Background

In 2021, Freedom House reported a significantly low scores in political rights and civil liberties for the Philippines. This came after the Duterte administration failed to secure peace treaties with the CPP-NPA and more, declared them as terrorists. Journalists, activities, and opposing political parties also became main targets of harassment and were labelled as communist sympathizers which resulted in increased risks of physical assault from state forces. The continuous deterioration of human rights in the country was further accelerated by the administration's draconian "War on Drugs", normalizing impunity especially in the smallest units of state forces.

  • Research Questions

    | Who are the specific actors behind cases of civilian casualty in the Philippines?

    | Do reports of civilian victimization in the Philippines show patterns (e.g. certain times of the year) on when they are more likely to occur?

  • Hypotheses

    Null hypothesis (H0): There are no specific actors behind civilian casualty. That is, it occurs randomly.

    Alternative hypothesis (Ha): There are specific actors behind civilian casualty in the Philippines; fatal violence against civilians is intentional.

  • Action Plan

    Our approach uses Data Science to analyze and visualize the factors of interconnection (location, date of occurrence, actors, types of event) of civilian victimization since the Duterte administration, as well as examine the timestamps and frequency of these incidents to identify the trends in their occurrence.

Collection

Sourced entirely from the Armed Conflict Location and Event Data Project (ACLED), the data set consists of information regarding conflicts that have transpired in the Philippines since the inauguration of former president Rodrigo Duterte on June 30, 2016.

Exploration

Our data exploration reads like a breaking news story. View and follow our live updates in each step of our data investigation.

  • Preprocessing

    The ACLE dataset underwent preprocessing in Google Colab, where we removed unnecessary data, managed missing values, encoded categorical data, and a little touch of natural language processing in order to arrive at the refined version utilized in our study.

  • Visualization

    We utilized Python alongside the Plotly Library to generate interactive plots and visualize our data, and obtain insights that answer our research questions.

  • Hypothesis Testing

    Since two categorical variables, inter1, which determines the type of actor, and is_fatal, which denotes whether an encounter resulted in a casualty, were being tested for association, the Chi-Square Test of Independence was utilized.

Research Question 1: Who are the specific actors behind cases of civilian casualty in the Philippines?

For the first graph, Sunburst was chosen to show the most common civilian groups that were targeted in conjunction with their frequent offenders without overcrowding the graph. While the graph is self-explanatory, there are two additional notes about it.

  1. The government group includes government workers, political candidates, and both former and incumbent officials.

  2. Civilians are victims with no specific designation.

Research Question 2: Do reports of civilian victimization in the Philippines show patterns (e.g. certain times of the year) on when they are more likely to occur?

Notice that the peaks all occurred during the first half of the year. The modal peak month was August which had been the peak for three consecutive years from 2017 to 2019 whereas the peak with the highest reports happened in July 2016.

To contextualize these data points, here are major news that occurred at the same time as the peaks in the graph:

July 2016 (642 cases)
The launch of Duterte’s gruesome “War on Drugs”. In its first month, 642 civilian victimization cases were reported.

August 2017 (254 cases)
The brutal murder of Kian Loyd Delos Santos, a victim of War on Drugs and one of 254 civilian victimization cases that month. Hundreds marched as Kian laid to rest.

August 2018 (165 cases)
Weeks after Duterte’s SONA Speech, where he mentions the continuous rampage of War on Drugs: “...the war against illegal drugs is far from over.” “...it will be as relentless and chilling, if you will, as on the day it began.”

August 2019 (150 cases)
“Spend it so that you can solve the problem,” Duterte said as he talked about the Billions he gave to the PNP for drug war intel work.

January 2020 (100 cases)
Duterte threatens to end military deals after the USA’s decision to deny entry of Bato de la Rosa due to allegations of extrajudicial killings.

January 2021 (87 cases)
With heightened powers during the COVID-19 pandemic, police forces continued to be entangled in violent encounters with drug suspects.

