According to Claude, terrorist reoffending is rare. The Terrorism Recidivism Study tracked US terrorism convictions post-9/11 and found that of 354 released offenders, about 3.1 percent reoffended, well below rates for ordinary criminal offenders. European practitioners put the global median for violent extremist and terrorist offenders near 5.5 percent, and a Belgian dataset of more than 500 jihadi convicts from 1990 to 2019 found the same low rate. So a program that takes eighty men and reports three failures has produced what leaving them alone produces. Almost every Western deradicalization program is too small for its own success rate to mean anything. You cannot measure a treatment effect on a rare event with a caseload that size, and the providers know it, which is why the reporting runs to testimonials and throughput counts.
The evaluation literature says the same thing in drier language. The Campbell evidence and gap map found the research piled up on policing interventions, few studies on custodial corrections, and none at all on community corrections. The Campbell review of case management concluded that effectiveness remains poorly understood because robust impact evaluations do not exist. A review of prevention programs screened close to twelve thousand documents and found 33 empirical evaluations, hampered by weak designs, small and biased samples, unclear definitions, and conflicts of interest. That same review found harmful effects clustering in programs that targeted particular ethnic or religious groups or leaned on surveillance. That is Prevent in a sentence.
The Tom Wolfe (1931-2018) reading fits the funding structure. Mau-mauing runs on an intermediary whose standing depends on his ability to represent a threat to the men holding the checkbook. The credible-messenger model builds that in by design. The former gang member’s qualification is his old rank in the thing he is now paid to dissolve. Success retires him. Nobody has to be cynical for the incentives to favor permanence, ambiguity in the outcome measure, and a threat that stays vivid.
Focused deterrence has evidence. Braga, Weisburd and Turchan pooled 24 quasi-experimental evaluations and found a statistically significant, moderate crime reduction effect. The engine there is the call-in, the credible promise of group liability, and a police department that follows through. Services ride along. Take away the deterrence half and nobody has shown the remainder does anything.
The second thing that works is cheap and has nothing to do with ideology. In Liberia, 999 high-risk men were randomized to eight weeks of therapy run by nonspecialists, two hundred dollars in cash, both, or neither. A decade later the men who got therapy, or therapy with cash, were about half as likely to be robbing, selling drugs, or fighting in the street, and the effects were concentrated in the highest-risk men. The result exceeded what expert forecasters predicted. Sara Heller’s Becoming a Man trial in Chicago points the same way. Self-regulation and a new self-description are teachable. Theology is not.
Disengagement might work better than deradicalization. Behavior against belief. Men leave violent movements for reasons that embarrass everyone: they get tired, they fall out with a comrade, they meet a woman, they have a child, they turn thirty. A program that lowers the cost of walking away and hands a man a story he can tell about why he left might help. A program that promises to correct his beliefs claims a power over conviction nobody has demonstrated.
Disengagement means a man stops. He quits the cell, the crew, the set. He stops carrying, stops attending, stops answering the phone. Deradicalization means he stops believing. He renounces the doctrine that justified the violence and adopts something else. Researchers separated these two ideas thirty years ago. Funders never did. Grant language, program names, and press releases keep promising the second thing while the world can only observe the first.
The promise survives because nobody can check it. Terrorist reoffending is rare. The Terrorism Recidivism Study followed post-9/11 American convictions and found that of 354 released offenders, roughly 3.1 percent reoffended, far under rates for ordinary criminals. European practitioners put the global median for violent extremist offenders near 5.5 percent. A Belgian dataset of more than five hundred jihadi convictions between 1990 and 2019 produced the same low figure. Run a program on eighty men, report two failures, and you have matched what neglect produces. No caseload in Europe or North America is large enough for its own success rate to carry information. The event is too rare and the samples too small. A provider who did nothing but hand out sandwiches could publish the same numbers.
The reviews confirm the vacuum. The Campbell evidence and gap map found the evaluation research bunched around policing, thin on custodial corrections, and empty on community corrections. The Campbell review of case management concluded that effectiveness stays poorly understood because robust impact evaluations do not exist. A review of prevention programs screened close to twelve thousand documents, kept thirty-three, and reported weak designs, small and biased samples, unclear definitions, and conflicts of interest. The same review found harmful effects clustered in programs that targeted particular ethnic or religious groups or leaned on surveillance.
So the field cannot tell you whether belief changed, and cannot tell you whether anything changed. What it can tell you, from a much older literature, is how men leave violence when nobody pays them to.
