When and why do people get involved in mass violence? What pushes individuals in a precariously tense region into brutality and blood — and what can prevent such slaughter? With gut-wrenching stories reaching us daily from Syria, Yemen, Sudan, South Sudan, North Korea, Central African Republic, Libya, Nigeria and Congo, that urgent public policy question is being tackled by a number of academics, researchers and policymakers, including the Early Warning Project, a public prevention system for mass atrocity and genocide that combines statistical modeling with insights from an expert opinion pool. A month ago, the EWP released a detailed risk assessment for such violence.

But in addition to preventing wholesale genocide, it’s important to understand when and why killers step back from killing. Everyone loves the story of Paul Rusesabagina, a Hutu who saved more than 1,000 Tutsi in the Rwandan genocide, made famous by the movie “Hotel Rwanda.”

But what about rescuers whose records aren’t so neat? How should we think about people who both kill and save others during the same genocide? War is messy and complicated. People who kill do not always and only kill. Rescuers sometimes murder or become victims themselves.

In a recent article published in Sociological Theory, I examined the idea that “once a killer, always a killer.” Drawing on several months of research in Rwanda, where Hutu civilians massacred their Tutsi neighbors in the 1994 genocide, I identified four factors that explain how and why some of these same Hutu also occasionally saved Tutsi instead.

1. Money is power.

In both peace and war, people with financial resources can do things others cannot. In a genocide, this means rich people can sometimes buy their way out of killing when poor people cannot escape the pressure to kill.

Consider the following account in “Leave None to Tell the Story” by Alison des Forges, a Human Rights Watch report on the Rwandan genocide. Two militiamen came upon a Tutsi girl — Marthe was her name — who had been pushed into a hole by a sergeant who had intended to kill her later. One of those soldiers wanted to kill Marthe; the other, who knew the girl, told him not to. Then, to ensure that Marthe would be protected, the soldier gave his peer 5,000 Rwandan francs (about U.S.$25) and pushed him away.

This soldier killed many other people during the genocide. That was his job. But when it came time to save someone he knew, being able to bribe his peer allowed the soldier to save someone without being killed himself.

2. Public examples: Were those who resisted the genocide punished or safe?

Some people shifted from peaceful to murderous actions during the genocide, as in Butare, or remained peaceful throughout, as in Giti, in the south and north of Rwanda respectively. Butare also remained peaceful for two weeks after the genocide first began in Rwanda’s capital, Kigali.

Many Tutsi believed they would be safe in Butare, in part because before the genocide, it had been well integrated, with Tutsi making up about 25 percent of the town’s population — a percent far higher than the national average.

However, once extremists took over the district, violence unfolded rapidly. The extremists not only began slaughtering Tutsi, but also publicly executed moderates in Butare who refused to join in the genocide and had called for peace in the past. Once other Hutus saw the high cost of resistance, they were far more likely to pick up machetes and join in massacring their neighbors.

By contrast, in Giti, most Tutsi survived the genocide. There were no group killings. The town’s leaders argued against slaughter — and faced no consequences. Extremists never entered the area, did not begin violence, and had no chance to target or publicly execute those moderates who refused to murder their neighbors. Where moderates remained in power and resisters did not see others punished for refusing to kill, genocide did not take place.

3. Social networks and the safety of privacy.

Knowing someone, of course, was one reason Rwandan civilians were willing to save them, despite the pressure to kill. But many were unwilling to save someone unless safely unobserved. When others involved in the killings were nearby, Hutu were less likely to try to save Tutsi, even one who was a relative or friend.

Here, Lee Ann Fujii recounts an interview with Olivier, a man who she describes as “every bit the willing executioner” in her book, “Killing Neighbors.” Olivier claims “it was impossible” to save someone during the genocide because if you tried, you would be pressured to kill them yourself. And yet Olivier saved a Tutsi during the genocide. One day, walking alone, Olivier saw a neighbor trying to escape and helped him.

When asked to explain, Olivier claimed that isolation was responsible. Not morals nor kindness, but the safety of seclusion.

4. Timing is everything.

At the start of the genocide, many were horrified by the violence. But as the massacres continued, people got used to it, and to the act of killing.

Jean Hatzfeld tells a powerful story in the book “Machete Season” about a Hutu man named Jean-Baptiste who was forced to participate. Génocidaires told him that if he did not start killing, they would murder his Tutsi wife. And so he killed. At first, Jean-Baptiste recoiled in horror. He explained, “When I saw the blood bubble up, I jumped back a step.” Later, however, he “got used to it.”

Similarly, a man interviewed by Scott Straus for his book “Intimate Enemy” with Robert Lyons, describes how, “As the days passed, people became increasingly habituated. We were no longer afraid, like in the beginning.”

What happened in Rwanda also happened in other genocides: Many people were inconsistent killers or saviors during the Armenian genocide, the Holocaust, and the mass murder of Bosnian Muslims following the breakup of Yugoslavia. Sometimes they would kill; sometimes they would rescue.

Knowing what prompts a person to kill or not as a genocide unfolds has powerful implications for intervention. For example, international organizations can identify moderate strongholds like Butare was in Rwanda and seek to offer financial or military support to maintain ongoing efforts at resistance. That timing matters for shaping civilians’ adaptations to genocide shows the importance of early intervention in a war, rather than later when participation in violence has become normalized. The Early Warning Project is an excellent start for understanding where to focus efforts to prevent genocide. Understanding how and why individuals waver between murderous and rescuing behaviors can help us understand how to effectively intervene.

Aliza Luft is a PhD candidate in sociology with a PhD minor in political science from the University of Wisconsin, Madison, and an MA in education from the University of California, Berkeley. Her research examines how people choose to support and participate in or mobilize resistance to violence.