Photo by Kimberly dela Cruz

Building a dataset of publicly available information on killings associated with the antidrug campaign

Clarissa C. David, Ronald U. Mendoza, Jenna Mae L. Atun, Radxeanel Cossid, and Cheryll Soriano


In May 2016, Rodrigo Duterte was elected president of the Philippines, winning decisively with 39 percent of the vote, which represented 16 million Filipinos. During the campaign, the former mayor of Davao City vowed to wage war against what he argued was the most dangerous threat to the country ̶ illegal drugs1. Through a singular platform of peace and order, he promised a swift end to drugs, crime, and corruption2 in the Philippines

Amid the rising death toll in acknowledged police operations, there was also a surge of violent killings throughout the country, particularly in the National Capital Region (NCR). This rash of violence led to the killing by unidentified assailants of thousands of Filipinos suspected to have been involved with drugs, in what have been labeled “vigilante-style” killings.

As soon as Duterte assumed office, the Philippine National Police (PNP) was tasked to operationalize and mobilize a nationwide antidrug campaign called Oplan Double Barrel3. This consisted of various forms of police action, including gathering intelligence for identifying drug users in communities, door-to-door visits, imposing random drug tests, and police operations such as entrapment, community sweeps, and armed raids4. Many police operations have resulted in the deaths of suspects alleged to have been users or dealers of drugs, usually methamphetamine, commonly known as shabu.

Amid the rising death toll in acknowledged police operations, there was also a surge of violent killings throughout the country, particularly in the National Capital Region (NCR). This rash of violence led to the killing by unidentified assailants of thousands of Filipinos suspected to have been involved with drugs, in what have been labeled “vigilante-style” killings5.

The “drug war” declared by the Duterte administration has attracted international attention amid allegations of widespread human rights abuses, including extrajudicial killings (EJKs) and summary executions by law enforcement officials6. Between 4,000 and 20,000 people were killed in the first 18 months of the administration, according to various estimates7. The PNP recently acknowledged 4,270 deaths resulting from police operations between July 1, 2016 and May 21, 20188. In addition, there were 22,983 possibly drug-related deaths under investigation during that period9.

The PNP recently acknowledged 4,270 deaths resulting from police operations between July 1, 2016 and May 21, 2018. In addition, there were 22,983 possibly drug-related deaths under investigation during that period.

The wide ranges in estimates of the death toll in the antidrug campaign are largely the result of an opaque system of collecting government data and statistics. This is exacerbated by government backtracking on some of its own statistics on drug-related deaths multiple times, often reclassifying counts into different categories10. The lack of reliable information raises important issues of accountability, especially for the killings, few of which have been independently investigated. Moreover, if the veracity of government data is an issue, then it would be difficult to assess fully how effective the antidrug campaign has been. Policymakers at the national level, therefore, may find this dataset useful because it contains granular information that may provide insights into how the antidrug campaign is being implemented by local government units.

Because an accurate count of the drug campaign deaths has been so elusive, several media organizations have compiled their own lists of those killed since the so-called “war on drugs” began. They have initiated in-depth examinations of reported government statistics11 and collated both media reports of violent incidents leading to deaths and reports from civil society groups. National news organizations such as the Philippine Daily Inquirer12 and ABS-CBN13 began publishing updated lists of deaths since late 2016. The Inquirer called its own page The Kill List. Both Rappler and the Philippine Center for Investigative Journalism (PCIJ) published stories that documented how the government changed its own official death counts. Many of these efforts have since stalled as news outfits trained their attention on other urgent stories. As of this writing, only ABS-CBN’s list remains online and updated, while the Inquirer’s The Kill List has not been updated since February 16, 2017.

Given the uncertainty as to the number of people killed in the government’s antidrug campaign, efforts to document individual cases of drug-related killings are important not only for short-term monitoring of violations of human rights and of due process, but also in any longer-term attempt to understand and evaluate the government campaign. This study assembles the first national victim-level dataset of individual drug-related deaths during most of the initial phase of the Philippines’ antidrug campaign. The data coverage spans May 10, 2016, to September 29, 2017.

“Efforts to document individual cases of drug-related killings are important not only for short-term monitoring of violations of human rights and of due process, but also in any longer-term attempt to understand and evaluate the government campaign.”