They age out. Adolphe Quetelet (1796-1874) noticed the curve in the 1830s and it has not moved since: offending rises through adolescence, peaks around nineteen or twenty, and falls for the rest of a man’s life. Sheldon Glueck (1896-1980) and Eleanor Glueck (1898-1972) followed five hundred Boston delinquents into middle age, and when Robert Sampson and John Laub reopened the files half a century later they found the turning points were mundane. A steady job. A wife. The Army. Mark Warr showed part of the wife effect runs through the calendar; a married man spends his evenings with a woman instead of his friends, and street crime is a group activity.
The terrorism interviews land in the same place. Tore Bjørgo’s Scandinavian right-wing extremists left because they got tired, because they feared prison, because their girlfriends threatened to leave, because the movement’s promises did not arrive. John Horgan’s interviews with Irish republicans and jihadists produced a list any human resources department would recognize: disillusionment with leadership, burnout, unmet expectations, small humiliations, family pressure. Altier, Thoroughgood and Horgan borrowed the turnover literature on purpose, because leaving a terrorist group looks like leaving a bad employer.
Notice what is missing from those accounts. Almost nobody says he changed his mind and therefore left. Men leave and keep their opinions for decades. Other men lose their faith in the doctrine and stay for the friends. Travis Hirschi (1935-2017) put belief last among his four social bonds, behind attachment, commitment, and involvement, and the ordering has held up. David Matza (1930-2018) described drift, and drift runs both ways. Conviction rides along with the life. It rarely drives.
That gives you the design principle. Belief follows the exit. Buy the exit.
The two things work on price and habit. Focused deterrence is the strongest result in this area: Braga, Weisburd and Turchan pooled twenty-four quasi-experimental evaluations and found a statistically significant, moderate crime reduction effect. The engine is the call-in, group liability, and a police department that follows through, with services on offer for the men who want out. It raises the price of the next shooting and lowers the price of quitting, and it says nothing about anyone’s soul. The second result comes from Liberia, where 999 high-risk men were randomized into eight weeks of therapy run by nonspecialists, two hundred dollars in cash, both, or nothing. Ten years later the men who got therapy, or therapy with cash, were about half as likely to be robbing, selling drugs, or fighting in the street. The gains concentrated in the highest-risk men and beat what expert forecasters predicted. Sara Heller’s Becoming a Man trials in Chicago point the same direction. What those programs teach is self-regulation and a new way to describe yourself. Doctrine never comes up.
Now the incentives.
Peter Rossi (1921-2006) gave us the iron law of evaluation: the expected net impact of any large-scale social program is zero. He also gave us the corollary that explains why programs survive anyway. The better the evaluation design, the smaller the measured effect. Providers learn this. They do not need to be dishonest to learn it. They need only notice which studies get them refunded.
Donald T. Campbell (1916-1996) supplied the second half. Any quantitative indicator used for social decision-making comes under pressure to corrupt, and corrupts the process it monitors. Charles Goodhart (b. 1936) said it for economists. In deradicalization the indicator is contacts, referrals, sessions delivered, mentors trained, youth reached. Those numbers rise with effort and never fall with failure. A program paid for contacts will produce contacts, and the men most likely to generate contacts are the men easiest to work with, who are also the men least likely to have offended in the first place.
George Akerlof (b. 1940) supplies the third. When buyers cannot observe quality, price stops tracking quality and the market selects on what buyers can observe. Here the buyer is a home office or a city council, the product is a change in a man’s future behavior, and that change is unobservable at the scale any single program operates. So the market selects on the observable substitute: a good story, a charismatic founder, a survivor’s testimony, a documentary crew. This produces exactly the industry we have. The best storytellers win the contracts. Efficacy floats free.
Tom Wolfe (1931-2018) described the last piece in 1970 and the structure has not changed. Money moving toward a threatened population moves through an intermediary whose standing depends on his ability to represent the threat. The credible-messenger model builds this in as a feature. The former gang member’s qualification is his old rank in the thing he is now paid to dissolve. Peace retires him. The flak catcher on the other side of the table wants the meeting to end without an incident and has no way to audit anything. Neither man has to be corrupt. The incentives point toward permanence, vividness, and an outcome measure nobody can fail.
There is a fourth cost the field rarely counts. Howard S. Becker (1928-2023) argued that deviance is conferred, not discovered, and that the label stabilizes the career. A referral program that names a seventeen-year-old an extremist has told him who he is, told his school, and told him which adults now find him fascinating. The prevention review’s finding of harm in programs targeting religious and ethnic categories reads as labeling theory confirmed in the field. Before you can deradicalize a boy you have to certify that he is radical, and the certificate has value in some markets.
