Why Bangladesh's murder statistics hide more than they reveal

Last August, four members of the same family were murdered in Rangpur. In the police's Crime Statistics for August 2026, their deaths will appear as one murder case, indistinguishable from a single killing. 

Nothing in the data will reveal that four people died. Because the police count the murder statistics by the number of recorded murder cases, not the number of victims. 

And this means some of the most basic questions about crime in Bangladesh cannot be answered: Where is violence concentrated? When does it peak? Does tougher punishment deter it? 

And therefore, researchers, policymakers and the police themselves are left setting their agenda from incomplete data.

Counting cases, not crimes

The previous Awami League regime stopped publishing monthly crime statistics after the 2018 national election. It was not until after the July Uprising that the interim government resumed publication and released data from January 2019 onwards. 

The figures show how many cases were filed each month under 10 categories, broken down by police unit. Cases may be filed long after the murder. 

Many crimes never reach the police at all. In Bangladesh, suspicious deaths are first registered as unnatural deaths and may later be converted into murder cases, sometimes in a different year and sometimes never. And a case filed today for a killing years ago is counted as a crime of the current month.

According to an additional superintendent of Bangladesh Police, they are still employing this method. 

Nikhil Chandra Nath, a doctoral researcher and adjunct faculty member in Criminal Justice and Criminology at the University of New Haven, said, "For law enforcement, the unit of analysis is the case file created for investigative purposes. Relying on investigative case files to measure actual crime yields several major limitations." 

The world counts victims

International practice settled the counting question long ago. 

The UN's International Classification of Crime for Statistical Purposes, endorsed in 2015, makes the victim the counting unit for acts leading to death. 

England and Wales follow a "one crime per victim" rule and maintain a Homicide Index in which every victim is a separate record, with age, sex, method of killing and relationship to the suspect. Australia, likewise, counts people harmed, not police paperwork.

The timing question has been solved too. 

Statistics Canada records the date each incident occurred and revises earlier figures as late reports arrive, instead of adding old crimes to the current month. Even then, late recognition still leaves a trace. 

In England and Wales, the 96 Hillsborough victims entered homicide statistics only after fresh inquests ruled them unlawfully killed in 2016, some 27 years after the disaster. Because the system records incidents separately, however, analysts can see exactly why that year's figure jumped.

The US offers the closest parallel to Bangladesh's situation. Its old Summary Reporting System tallied monthly aggregates and applied a "hierarchy rule" that counted only the most serious offence in any incident. 

On 1 January 2021, the FBI moved to the National Incident-Based Reporting System (NIBRS). NIBRS records up to 10 offences per incident, details of every victim, and both the date a crime occurred and the date it was reported.

Two other American tools address flaws Bangladesh shares. The FBI's Supplementary Homicide Report has long recorded victims separately from incidents. The National Violent Death Reporting System cross-references medical examiner files, death certificates and police records, so a death does not vanish when it is reclassified.

India shows the opposite limitation. 

Its National Crime Records Bureau publishes both cases and victims for murder, and most police stations log reports digitally. But the bureau still counts only the gravest offence in a multi-crime incident. It publishes once a year, one to two years late, in static PDFs. Digitisation alone did not make the data useful in this case. 

The research Bangladesh cannot do

The deeper cost is lost knowledge. Sociologists John Kitsuse and Aaron Cicourel warned in 1963 that official statistics describe the institutions that produce them as much as the behaviour they claim to measure. Bangladesh has done little to correct for that.

Nikhil Chandra Nath knows of no study testing whether harsher punishment in Bangladesh actually reduces crime, and no index measuring the accuracy of police data. "Unlike the US, where police departments employ research scientists to work on indexing and measurements," he said, "it is unclear whether any scholar here has attempted such systemic analysis." 

He sees the timing flaw as the greater threat to policy. 

"Counting cases rather than victims is a major problem if the goal is determining the actual death toll," Nikhil said. "But recording crimes by the filing date rather than the incident date creates a critical disruption in criminal justice policymaking." 

The gap can still be measured. Researchers can take a representative sample of case files from a specific year and divide the total number of actual murders by the total number of case filings, he said. 

"A correction factor of 1.3 would mean 130 deaths for every 100 recorded cases," Nikhil added.

Another method, capture-recapture, cross-references police, media, human rights and investigation records to estimate deaths every source missed. Peru's Truth and Reconciliation Commission used it to establish the true toll of the country's internal conflict.

Md Mazharul Haque, a retired additional deputy inspector general of police and now chief executive officer of the National Security Cell, said the absence runs through the force. "Data is a major factor for the police. The more data you have, the more you can analyse it to study crime patterns and conduct research," he said. 

"Currently, our police force lacks a strong research culture; the practice simply is not there. But I believe there is immense potential." 

Japan and Singapore built policing on data

Mazharul points to Japan's koban system, which Singapore replicated and which he observed first-hand in both countries. "What is their primary focus? Data. Data is their core strength," he said. 

"The officer in charge of a local police box knows exactly who lives within their jurisdiction. Consequently, their community data is perpetually up to date," he added. 

Mazharul added digital coverage as a second pillar. 

But, he also cautioned, in a country where surveillance has been turned against dissidents, that pillar needs firm legal safeguards. 

A parallel system, not a demolition

Dewan Tahmid, chief coordinator of the Crisis Documentation Center (CRIDO), said a parallel system would be more useful. 

"In a parallel statistical model, the database could be updated based on when the crime actually occurred rather than when the case was filed," he said. "If a new case is registered for an old incident, the statistics for that specific previous month would be retroactively updated."

He also called for data by division, district and upazila, published on a webpage with Excel and CSV downloads instead of PDFs.

"Such data would not only be invaluable for researchers and analysts but would also assist the government and law enforcement agencies in formulating effective policies," Tahmid said.

Nikhil Chandra Nath suggested adding disclosure of whether figures count cases or victims, a clear distinction between missing data and a true zero, and a published correction factor with a margin of error. 

In the long run, he proposes a national victimisation survey run by the Bangladesh Bureau of Statistics.