The study uses data culled from publicly available information, mostly from media reports and others from publicly accessible court filings. To the best of our knowledge, this represents the most comprehensive available victim-level list of killings in the Duterte administration’s antidrug campaign known popularly as Oplan Tokhang. We have collected data on 5,021 people killed in the campaign. They represent only a small portion of the total number reported as aggregate statistics by government and nongovernmental organizations (NGOs). While this initial effort represents only a fraction of the total, it currently captures a significant portion of publicly available incident and victim information on these deaths.

In this paper, we introduce the dataset and describe the procedures for building it, alongside a presentation of statistical analyses examining the geospatial and temporal spread of the deaths. We also empirically examine differences across types of killings. The objective of this research is to build a victim-level dataset of documented killings during the Duterte administration’s antidrug campaign. The analyses have the following research objectives: (1) to describe the spatial spread of the killings across provinces of the country and within the cities of the National Capital Region, (2) to examine patterns of types of incidence that are linked to the deaths, namely police operations as opposed to deaths resulting from attacks by unidentified assailants, and (3) to document the temporal spread of incidences and situate them in the historical developments related to shifts in the antidrug operations.


Documentation collected for this dataset comes almost exclusively from media reports, with others collated from various groups’ monitoring of media. All information on drug-related killings included here are from publicly available broadcast, print, or online news publications of reputable media agencies in the Philippines. The original data sources, comprised of web pages of news stories and videos of television broadcasts, were archived and catalogued in a manner consistent with the victim case ID number in the datafile.

The initial lists of media stories and identifiable individual victims were sourced from the two largest lists compiled by ABS-CBN and the Philippine Daily Inquirer. Both lists were built from automated Google Alerts set up by monitoring teams at each media organization and from internal news reports filed by their staff.

Inclusion criteria. Deaths related to the government’s antidrug campaign include those of people who allegedly had ties to drugs. We sought to collect information on deaths that took place from the day after Mr. Duterte was elected president, May 10, 2016, to December 31, 2017. Due to incomplete information, however, the dataset here covers the period May 10, 2016, to September 29, 2017, when the project stopped adding new cases to the dataset. Victims who were injured in drug-related incidences and survived are not included in the analysis.

A death is included in the dataset if the following criteria are evident: (1) a person was killed by violent means such as gunshot, stabbing, or beating; (2) the report or source mentions that the victim is alleged to have had a history of drug use; (3) the report or source identifies the victim as someone who was linked to drugs as a pusher or dealer; (4) the report or source mentions that the victim was on the drug watch list, or had a drug-related case filed against him/her, and: (5) the report or source states that the death is linked to, or is a result of, the government’s antidrug campaign.

Sources of cases. The vast majority of the victims’ information and data points in the June 2018 version of the dataset are derived from stories published in the news (98.6%). These news outlets include online websites and channels that broadcast content from primetime network television news programs (from YouTube), newspapers, tabloids, radio stations, and web-only news outlets. A much smaller proportion of the victims’ information in this datafile comes from publicly available legal case filings (e.g., Morillo case, San Andres Bukid case) which include sworn affidavits by the victims’ family members and witnesses of incidents. A very small amount of information on victims comes from press releases of regional or provincial police offices.

All information on victims in this dataset comes from sources that are either verified by media through their reporting and editorial fact-checking, or sworn statements from legal documents. This means every case in the dataset represents an individual victim, and the total number represents a conservative minimum of documented drug-related deaths since the Duterte administration started. Incidents that were not covered by media, those that were not reported by families, and in cases where the incidents are geographically far from the reach of national media, have a lower likelihood of being included in this dataset. For these reasons, we acknowledge that this dataset is likely an underestimation of the true number of deaths linked to the government’s antidrug campaign during the period.

Coding procedure. Members of the research team conducted content coding of all documentation related to victims; the same coding procedure is followed regardless of type of source (i.e., news or narratives). Four coders received training on how to use the code sheet and how to enter data. The same code sheet guide is used by all coders, and additions and revisions to the code sheet are documented and circulated to the rest of the team for adoption. Each coder reads, views, or listens to the news story or narrative and enters information indicated in the code sheet when available in the story. Cells are left blank when no information is available. Almost all variables are specified as binary (e.g., 1 or 0). Not all variables specified in the code sheet have available values. A summary table of the percentage of cases with valid values can be found in the project documentation files.