An incentive-compatible design falls out of all this.
Pay for exit, and price it like any other transition. Tattoo removal, relocation, protection from the men he is leaving, a driver’s license, the fines and warrants cleared, work that starts Monday, a story he can tell his mother and his old friends that does not require him to grovel. Every one of these lowers the cost of walking away, and every one of them can be counted.
Never require recantation. A man who must denounce his beliefs to receive help will denounce them, and the reward goes to the fluent liar over the exhausted true believer who is actually leaving. Recantation is cheap talk with a public relations value to the program and a negative value to the participant, since it burns his bridges back to the community he still lives in.
Measure behavior with records collected by someone with no stake in the result. Arrests, hospital admissions, employment, returns to custody. Self-report scales administered by the staff whose funding depends on the score are theater.
Randomize, because these programs are oversubscribed anyway. A waitlist lottery costs nothing, treats applicants fairly, and turns twenty years of anecdote into a finding. A provider who refuses a lottery on a program with more applicants than slots is telling you what he expects the answer to be.
Target behavior, never a category. Men who have shot at someone, or been shot at, are a defensible population. Muslim teenagers, or working-class White teenagers, are not, and the evidence says targeting them that way does harm.
Sunset the contracts. Any body funded to manage a threat will discover the threat is chronic. Write the expiry into the grant and make renewal contingent on outcomes measured elsewhere.
The general lesson runs past terrorism and gangs. Programs that promise to change what people believe are funded by principals who cannot observe belief, which guarantees payment for performance of belief. That describes drug education, scared-straight tours, most workplace training, and a good deal of what passes for prevention. Programs that change prices, opportunities, and daily company can be measured, can fail visibly, and sometimes work. Gary Becker (1930-2014) was closer to the truth about crime than any curriculum designer, and he was writing about arithmetic a man does without knowing he is doing it.
We cannot buy a man’s convictions. We can make the door out of his old life cheap, quick, and unhumiliating, and we can make the door back in expensive. Do that at scale and the beliefs sort themselves out over ten years, quietly, in the direction the beliefs of aging men always drift. Nobody gets to hold a press conference about a soul reclaimed. That is the cost of a program that works.
The Atomistic Extremist
Take John J. Mearsheimer (b. 1947) at his word and the deradicalization industry looks like a category error running on schedule.
His claim is about human beings before it is about states. We are social from birth to death. Individualism comes second. A man is born into a group that shapes his identity before he can assert one, and by the time his reasoning has matured his family and his society have already loaded him with an enormous value infusion. Reason ranks third among the forces that fix his preferences, behind inborn sentiment and behind socialization. He has limited choice in formulating his moral code.
Set that anthropology beside a deradicalization program and the mismatch is total. The program addresses a man alone in a room. It offers him arguments: a mentor with better theology, a curriculum on democratic values, a survivor’s testimony, a counter-narrative video. It asks him to reconsider. It treats him as the bearer of a mistaken opinion, correctable through discussion, because that is what political liberalism takes a person to be. Mearsheimer’s man is not available for that conversation. His convictions arrived through a group, they are held on behalf of a group, and they will change when the group he lives among changes, which no forty-minute session accomplishes.
This predicts the pattern in the evidence. The Campbell review of case management approaches concluded that their effectiveness remains poorly understood because robust impact evaluations have not been done. A review of prevention programs kept thirty-three studies out of nearly twelve thousand documents and reported weak designs, small and biased samples, unclear definitions, and conflicts of interest. The same review found harm clustering in programs aimed at particular ethnic or religious groups and in programs built around surveillance. On Mearsheimer’s account that harm has an obvious source. A program that selects a boy by his religion has told him and everyone around him which group he belongs to, and has done so at the moment he is deciding how much that membership costs him. The state confirms the extremist’s central claim about the world.
The two things with evidence behind them work on the group, not the argument.
Focused deterrence is the strongest result in this area. Braga, Weisburd and Turchan pooled twenty-four quasi-experimental evaluations and found a statistically significant, moderate crime reduction effect. Look at what the call-in does. It brings a gang into a room as a gang, tells its members that the next shooting brings sanction on all of them, and puts mothers and grandmothers and clergy in front of them to say the same thing. It assigns collective liability and it recruits the men’s own social ties as enforcers. Nobody is asked to change his mind about anything. The strategy assumes Mearsheimer’s man: embedded, loyal, responsive to the standing of his people, willing to sacrifice for fellow members and therefore reachable through them.
