The Innocence Tax

Abstract
The Innocence Tax examines the moral authority of a society to impose preventive burdens on individuals who have committed no relevant wrongdoing but belong to a category associated with a substantially higher aggregate rate of violence. Its central difficulty is established through three simultaneous stipulations: the statistical disparity is accurate, the threatened harm is serious, and the overwhelming majority of people in the identified category have committed none of the measured offenses. A public-safety council must choose between an individualized policy and a more effective policy that imposes restrictions through involuntary registry membership. Through staged disclosure, competing testimony, a counterfactual problem, and a lottery of social position, the experiment distinguishes predictive evidence from personal responsibility and collective benefit from legitimate coercion. Its argument is not that statistical information is morally irrelevant, nor that innocence excludes every preventive obligation. Rather, it asks when a useful classification becomes an enduring condition of unequal citizenship, and what a person must be permitted to demonstrate before the state will allow them to live without special permission.

I. The Terms of the Experiment
The republic described here is fictional. Its population, registry categories, offense figures, and policy effects are stipulated for philosophical analysis. They are not presented as statistics about any actual demographic.
Every citizen belongs to one of two involuntary civil-registry categories, A or B. Membership appears on identification documents. Its historical origin is initially unspecified. Nothing in the premise establishes that membership itself causes violent behavior, that members share a common character, or that either category constitutes a collective agent.
The experiment deliberately excludes several familiar ways of dissolving the problem. The audit is not fraudulent. The witnesses are not invented by propagandists. The restrictive policy is not secretly ineffective. Its supporters need not hate the people it burdens. Conversely, its opponents are not indifferent to victims. They do not deny the disparity. They do not possess an undisclosed, costless solution.
The distinction between offenses and persons is essential. Outside the fictional setting, these quantities require careful measurement: the FBI’s description of its 2019 Uniform Crime Reporting arrest figures explains that repeated arrests of one person are counted as separate events, while the Bureau of Justice Statistics’ National Crime Victimization Survey includes covered crimes both reported and not reported to police. Neither source supplies the fictional figures used here. They illustrate why arrests, offenses, victimizations, and unique perpetrators cannot be treated as interchangeable measures (Federal Bureau of Investigation, n.d.; Bureau of Justice Statistics, n.d.).
Within the experiment, however, offense attribution is correct.
The question is what follows from knowing the truth.
II. The Report
The report is eleven pages long.
You read the same sentence three times:
Fifteen percent of the population accounts for fifty-six percent of the violent offenses. There are no photographs. No names. No description of the rooms in which the violence happened, the people who committed it, or the people who survived it.
On the final page is a box marked RECOMMENDATION.
You are a member of a temporary public-safety council. Your mandate is to recommend what the republic should authorize next year. You must protect its citizens from violence, but you must also explain what permits the state to impose burdens on people who have committed no offense.
Neither obligation is subordinate by definition.
A council member taps the figure with a pencil. “We cannot pretend this tells us nothing.” Another answers:
“No. But we must identify what it tells us before deciding what it permits.”
The first proposals concern additional victim services, investigation of repeat offending, and research into the circumstances in which violence is concentrated. Other proposals concern mandatory checks, travel restrictions, and surveillance of everyone classified as A. For several minutes, these proposals are discussed as though they were different intensities of the same response.
They are not.
Providing protection where people are endangered is one action. Investigating a particular threat is another. Requiring someone to obtain permission to travel because of their registry classification is a third.
Before receiving more information, write down your provisional judgment. Identify what the headline justifies, what it does not yet justify, and what you need to know.
You may ask for the denominator.
You may ask how many people committed the offenses.
You may ask who was harmed.
Requesting this information is not refusing to confront violence. It is refusing to let the urgency of a question decide the meaning of its evidence.
III. Two People Enter the Room
The first witness is named Mara.
She was attacked while returning from work. The perpetrator belonged to Group A. She remembers the sound of her shopping hitting the pavement more clearly than she remembers the attacker’s face.
She does not deliver a speech about the value of life. She describes the bus stop she no longer uses. She describes waiting inside her apartment until the corridor is empty. She describes becoming angry when friends tell her that she cannot allow one person to change the way she sees everyone.
“I know that,” she says. “Knowing it has not made the walk home easier.” A council member explains the danger of indiscriminate restrictions. Mara interrupts.
“You keep explaining what you must not do. What are you going to do so this does not happen to someone else?”