Variables in the dataset

There are broad categories of information coded into the dataset; each set is explained below. Specific definitions of each variable are available in the content coding sheet which can be found among the project documentation files.

Victim information. Standard information on the victims is coded for in the data sheet. It includes full names when available, aliases, age, gender, and occupation. (Some of these details, however, are not revealed in this study to protect the privacy of the victims and their families.) Victims can be “unidentified,” as they often are in cases where bodies are found away from the site of the killing and outside their community of residence. Unidentified victims are labeled as “unidentified,” and cross-checking is done to minimize the risk of a single unidentified victim appearing more than once in the dataset. This is done by triangulating various data sources to ensure that the victims are unique and distinct and that there is no double-counting in the dataset we developed.

Incident information. Details on the circumstances of the killing are coded for, with a different set of variables for those killed as a result of a police operation, those killed by unidentified assailants, and those whose remains were found away from the scene of the killing. The last of these, sometimes referred to here as “body dumps”— a term coined by the media for dead bodies discovered, usually on the street -̶ include the initial rash of killings in 2016 when men were found dead on streets, their heads covered with packaging tape and their hands and feet bound. Basic information for each incident includes date and time, location, site of killing, and context (e.g., outside business establishment, inside or in front of home).

More than half of the incidents documented here are killings that happened in the context of an acknowledged police operation, and details of the police operations are recorded only as they were reported by the police and the media stories. These include the type of police operation (e.g., buy-bust operation, serving a warrant, sweep or one-time-big-time operation, and others), whether a police officer was interviewed for the news story, and whether drugs or guns were found with the victim.

Data access statement

Access to the data can be arranged through research partnerships with the primary funding institution and through the corresponding author. The data cannot be made openly available to the public to protect the privacy and safety of the victims’ families and the safety of entire communities which are vulnerable to further violence while the anti-illegal drug campaign is in effect. Aggregate tables of all variables used in the analyses are included as supporting materials, as a form of limited data release.


Total deaths and dates of deaths

The dataset covers 5,021 individual victims. Of this total, 79.8% are identified by their real names, while another 9.3% are known only by their aliases; the rest are unidentified (10.9%). Over half of the deaths, or 55%, are linked to police operations. These deaths were reported in news stories that quoted police officers, including police chiefs, who recounted the circumstances of the killings. The remaining 45% of deaths were killings  by unidentified assailants: 38% involved witnesses to the killings; and another 7% involved so-called “body dumps,” victims whose bodies were discovered away from where they were killed.

In media reports, each victim’s alleged drug connection was described in various ways, such as: person killed appears in the local drug watch list (22.8%), victim was an alleged drug user or dealer, person previously surrendered in antidrug operations (10.6%), or a friend, family, or community member of the victim said the person used drugs. The labels used by the media or the police include: drug pusher or dealer (46.6%), drug user or addict (8%), drug courier or runner (1.1%), narco-politician (1.3%), narco-police (1.2%), or drug lord (0.6%). In the vast majority of these deaths, no charges have been filed against the alleged drug personality.

Figure 1 shows the number of deaths each day between May 10, 2016, when Duterte was proclaimed president-elect, and September 29, 2017. During this 16-month period, there was a clear pattern of surges in killings and periods of relative silence. Rodrigo Duterte won the election held on May 9, 2016, and was proclaimed president on June 30. The very next day, July 1, 2016, there were 39 recorded deaths. But there were reports of killings even before he officially took office. The first high death-count period was June 19 and 20, 2016, when the tally went from single digits per night to 16 and 11 deaths, respectively, on each of those days. The high rate of deaths continued throughout the months of July until September of 2016. To some extent, there was some reduction in nightly killings after this initial period even though there were still days when more than 10 deaths were reported.

Figure 1. Number of deaths by day from May 10, 2016, to September 29, 2017


Note: Total cases in Philippines: 5,020; Total cases in NCR: 1,999. One case was excluded because the exact date of the incident was not available.

Death counts per night in Metro Manila are separated out in orange in Figure 1. The pattern suggests that a disproportional number of deaths per day occurred in Metro Manila compared to the rest of the country. In the early months of the campaign, there were more killings in other parts of the country. By around November 2016, however, the rate of killings apparently abated in the rest of the Philippines, but remained high in Metro Manila. Further examination of the data in the provinces reveals that areas directly neighboring NCR also registered high death rates, namely Bulacan in the north of Metro Manila.

