Alex Nowrasteh
My recent posts about politically motivated terrorist killings in the United States revealed surprising findings. There are a few politically motivated killings; Islamist terrorists are still the deadliest, and, most relevant for the recent assassination of Charlie Kirk, the left-right political distribution of murders skews right, even in recent years, but the numbers are minuscule. Those findings and others prompted questions and commentary by politicians, journalists, academics, and social media users, including Elon Musk and Catturd2. Below are the most common questions and criticisms, followed by my responses.
Why did you only count fatalities? Why not injuries, property crimes, incidents, or other variables?
My posts focused on homicides committed by politically motivated terrorist killers, with a secondary focus on perpetrators. Some critics thought I should have included incidents, injuries, arrests, property crimes, or other measures of politically motivated violence. Ultimately, those critics claim that if we widen the measurement of politically motivated terrorism to include noisier metrics, then the results would change. They certainly could change, but murders are the best measure.
Deaths are the best measure of violence because all deaths report the same level of harm. They are discrete and equal, which eases comparison between terrorist attacks. All other measures have extreme variance or are otherwise vague, complicating comparison between attacks.
For instance, injuries range from scratches to amputations. Even if you still wanted to include injuries, they are correlated with the number of murders during the shorter period of 1975–2017, when I have complete data for all terrorists on US soil. Incidents are also a worse variable than the fatalities because some incidents result in more people murdered or injured than others. An incident could be a ding-dong-ditch or blowing up the house, a variance in damage so high that it’s about as useful as describing the Grand Canyon and my local creek as “water holes.” Arrests of plotters before they could attack are also a weak measure because they treat all plotters as equals when there are major differences in competence, seriousness, or potential harm. Killing somebody in an attack is also a decent proxy for the intensity of the killer’s belief.
The difficulty of finding those who did not commit attacks or only injured somebody is the mundane reason for only reporting politically motivated killers and their victims. My work documents them all for my research on foreign-born terrorism, but there are fewer. One of my reports includes extensive research on native-born terrorists, but it took months to produce, and no one was interested. Never again. These other measures of violence matter, but they matter less than murder, are more difficult to find, and are often difficult to compare between attacks.
The obvious reason why fatalities are the best metric to discuss politically motivated violence is that we are only debating this issue because of Charlie Kirk’s murder. If he were merely injured, if there were an attempt on his life that didn’t touch him, if his property was damaged, or an arrest was made in a plot that resulted in no harm, then the attention would be over 99 percent less.
Why didn’t you include people who died in riots?
Some critics argue I should have included all the people killed in the BLM riots and others as politically motivated terrorist murders. I looked at the 19 most prominent people murdered during the BLM riots. They were mainly looters shot by the police or property owners, or victims of normal crime during the riots. It would be nonsensical to count the lawful killing of a looter as a politically motivated left-wing death, especially if he were a politically motivated left-wing looter. In such a case, the looter is the attacker, not the victim.
Some critics who want looters counted as victims of left-wing politically motivated violence also want me to include Ashli Babbitt as a victim of left-wing violence, even though a police officer shot her while rioting at the Capitol on January 6, 2021. Including neither of them makes sense.
Even so, my data are a measure of politically motivated terrorist killings. State actors do not commit those crimes. Thus, the police cannot commit politically motivated terrorist killings under this definition if they’re doing their jobs as they were during the BLM riots and while defending the Capitol on January 6, 2021. If a police officer were operating independently as part of a paramilitary organization, as a member of a terroristic conspiracy, or murdered somebody for political reasons while off duty, then the killing could be politically motivated terrorism. If the police officer assassinated somebody on orders from his bosses or the president, it would not be a politically motivated terrorist killing because the government did it and not a private actor. President Trump ordered the US military to drop bombs on boats in the Caribbean. While that’s probably a war crime, illegal, unethical, and bad for many other reasons, it’s not terrorism or a politically motivated killer because it was committed by state actors who were just following orders.
The argument from riotous contagion is even more tenuous. Normal criminals who took advantage of the initial unrest committed most of the BLM riot violence. It’s tenuous to blame that on left-wingers just because they mostly supported BLM and weren’t as quick to condemn the rioting as they maybe should have been. Using the same logic, should libertarians be blamed for some increase in civil disorder because we tend to support drug legalization and some other criminal law reforms that could plausibly be linked to urban annoyances? Conservatives rightly oppose most gun restrictions, and left-wingers blame guns for gun crime. Would they be justified in blaming conservatives for all deaths by gun crime just because they support the private right to keep and bear arms? No to both.