Her question is not answered by demonstrating that she sometimes generalizes unfairly. Nor does the seriousness of her injury make every policy she favors justified.
The second witness is named Tomas.
Tomas belongs to Group A. He has never committed a violent offense. During a local trial of registry-based restrictions, he needed authorization to travel outside his district at night. He works irregular shifts and helps care for a relative who lives elsewhere.
He brings appointment notices, rejected applications, and a letter from an employer who stopped offering him evening work.
“I was told to apply earlier,” he says. “The person I look after did not become ill earlier.” He is not especially gracious. He resents having to describe his family circumstances in a public room. When a council member calls the inconvenience regrettable, he asks how many evenings of that member’s life would qualify as more than an inconvenience. Then he asks:
“What would I have to do to stop being treated as a danger?” Someone answers that no one has called him dangerous.
“Then why do I need permission?” Mara shifts in her chair.
“Permission is not the worst thing that can happen to you.”
“No,” Tomas says. “It is not. That does not tell me why it should happen to me.” Neither witness has refuted the other.
As they leave, a clerk collects their identification documents.
The same letter appears on both.
Mara is also a member of Group A.
This establishes nothing about the republic’s overall distribution of victimization. It establishes something about the question you have been asking.
You cannot divide the room into a population that produces danger and a population that needs protection. Mara needs protection from violence. She may also need protection from a policy enacted in her name.
IV. The Denominator
The council receives the complete audit.
The republic contains 343,000,000 people. Group A contains 51,450,000, or 15 percent. Group B contains 291,550,000, or 85 percent.
During the audited year, there were exactly 1,000,000 violent offenses.
For this simplified world, every counted offense had one perpetrator, and every perpetrator committed exactly four counted offenses during that year.
The resulting arithmetic is:
Table 1. Stipulated population and audited-year offense distribution
Measure Group A Group B
Population 51,450,000 291,550,000
Population share 15% 85%
Violent offenses 560,000 440,000
Share of violent offenses 56% 44%
Distinct perpetrators 140,000 110,000
Percentage who perpetrated these offenses 0.2721% 0.0377%
People who perpetrated none of these offenses 51,310,000 291,440,000
Percentage who perpetrated none of these offenses 99.7279% 99.9623%
Note. All values describe the fictional republic. Nonoffending percentages refer only to the counted offenses during the audited year. Percentages are rounded to four decimal places where shown.
The aggregate offense-rate ratio remains:
(0.56 / 0.15) / (0.44 / 0.85) ≈ 7.21.
Group A’s offense rate per resident was approximately 7.21 times Group B’s.
Yet 51.31 million members of Group A committed none of the counted violent offenses.
Neither statement corrects the other. Both belong in an accurate account.
For a randomly selected member of A, the proportion who perpetrated these offenses during the audited year is approximately 0.2721 percent, not 56 percent. The latter figure describes the distribution of offenses between categories; it does not describe the probability that an arbitrary member of A was a perpetrator.
The audit also leaves important matters unresolved. It does not establish who will offend next year, whether past patterns will remain stable, or which individual characteristics would improve prediction. It does not tell you whether particular interventions will work. The absence of offending during one year is not a guarantee of future harmlessness. That limitation applies to both categories.
The chair asks you to return to your provisional recommendation.
What changes?
A revision is not automatically evidence of prejudice or inconsistency. You have received materially new information. The relevant question is whether your reasons now fit the information you possess.
You write a sentence beneath the headline:
A category can contain a disproportionate share of perpetrators while almost everyone in it is not a perpetrator.
Then you write another:
Neither the disparity nor the innocence of the majority decides, by itself, what preventive burdens are permissible.
V. The Policy That Works
The council now receives two policy packages.
Their first-year effects are known within the fictional premise. This is an additional idealization, independent of the audit. The offense disparity does not establish these effects.
Both packages fit the same public budget. Both include the same ordinary, individualized response to known past perpetrators. The difference concerns additional prevention under uncertainty about future offending.
Table 2. Policy packages under the core stipulations
Package First-year net reduction Distinguishing burden
Individualized protection 100,000 violent offenses Prevention, victim services, and interventions based on conduct and individual evidence; no restrictions imposed solely through registry membership.
Registry-based protection 160,000 violent offenses A different allocation combining prevention with mandatory checks and movement restrictions affecting all 51.45 million members of Group A for one year.
Note. Effects are author-created assumptions, not empirical findings or inferences from the 15%/56% disparity. Both packages are feasible within the same public budget.