The first sustained drop in rate of killings per night began in late January 2017, when Oplan Tokhang was suspended temporarily as a reaction to revelations that the kidnapping and murder of a Korean businessman in October 2016 was perpetrated by police officers inside police premises14. This suspension was declared on January 30, according to a Reuters report. The sharp reduction in deaths, below 10 per night on most nights, started on January 27. On March 6, operations were relaunched under a new moniker, Oplan Double Barrel Reloaded. Ten people were reported killed on the same day.

These recalibrations in government policy are revealing because the bulk of the deaths follow the period of intense government-led operations. Yet, during the same period, a considerable number of deaths were not from police operations, including body dumps.

There was another major surge of killings in August 2017. On the first three days of that month, 44 people were reported killed. A high death rate continued through the next two weeks, with a peak of 49 confirmed killed on August 15, many resulting from community sweep operations in Quezon City. This was the day before Kian Lloyd delos Santos was killed. Kian was a 17-year-old Grade 11 student whose death was captured by CCTV cameras as police officers dragged the struggling teenager to an alley where he was shot to death. Kian’s brutal murder inspired a wave of criticism and protest against the government’s war on drugs. And throughout the rest of August and September 2017, there were markedly fewer deaths per night. On October 12, 2017, a few days beyond the data included in this paper, the antidrug operations were taken out of the purview of the PNP and placed under the Philippine Drug Enforcement Agency (PDEA).

In the 456 days since the administration was installed, there were only 17 days with no deaths reported in the media. These days included Christmas (December 25, 2016), 4 days in February 2017 after Tokhang was suspended the first time, and 2 days in August 2017, after the news about the killing of Kian delos Santos became controversial. The curious coincidence between government decisions to stop or scale down police operations and the relative intensity of death incidences (even those linked to body dumps and unidentified assailants) highlights questions on the role of government in the killings.

Figure 2 shows the distribution of deaths by type of incident by month, namely whether these were a result of acknowledged police operations or of murders by unidentified assailants. The deadliest months were the first three after Duterte was sworn in: July 2016 (738 deaths), August 2016 (669), and September 2016 (690). The least deadly month was February 2017 (88), the month after the suspension of police operations.

In the period between June and August 2016, there was a rash of killings that also featured bodies found in various parts of cities, bound, gagged, or with heads wrapped in packaging tape and often with signboards that said “I’m a drug dealer.” Many of these victims remain unidentified. In some cases, the victims were not even known to residents of the communities where they were found in what appeared to be “body-dump” cases. These incidences starting dropping off after September 2016 and remain rare to this day.

Figure 2. Number of deaths by type incident, by month


*Numbers for May 2016 cover only May 10-31 and for September 2017, only up to the 29th of that month. One killing had no date so is not included on this chart. There was not enough available information on six cases, so these could not be categorized by type.

A number of deaths resulted from attacks by unidentified assailants, which we designated as “killings not due to police operations.” This study shows a trend line that closely tracks the monthly death counts due to police operations. In February 2017, there was a complete drop-off of police operations deaths while deaths from other killings declined as well, but less sharply. Once Tokhang was reinstated in March, deaths due to police operations began to rise again.

Who are the victims?

According to the data, the vast majority of victims were males (94%) between the ages of 20 and 50 (30%). Almost all victims died of gunshot wounds. regardless of whether they were killed during police operations or by unidentified assailants. The victims’ ages were reported in only 1,764 cases. Of these, 22 victims were below the age of 18, another 228 people were between 18 and 25 years old, and 225 were at least 50 years old, including 33 who were 60 or older. The average age of the victims whose ages are known is 37.

The data indicated an occupation for only a small portion of the total number of victims (15.8%). In cases in which the victims’ occupation was known, the majority were low-paying, low-skilled work. There were 98 tricycle drivers, 32 construction workers or carpenters, 24 street vendors, 19 jeepney barkers or dispatchers, 16 farmers, 12 jeepney drivers, 15 habalhabal and pedicab (bicycle) drivers, and 7 garbage collectors. Thirty-eight were reported as unemployed. Based on their place of residence or their occupation, it is clear that most of the victims were poor.