May 2022 (58 cases)
Voters and government personnels were caught in the crossfire between fierce political tensions as the historic election, involving a Marcos scion, engulfed the entire nation.

Apr 2023 (47 cases)
The shift to the Marcos administration witnessed government workers become the new targets of civilian victimization.

Null hypothesis (H0): There are no specific actors behind civilian casualty. That is, it occurs randomly.

Alternative hypothesis (Ha): There are specific actors behind civilian casualty in the Philippines; fatal violence against civilians is intentional.

To address the hypotheses, two variables will be compared, inter1, which determines the type of actor, and is_fatal, which denotes whether an encounter resulted in a casualty. As both variables are categorical, they will be subjected to the Chi-Square Test of Independence to test the relationship between them.

Actor Not Fatal Fatal
State Force 150 3833
Rebel Group 98 303
Political Militia 383 3896
Identity Militia 16 72

The test results, whose computation is available on the Colab link below, show that a p-value < 0.01 < 0.05 with dof = 3. Hence, the null hypothesis is rejected. That is, there are indeed specific actors that enact violence on civilians, causing their deaths.

Model

Imagine walking down a dark alley in the city and encountering a police officer…

How safe do you think your life is during this encounter?

Here’s our machine learning model, designed to predict whether your encounter is fatal or not based on some parameters.

Logistic Regression was used to predict whether a certain event was fatal or not based on some features of the dataset.

  • The model was able to provide around 93% accuracy of response to unseen data based on its given features.
  • Some of the features influencing the model's predictions are whether the parties involved were drug suspects, police forces, or civilians.
  • This indicates that the fatality of a situation highly depends on the presence of these groups in an encounter.
  • This is further solidified by the findings of our Exploratory Data Analysis (EDA), where drug suspects are the largest targeted group and the police forces being the largest group among their offenders.
  • Given the accuracy of these predictions, these features become crucial in understanding the dynamics of fatal encounters, shedding light on who the real victims and suspects are, and offering a clearer view of who Filipinos are really dying from.

Conclusion

State authorities, who are tasked with enforcing the law and protecting life are revealed as the primary perpetrators of civilian victimization and casualty in the Philippines. Additionally, the present state of Philippine society exposes the common Filipino to political violence more if they are a member of a particular group, such as government employees or suspected drug users.

What does this mean for us?

The moment has come for the Filipino people to hold those in authority responsible. As the credibility of justice institutions in the Philippines slowly corrodes, we must advocate for independent, transparent, and people-centered investigations into these human rights issues. We must, however, proceed cautiously because, given the current political climate and the legal validation of red-tagging in the form of the Anti-Terror Act of 2022, public dissent is likely to be perceived as a threat. As individuals, the least we can do is to initiate online and offline campaigns, create and join support groups for the victims, and exercise caution when interacting with the reported offenders.

What can we do more?

  1. Although ACLED labels the actors involved in a case with sufficient clarity, some data points lack information on the demographics of the victims. Manual labeling of such is recommended to reach more accurate and comprehensive conclusions.
  2. Columns that were dropped, such as the notes column that provides additional context about the event, can undergo Natural Language Processing in order to more features to and further improve the overall accuracy of the logistic regression model.

About Us

Curious for more? Feel free to contact us or visit our repo!

  • David Andal

    dcandal@up.edu.ph

    Hallo! I'm David, currently in my second year, taking up BS Computer Science at the University of the Philippines - Diliman. My dream is to become a game developer, using Data Science to enhance gameplay and create unforgettable experiences for future players. You'll find me working on small game projects or playing online games.

  • Seb Soriano

    jlsoriano1@up.edu.ph

    Hiii, I'm Seb, a Sophomore (see: third-year) Computer Science undergrad from the University of the Philippines - Diliman. I hope to contribute to medical diagnostics some time in the future. Outside acads, I enjoy slice-of-life lighthearted dramas and reading, sometimes writing, poetry!