What about the lives saved from policy decisions? It would be incorrect only to count the downsides. Should those be counted on the positive side of the ledger? Do right-wingers really want to be credited for the lives saved by Trump’s Operation Warp Speed? Or would right-wing vaccine skeptics who claim the COVID vaccines killed millions want President Trump blamed for that fictitious death toll? Expanding the methodological scope to include deaths in riots and the other social effects from policies becomes less tenable the more one looks at it.
More deeply, these critics want me to include deaths plausibly downstream from political decisions. Such a methodological adjustment would amount to a statistical analysis of the effects of policies on death rates more broadly rather than a count of politically motivated killings. Such a macrosocial analysis is likely impossible, and the results would be uncertain because then I’d have to consider other tendentious social science theories, like whether gun deaths should be blamed on right-wingers because of their opposition to gun control. This is especially thorny regarding Kirk’s murder because he’s well known for a quote, “I think it’s worth to have a cost of, unfortunately, some gun deaths every single year so that we can have the Second Amendment to protect our other God-given rights. That is a prudent deal. It is rational. Nobody talks like this. They live in a completely alternate universe.”
I agree with Kirk, for what it’s worth, but does that mean right-wingers are responsible for Kirk’s death? No. Such a result would be absurd but logical under the methodological change my critics support, which is another reason we shouldn’t embrace it.
The critics are very confused here and didn’t think through the issue. Many of them want to judge whether a killing was politically motivated based on the politics of the victim or their partisan view of different riots (BLM bad, January 6 good). I’m doing the opposite by examining the motivations of the killers. The only way to understand the motivations of the killers is to investigate the motivations of the killers.
Surveys and polling show increasing support for political violence. Why didn’t you include them?
Talk is cheap; actions reveal actual preferences and desires better than survey responses. Besides other reasons to doubt the veracity of opinion surveys, there is no good reason to rely on them when we have actual data on violence. Consider them if you want, but don’t say they are nearly as revealing as the fatalities.
Why didn’t you count rhetoric as violence?
Calling people Nazis, communists, child molesters, or traitors isn’t nice, but it is not violence. Sticks and stones will break my bones but words will only hurt my feelings, which is why adults usually get over them. Harsh words may convince some people to commit violence, but so might bad traffic or indigestion. Even if I believed that words were as bad as deeds or worthy of inclusion as a less weighty variable in my analysis of politically motivated killings, how could I possibly count this? Would I include my hate mail where right-wingers say I’m guilty of treason, which is a crime punishable by death? Should I include left-wing hate mail to my other colleagues that argues they’re killing people for opposing more government control of health care or energy, because murder is punishable by death? Should I run a Google n‑gram search for the word “Nazi” or “fascist” and see if it was written within 10 words of “Trump”? I’ve not heard a good idea about how to measure invective, even if I wanted to.
Why didn’t you go back further in time before 1975? There were many left-wing terrorists in the 1960s and early 1970s. These data are t‑hacked.
It is harder to gather biographical, motivational, and other data about killers for years in the more distant past. Years ago, I tried to go back to 1970 to analyze foreign-born terrorism in the US and hit several insurmountable data limitations, so I gave it up. Including the earlier period wouldn’t change much when it comes to deaths. For instance, the left-wing terrorist group the Weathermen (also known as Weather Underground) didn’t kill anybody in their bombings. Three Weathermen terrorists died when a bomb they were making exploded, but I don’t count the terrorists as victims when they die in their own attacks because doing so wouldn’t make any sense. The Black Liberation Army was much deadlier but killed 16 people at most. The Black Liberation Army and the Weatherman members who murdered two Brink’s security guards in 1981 are in my data.
The combined fatalities by the Black Liberation Army and the Weathermen were fewer than the 23 murdered by the right-wing nativist attacker Patrick Crusius, who wrote a manifesto, on one day in August 2019 at a Walmart in El Paso, Texas. Sometimes, when people mention the Weathermen, the bombings, and other terrorists of that period, they append the adjective “forgotten” to describe their campaigns of terror. It’s true, those attacks were largely forgotten because there were so few killings. Still, nothing is stopping the critics from extending my research backward. It would be great if they succeeded.
Your data are p‑hacked.
P‑hacking is when researchers manipulate how they analyze data until they get a statistically significant result. P‑hacking is only possible when researchers exploit sampling variability to lie with statistics until they get the desired answers. All statistical studies rely on samples. However, my data are the population and not the sample. P‑hacking is thus impossible when analyzing the entire population because a p‑value is worthless with such complete data. In fact, the numbers used to describe a population aren’t even called statistics; they are called parameters. Furthermore, there are no regressions in my analysis, and thus no p‑values were ever produced or reported.