Some of the victims were government officials, particularly at the level of local government. Many were current or former barangay officials. The dataset shows that  these victims include 38 barangay chairpersons, 2 former barangay chairpersons, and 46 current or former barangay councilors. Another 28 barangay employees, including tanod (watchmen) and treasurers, were also killed. In addition, a total of 107 current and former police officers died during the antidrug campaign. Most of them (65) were active-duty officers at the time of death. Of these, 11 were killed while conducting police operations. The rest were killed as “targets” of police operations or died in the hands by unidentified assailants. The list of casualties also includes 11 current and former soldiers and 9  police assets or informants.

Overall national and NCR city distribution of deaths

Killings in the National Capital Region (NCR), which covers 16 contiguous cities and 1 municipality, comprise 40% of all the victims included in this dataset. This is a disproportionate share, considering only 13% of the country’s population resides in the NCR.

Table 1 shows regional differences in total killings and by type of incident. While at least some of this heavy representation of NCR in the death count may be due to access by national media outfits, all of whom are based in Metro Manila, there are some independent accounts that the drug war death toll has indeed been more focused on this region. Outside of the NCR, the regions with a high number of deaths are neighboring Region III (Central Luzon) and Region IV-A (CALABARZON), a densely populated set of provinces with large urbanized communities, and Region VII (Central Visayas), which is where one of the largest cities outside of the NCR is located.

Table 1. Percent of Deaths by Region and by Type of Incident within Region

REGION No. of deaths % of all deaths % of killed in police operations in region % of killed in non-police operations in region % of region population to national populationa
NCR (National Capital Region) 2,000 39.8 49.3b 50.7b 12.75
I – Ilocos Region 224 4.5 21.9 78.1 4.98
CAR (Cordillera Administrative Region) 44 0.9 20.5 79.5 1.71
II – Cagayan Valley 141 2.8 25.5 74.5 3.42
III – Central Luzon 916 18.2 69.9 30.1 11.11
IVA – CALABARZON 517 10.3 64.0 36.0 14.27
IVB – MIMAROPA 11 0.2 72.7 27.3 2.93
V – Bicol Region 121 2.4 72.5c 27.5c 5.74
VI – Western Visayas 83 1.7 55.3 44.6 7.46
VII – Central Visayas 460 9.2 52 48 7.33
VIII – Eastern Visayas 65 1.3 52.3 47.7 4.4
IX – Zamboanga Peninsula 65 1.3 56.9 43.1 3.59
X – Northern Mindanao 70 1.4 47.1 52.9 4.64
XI – Davao Region 92 1.8 76.1 23.9 4.85
XII – SOCCSKSARGEN 106 2.1 80.2 19.8 4.5
XIII – CARAGA 73 1.5 50.7 49.3 2.57
ARMM (Autonomous Region in Muslim Mindanao) 33 0.7 87.9 12.1 3.74
Total 5,021 100.00 54.9 45.1 100

a. Population estimates are based on the 2015 Census of Population.

b. Base for NCR is 1,995 cases. Five cases have not been categorized by type of incident.

c. Base for Bicol Region is 120 cases. One case has not been categorized by type of incident.

Compared to the rest of the country where most of the deaths are attributable to police operations, the killings in the NCR were almost equally split between police operations (49%) and non-operations (51%). In contrast, over 60% of deaths in the regions of Central Luzon, Bicol, Davao, SOCCSKSARGEN, ARMM, and MIMAROPA were due to police operations. In the CAR, Cagayan Valley, and Ilocos regions, less than 30% of deaths were linked to police operations, and the majority of victims were killed by unidentified assailants.

Within the NCR, the distribution of the 2,000 deaths was also significant (see Table 2 and Figure 4). Drug-related deaths in the NCR were concentrated in Manila City (23%), Quezon City (20%), and Caloocan (19%).  Overall, deaths in these 3 cities comprise 62% of the total deaths in the NCR and 25%, or one-fourth, of total drug-related deaths in the dataset. Even within these large cities, the spatial spread of the killings indicates concentration in some barangays, while others have remained unaffected.

Outside NCR, as shown in Figure 3, Bulacan has the highest number of deaths (644), of which 73% were the result of police operations.

