted under the UN's International Classification of Crime for Statistical Purposes (ICCS) definition of homicide. The unit of measurement is the individual victim rather than the police case;
- Standardization of information on victims and violent incidents; and
- Production of analytical outputs such as concentration mapping and local trend assessments, published in annual homicide reports, including a retroactive adjustment of official data back to 2018 under the new methodology.
Chile's experience in combining data from CPHDV, the Prosecutor's Office, and the Policía de Investigaciones de Chile (Chilean Investigations Police, PDI) shows that much of the inconsistency in crime statistics is not inherent to crime itself; it comes from how institutions independently produce and reconcile their own administrative data.
This work builds on broader IDB engagement with Chile's citizen security agenda. The IDB country strategy provides a diagnosis that includes rising victimization, a widening gap between perceived and objective crime rates, and low reporting rates. It is paired with an uptick in cybercrime, with a portfolio of instruments proportionate to that challenge, including the Program to Strengthen the Strategic Management of Public Security (CH-L1142) and technical cooperation on police modernization, cybercrime incident management, and social reintegration for vulnerable populations. Better crime measurement is one piece of that larger effort to restore institutional legitimacy and strengthen the management capacity of the agencies responsible for public safety.
Crime statistics are far less objective than often assumed. Across our region, agencies count crimes under different classifications, yet their data shape police deployments, security budgets, and public perceptions of success or failure. This is fixable. Clearer definitions, integrated registries, and transparency about what is and what is not being measured can substantially improve the quality of security policymaking.
The IDB works together with countries using its lending instruments and producing knowledge products to harmonize violence measurement. Explore the IDB's evidence on the costs of crime and violence.