The critics who wield this criticism likely mean that my definition of the parameters of the population is flawed, or some other mistakes were made, like over-including, under-including, or misclassifying the political motivations of some killers. I could have made different mistakes (please let me know), but that’s not p‑hacking. One mistake already discovered was that I overcounted murder victims by Shamsud-Din Jabbar in his Islamist-inspired vehicle ramming attack earlier this year. I originally recorded that he murdered 16 people when the real number was 14. So far, my biggest error was to overcount the number of people murdered by Islamist terrorists.
The Trump administration deleted a Department of Justice (DOJ) research paper that supported my findings. In fact, the DOJ report found that right-wing terrorists killed 520 people and left-wing terrorists killed 78 since 1990. If the DOJ numbers are to be believed, my research undercounted left-wing politically motivated terrorist killings by 37 and right-wing killings by 168. According to them, I missed an entire additional equivalent of an Oklahoma City Bombing committed by right-wing killers. I’m currently going through this report in detail, and it looks like it includes hate crime perpetrators whom I don’t want to include.
Disagreements Over the Ideological Motivations of Some Killers
My main message is that politically motivated violence and killings are rare, and few seem to disagree in writing. The best counterargument is that they may be rare, but they are destabilizing when important people are murdered, which I agree with. However, some readers also disagree with how I classified the ideological motivations of some of the killers. My first piece on the politically motivated killers included this line:
The motivated reader can slice and dice these numbers in different ways, count marginal hate crimes as politically motivated terrorist attacks, assign different ideological motivations to the individual attacker, and must still conclude that the threat to human life from these types of attacks is relatively small.
Indeed, readers or other researchers should use my data, create their own, develop their own methods, and make different choices. Then we can compare and hopefully discover better methods in the process. A common erroneous omission that some think I made was to exclude Devin Patrick Kelley entirely from my data set. He murdered 26 people in a truly ghastly church shooting in Texas in 2017. If he were included, critics think I should have counted him as a politically motivated left-wing killer because he was an atheist.
However, investigators’ convincing evidence showed that he perpetrated the shooting because he was in the middle of an argument with his mother-in-law, sent her threatening texts that morning, and then went to her church to kill her and others. He was an atheist who murdered 26 people because of a domestic dispute, not because of his theological skepticism. It’s good that critics bring up these cases, though, so we can all check each other.
Critics can also waste your time, such as the one who claimed that the Oklahoma City bombers, Patrick Crusius and Peyton Gendron, weren’t motivated by right-wing ideologies. Absurd. The Oklahoma City bombers were heavily into the militia movement. Crusius and Gendron were great replacement theory supporters and racial nationalists who believed in blood and soil. The critic says that they were actually “eco-fascists,” so they weren’t right-wing. What does the critic think the “soil” in “blood and soil” means? There is more ideological diversity on the political right than on the left, so the belief set is larger.
Other Reasons to Think that Politically Motivated Killings Are Historically Low
The quantity of political violence is historically even lower if you consider the potential amount of violence possible in modern America. The murder rate in politically motivated terrorist attacks is 0.01 per 100,000 in 2025, compared to a likely total murder rate of around 3 per 100,000 so far this year. Those are extraordinarily low rates when you consider the high potential for politically motivated killings in this country. The inputs for politically motivated violence have never been higher relative to the outputs. Most adults in the US can buy guns and ammunition without restriction and carry firearms in most places because the laws are less onerous than they were even a few years ago. Because of the frequency of firearms ownership, large numbers of teenagers and ineligible adults have easy access to them. Online bomb making instructions are ubiquitous. The determined killer can use the internet to plot against anyone they hate. There have never been more people and crazies than ever before who have free time. There’s such demand for politically motivated killings that demented commentators see political motives where there are none.
Government security and police forces have also never appeared more incompetent and ineffective, so there’s a better chance that the committed killer will get away with it. If you don’t believe the last point, just read how exposed Thomas Crooks was when he tried to kill President Trump, watch videos of ICE agents on raids, or glance at homicide clearance rates. Police fecklessness is my weakest point, but if the perception is that the police are bad at their jobs, then more people will try to commit politically motivated attacks, and we don’t see that.
Despite the reduced cost of committing politically motivated murder, there is still a tiny supply of politically motivated killers. What can we conclude about such a constrained supply of killers in the face of increasing demand for politically motivated murder, the Herostratusian rewards, and lower barriers to entry for killers? The supply is inelastic, practically immovable, and perhaps has even contracted in recent years relative to demand.
Conclusion
The objections to my work above are the main ones sent to me. Some have a kernel of truth that disappears after a moment of thought, and others are entirely off the mark. Why are there so many poor criticisms? Many people want to talk about how upset they are with the few killings that do occur, have another reason to complain about the other political party they dislike, or yell at the people who do not operate in a pure feelings-based political debate. Those reasons are perfectly understandable and entirely uninteresting to me. And although it’s easier to emote without knowledge, my job is to create and discuss knowledge.