Analysts have raised important governance issues, given the way some local governments appear to have been less affected by antidrug campaign deaths compared to others. (Mendoza et al, 2018). For instance, Muntinlupa, Pateros, Navotas, San Juan, and Valenzuela cities have far smaller numbers of total deaths, compared to both Manila and Quezon City. However, normalizing by population size (number of killings per 100,000 population) shows that Pateros and Navotas both have high rates of killings given their small population. Quezon City, on the other hand, which has a high total death count but large population, is not among the deadliest cities in NCR..

Figure 3. Map of Philippine provinces and NCR cities with death counts


Similarly, there are marked differences in the proportion of deaths by city in the NCR attributable to police operations and those murders committed by unidentified assassins. Table 2 below reveals that while in Manila City more than three-quarters of all deaths (78%) resulted from police operations, in neighboring Makati City it is only 10%. In Quezon City, police operations resulted in 258 deaths, comprising 65% of all drug-related killings in that city; in the neighboring city of Pasig, police killed 20 people, which is 13% of the total death toll of Pasig.

Table 2. Percent of Deaths in NCR Cities by Type of Incident

City/Muni-cipality No. of deaths % of all deaths in NCR % of killed in police operations in city % of killed in non-police opera-tions in city % of city popula-tion to NCR popu-lationa Deaths per 100,000 peoplea
Manila 463 23.2 78.2b 21.8b 13.8 26.0
Quezon City 400 20.0 64.5 35.5 22.8 13.6
Caloocan City 373 18.7 51.2 48.8 12.3 23.5
Pasig City 156 7.8 12.8 87.2 5.9 20.7
Pasay City 118 5.9 19.7c 80.3c 3.2 28.3
Navotas City 79 4.0 53.2 46.8 1.9 31.7
Marikina City 60 3.0 23.3 76.7 3.5 13.3
Mandaluyong City 57 2.9 17.5 82.5 3.0 14.8
Makati City 52 2.6 9.6 90.4 4.5 8.9
Taguig City 43 2.2 25.6 74.4 5.2 5.3
Parañaque City 43 2.2 16.3 83.7 6.3 6.5
Malabon City 36 1.8 16.7 83.3 2.8 9.8
Las Piñas City 34 1.7 35.3 64.7 4.6 5.8
Muntinlupa City 25 1.3 36.0 64.0 0.5 5.0
Pateros 24 1.2 8.3 91.7 0.9 37.6
San Juan City 20 1.0 10.0 90.0 3.9 16.4
Valenzuela City 17 0.9 70.6 29.4 4.8 2.7
Total 2,000 100 49.3 50.7 100 15.5

a. Population estimates are based on 2015 Census of Population.

b. Base for Manila is 459 cases. Four cases have not been categorized by type of incident.

c. Base for Pasay City is 117 cases. One case has not been categorized by of incident.

People killed in the context of acknowledged police operations

We take a closer look at deaths resulting from police operations, of which there are 2,753 recorded in the dataset. Not all circumstances and details of the operations were recorded in the original data sources; however, for those where details are provided, we coded for the type of antidrug operation being conducted, the location of deaths, and allegations of the presence of drugs and guns at the scene.

Figure 4. Deaths by police operations and unidentified assailants, NCR by city


The majority of police operation deaths recorded in this dataset were in the context of buy-bust operations (Table 3), where members of the police force pose as dealers or buyers and capture those who attempt to buy or deal (58%). A much smaller share of deaths (15%) was in the context of serving a warrant. These are cases that seem similar to the killing of Ozamiz City Mayor Parojinog who was killed in his home together with 14 other people while the police were serving a warrant15. Very small percentages of the deaths were linked to other forms of operations such as raids (6%), checkpoints (4%), and police sweeps in communities (3%).

Table 3. Percent of killed in police operations, by type of police operation (multiple responses)

Type of police operation % of killed in police operations
Buy-bust operation 58.3
Serving of warrant 14.7
Police raid 6.0
Checkpoint 3.6
Police sweep or search 3.2
Others 4.3
No specific type of police operation mentioned 3.9
Total killed in police operations 2,753

In around 90% of the deaths linked to police operations, a reason was given for the killing of a suspect, typically by police officials interviewed by media at the scene (Table 4). In some cases, police offered multiple reasons for a death in the context of an operation. Each instance in which a reason was given for the killing was coded in our dataset. Seven times out of 10, police said there was a shootout that led to the killing of a suspect. Two times out of 10, police claimed they acted in self-defense because the suspects  posed threats to the officers’ lives (e.g., they pulled out a gun, grabbed the officers’ gun).  In media coverage of the antidrug campaign, these explanations became known as “nanlaban” or “resisted” or “fought back,” a general label the police use to mean any of these reasons.