Office of Justice Programs
The Liberia trial reads the same way. Nine hundred ninety-nine high-risk men were randomized into eight weeks of therapy, two hundred dollars, both, or nothing, and a decade later the therapy arms were about half as likely to be robbing, selling drugs, or fighting in the street. The effects concentrated in the highest-risk men and beat expert forecasts. The sessions ran in groups of about twenty, with men practicing a different self-presentation together: new clothes, new haircuts, new places to sit. The cash bought the material footing to hold the new position. What the men received was a small society with different membership rules and the means to stay in it.
Now the exits, which have always been social. The interview literature on men leaving violent groups turns up exhaustion, a falling-out with a leader, fear of prison, a woman, a child, a birthday. Almost nobody reports leaving because he was refuted. That looks like weak conviction until you read it through the frame. He did not exit into individualism. He exited into a different group with a competing claim on him: a wife, a household, a job crew, a congregation, an aging body that made him useless to the old crew and welcome somewhere else. The man never stopped being embedded. He changed which embedding governed his week.
Which means the design question is not what to say to him. It is what group will have him, on what terms, and how fast.
The lever with the best record is collective. Saudi Arabia’s program routes releases through family guarantors and tribal surety; a relative signs for the man and bears standing risk for his conduct. Denmark’s municipal approach works through parents, siblings, employers, and a caseworker who knows the household. Focused deterrence puts the group on notice as a group. All three take the family or the clan or the crew as the unit of accountability, because that is the unit that governs behavior.
A liberal state cannot say this out loud. Mearsheimer’s argument is that political liberalism is individualist at its core and organized around inalienable rights held by each person as a person. Collective liability offends that premise at the root. You cannot formally punish a brother for a brother, or lean on a mosque for a congregant, or make a father’s employment contingent on his son, without abandoning the idea that rights attach to individuals and only to individuals. So the liberal state does the thing informally, deniably, through community engagement contracts and quiet pressure, and then cannot evaluate what it will not name. It also cannot empower the thick communities that would do the governing, because those communities run their internal affairs on lines the state finds intolerable: patriarchal authority, honor, shunning, arranged marriage, deference to men with no democratic mandate. The liberal state is being asked to subsidize illiberal social power because illiberal social power is what moves men. It does the safe thing instead. It funds curriculum.
The universalist reflex compounds this abroad. Western donors fund moderate-religion programming in Kenya, Nigeria, Indonesia and Pakistan on the theory that the right doctrine, taught to the right individuals, produces the right citizens anywhere, since everyone carries the same rights and therefore, presumably, the same latent liberal self. Mearsheimer spent a book explaining why that project fails between states. The domestic version is the same project on a caseload of one. Samuel Moyn (b. 1972) describes human rights as the elevated aspiration of the age, evoking hope and provoking action. Deradicalization is what that aspiration looks like when it is handed a budget and a list of names.
If men need thick belonging, then extremist movements are supplying something that fills men up. The cell offers what the recruiting sergeant and the parish and the union hall used to offer: a name, a rank, brothers, a code, a task worth dying for. A liberal society that has thinned out its own particular attachments in favor of universal ones has fewer competitors in that market. This does not excuse the movement. It explains the demand and locates the shortage. On this reading the second-generation immigrant who joins a foreign fight is not failing to become an individual. He is looking for a nation, and the one he lives in has offered him rights and a passport while withholding membership, and a movement has offered him membership on the spot.
The counter-argument deserves a hearing. If socialization runs as deep as Mearsheimer says, exit should be rare, and it is not. Terrorist reoffending in Western samples runs low. The Terrorism Recidivism Study found about 3.1 percent of 354 released American offenders reoffended. European practitioners report a median near 5.5 percent. Men walk away constantly, cheaply, and without ceremony. A strong tribalism thesis has to explain that. It can: the tie that broke was to a small, young, recently assembled group, and it lost to older and denser ties that were waiting. What it cannot explain is a man who leaves into nothing, and those are the cases that recur in the reoffending files.
So the program design falls out of the anthropology, and it looks nothing like the industry we built.
Do not argue with him. Do not require recantation; a man asked to denounce his people to receive help will either lie fluently or refuse, and the lie is worth more to the program than to him. Do not select him by category, because naming his group is how you strengthen its claim. Work through the ties he already has, and pay for the ones he needs: the wife, the child, the mother, the job that puts him beside three men who are not from his old crew, the housing that is not on the old street. Where the family is willing, make it the unit and say so. Where the state must offer belonging, it has to have some to offer, which means the shortage runs deeper than the counterterrorism budget.
And measure behavior only. Belief was installed by a society before the man could think, and it will be revised by a society after he stops paying attention.