Table 4. Percent of killed in police operations, by reason cited for killing (multiple responses) 

Reason cited for killing % of killed in police operations
Person engaged in a shootout 73.7
Police acted on self-defense 19.4
Person resisted arrest or search 14.8
Person attempted to run or flee 9.1
No reason mentioned 9.7
Total killed in police operations 2,753

When a killing occurs as a result of police operations, the police often reported the presence of a gun or guns at the scene (76%) and the presence of drugs (64%). In some cases, they also reported the presence of drug paraphernalia (15%). In cases of murder by unidentified assailants the numbers are much fewer—only 20% of the crime scenes have guns, only 11% have physical evidence of unused drugs, and only 3.5% have drug paraphernalia.

Table 5. Presence of guns, drugs, and drug paraphernalia

Presence of guns, drugs and drug paraphernalia % of all types % of killed in police operations % of killed by assailants
Presence of gun/s 50.0 76.1 20.2
Presence of drugs 40.5 63.9 10.5
Presence of drug paraphernalia 9.6 14.7 3.5
Total number of deaths 5,021 2,753 1,907

Total number of deaths in the dataset is 5,021 but six cases have not been categorized by type of incident.

People killed by unidentified assailants

Persons killed in non-police operations make up 2,262 cases (45%) in the dataset. These include victims who were killed (usually shot) by assailants on a motorcycle and found at the spot where they had been slain (38%). It also includes victims who were “dumped” in an area but apparently killed in another location (7%). The scale of these type of killings started rising in July 2016 after the Duterte administration was officially installed. There was a rash of incidents in the first two months of Duterte’s presidency that featured dead bodies turning up on public streets, bound and gagged, with signboards saying they were drug dealers or users.

These incidents underscored the fact that the antidrug campaign involved not just official police operations but also some form of vigilante action. Eventually, a rash of killings by unidentified men riding tandem on a motorcycle began. These were also connected to drugs since the victims were later alleged to have been into drugs in some way. The drug connection was also established in news reports in which the victims were identified as having been either suspected users or dealers. Some were reported to have been on the local drug watch list. Others had handwritten signs on them accusing them of being a drug dealer or user. In some cases, new accounts quoted the victims’ families mentioning alleged drug connections.

Of the cases where killings were committed by unidentified assailants, 46% involved assailants who used a motorcycle for the attack. In 28% of the incidents, the assailants were on foot, and in 5%, assailants were in a van or a car. Almost all of the victims (99%) were shot dead. In 14% of the cases, the killing involved one assailant, while in 27%, the attack was carried out by two people. In 17%, 3 or more assailants were involved. It is unclear in many of the cases (42%) how many assailants were involved.

Of the 355 cases categorized as “body dumps,” 45% had gunshot wounds, 37% had their heads wrapped in tape or their mouths gagged, and 35% had their hands or feet bound when they were found. Some victims had knife or stab wounds (12%) while others were stuffed in a box or sack, or wrapped in a piece of cloth (11%)

Most of the killings happened on a street or in an alley (27%). Many of the victims were killed in, or in front of, their houses (24%). If the murder was committed by unidentified assailants, 44% of such cases occurred on a street, and 29% in a house. In police operations, most of the killings happened in a house (23%). Many took place on a street (13%) and in areas identified as “drug dens” (5%). Most of the “body dumps” were on a street or alley (52%) and a small number were on vacant lots (12%), Among the victims who were killed in houses, when these were done by unidentified assailants, 79% of the dead resided in the house; in police operations, 63% of the victims resided in the house. Some of the victims killed in houses were visitors of the house-owner.

Table 6. Location of incident

Location of incident % of all types % of killed in police operations % of killed by assailants % of bodies found
Street or alley 27.4 13.0 43.7 51.5
In or near house

(% of killed in or near house who are staying or living in house)








Vehicle 8.6 3.9 16.5 2.5
Establishment 5.0 1.2 10.9 3.1
“Drug den” 3.7 5.3 2.1
Public space 2.9 2.9 2.7 3.4
Vacant lot 1.2 0.6 0.3 11.5
Bridge 0.8 0.8 0.3 3.7
Holding facility 0.7 1.1 0.2
Body of water 0.7 0.4 0.2 5.9
Total number of deaths 5,021 2,753 1,907 355

*Total number excludes six cases that have not been categorized by type of incident.

Discussion and Conclusions

This paper introduces an initiative to build an individual-victim level database of drug-related deaths that occurred during the first [NUMBER OF DAYS] of President Duterte’s administration. Creating a credible and reliable system to track and collect data on the killings is of critical importance based on broad goals of accountability and truth-telling. The Duterte government has been inconsistent in providing official statistics and information related to deaths in connection with the anti-illegal drugs campaign. That makes an independent empirical monitoring even more relevant.

This is the first research-based effort to collect data on the killings related to the antidrug campaign, including information on the casualties. This initiative seeks to develop designs systems and procedures as a base for expansion of the current data. Supporting materials such as the coding procedures and variable definitions are provided here, enabling replication by other groups in the Philippines using other types of data sources, or to document and analyze similar events in other contexts and countries.

Beyond establishing the dataset through a triangulation of various publicly available data sources, this study also seeks to expose some patterns surrounding the drug deaths. Our analysis of the temporal pattern of drug-related deaths reveals the scale of killings in the country, with rapid escalation starting in July 2016 and lasting throughout the rest of that year. Noticeable abatement can be seen after key historical points in the anti-drug campaign, with slowdowns in the rate of nightly deaths during periods when Oplan Tokhang was suspended and operations were moved to the PDEA from the PNP, as the PDEA is a much smaller enforcement unit.

Spatial distribution of the deaths was also examined relative to sub-national political units, regions for the entire country, and by city for the major hotspot of the NCR. Analyses indicate a large concentration of deaths in the NCR compared to the rest of the country, although caution should be exercised in interpreting this. Given the reliance on media-based source material, there may be a larger proportion of unreported deaths in areas outside of the NCR. Further to spatial differences, the distribution of type of incident, whether deaths resulted from police operations or unidentified assailants, also varies widely across cities and regions. Analysts are already turning to an analysis of these patterns to try and establish whether and to what extent some of the deaths were linked to the failure of local government units to respond pragmatically to protect their citizens.

We identified worrisome trends in the way many of the killings were committed. Most notable is that a large percentage of them occurred in homes, regardless of whether the incident was a police operation or attributed to an unidentified assailant. This is even more significant in official operations. The killing of suspects in private homes during official police actions raises important issues of accountability and the effectiveness of official police policy in conducting such operations. There are questions of whether these operations were conducted in contravention of rules of procedure whereby these officials are prohibited from entering houses without a warrant.

Further, the spatial differences in distribution of incident type, whether police operations or not, are stark, both at the provincial level and at the city level within the NCR. Without more in-depth detailed information, it is difficult to make conclusions about why these differences occur. It is an open question whether the differences by political boundaries, along which local government policies or economic conditions likewise vary, may provide key explanations for the variations in types of incidents.

This data-building project is an initial listing and analysis of the victims of the Philippine antidrug campaign, representing only a portion of the drug-related deaths acknowledged by the government16. For the purposes of expediency in systematizing the databasing initiative, this first set of data relied on publicly accessible information, mostly media reports. As such, the cases which make it to this data are more likely to be in the NCR, where national media offices are based, and more likely to be police operation deaths rather than killings by unidentified assailants. The latter is an important limitation that must be addressed by further work in the documentation of murders, especially since some estimates by civil society organizations in the human rights space allege that there are up to 20,000 deaths to date, a figure that is consistent with one of the later versions of the government’s own “accomplishment report.”

This study relies mainly on media sources, a limitation of this project. We speculate that the current version of the data provides a very conservative listing of victims. That is, all of the deaths in the data here can be traced to identifiable and verifiable incidences in media reports, in many cases with photographs or video to support the existence of a victim. Given the apparent magnitude of the antidrug campaign launched by the Duterte administration two years ago, it is highly likely that many more killings were not reported by media and legal groups. Our hope is to expand the scope of this initiative so that even more of the killings are eventually reported and documented.



3. Coronel, S. (2017). Murder as enterprise: Police profiteering in Duterte’s War on Drugs. In A Duterte Reader: Critical essays on Rodrigo Duterte’s Early Presidency, Bughaw, Ateneo University Press: Quezon City, Philippines.