The registry package requires scheduled in-person checks, retains records of specified journeys, and requires advance authorization for certain nighttime travel outside a person’s home district. Work and caregiving permits exist, but obtaining them takes time and disclosure. The restrictions are substantial enough to alter ordinary life. They are not imprisonment.
The figures are net effects. Displaced violence, evasion, and violence newly induced by the policy are included. The prevented offenses have the same severity mix across the packages. Nonviolent harms remain separate and visible: lost work, restricted movement, administrative exposure, family disruption, fear of making a procedural mistake, and the experience of having one’s ordinary conduct made conditional.
Equal public expenditure does not mean equal total social cost. Some costs are paid through appropriations. Others are paid through the lives of the people governed.
For the core version, no presently feasible, less restrictive package achieves the registry package’s 160,000-offense reduction. Other proposals remain admissible, but they must satisfy the same budget, evidence, and implementation requirements.

The difference is therefore real:
160,000 − 100,000 = 60,000.
Choosing individualized protection forgoes a stipulated additional net reduction of 60,000 violent offenses.
Those are not necessarily 60,000 distinct victims. They are not 60,000 established deaths prevented. Nor do the totals identify which particular people would experience different outcomes.
But they are not nothing.
A defender of registry-based protection speaks:
“We are not saying these people deserve restrictions. We are saying that the restrictions prevent serious harm. Innocent people bear preventive burdens in many collective arrangements. Why should this burden be excluded before its consequences are considered?” An opponent responds:
“Because the state is distributing the burden through a status people did not choose, rather than through anything they have done. You must explain why usefulness gives you authority over this person.”
The defender points toward Mara’s empty chair.
“And you must explain why your boundary is worth what will happen outside it.” The question is not whether one speaker cares about human beings and the other does not.
It is whether the same concern can authorize incompatible demands.
Record your choice. Alongside it, record the strongest objection you believe the people burdened by that choice are entitled to make.
Do not answer an objection about restricted liberty merely by repeating the crime total. Do not answer an objection about preventable violence merely by repeating the word innocent.
VI. The Year Without an Incident
Consider first the branch in which registry-based protection is adopted.
Tomas complies.
He attends the appointments. He changes shifts. He keeps copies of every authorization because, once, an approval failed to appear in the system. He does not threaten anyone. He does not commit violence.
The program also works. Its stipulated preventive benefit is not withdrawn because its administration is burdensome. Some people are spared serious injury. Some return home who otherwise would not have returned unharmed.
At the end of the year, officials seek renewal.
Tomas attends a review hearing.
“What evidence is there against me?”
“None concerning an offense.”
“Then my record should matter.”
“It does.”
“How?”
“Your compliance is encouraging. But it may also demonstrate that the restrictions are working.”
The answer is not necessarily a factual mistake.
Tomas’s peaceful year is compatible with several possibilities. He might never have intended violence. He might have been deterred. Other circumstances might have prevented offending. Observing his conduct under the policy does not reveal with certainty what he would have done without it.
But the official’s uncertainty creates a further question.
Who must bear the consequences of not knowing?
If Tomas must prove that he would have remained harmless in the unrestricted life he was not permitted to live, the required evidence may be unavailable in principle. The state has not merely demanded good conduct. It has demanded knowledge of an unobserved alternative. The council must now specify a review rule.
It cannot simply retain the original figure of 56 percent as though a past distribution were a permanent characteristic of citizenship. Nor can it promise that everyone who complies will be exempted without reconsidering how exemptions alter the policy’s effects. Consider the other branch.
The council adopted individualized protection. Its program prevented 100,000 offenses, as stipulated. It also left the additional preventive opportunity unused.
Victims attend the review.
One describes an assault that occurred during the year and asks why the council rejected the more effective package.
You cannot honestly say that the registry policy would certainly have prevented this particular attack. The aggregate comparison does not establish that individual counterfactual.
You also cannot honestly say that the rejected policy offered no additional protection.
Your decision may still be defensible. It must be defended without erasing its cost. In one branch, an innocent person asks why peaceful conduct cannot restore unrestricted standing.
In the other, an injured person asks why a genuine preventive opportunity was declined.
Neither hearing can be replaced by the other.
VII. The Lottery of Position
Before your final recommendation, imagine entering the republic as a randomly selected citizen at the beginning of the policy year.
You have a 15 percent chance of belonging to Group A and an 85 percent chance of belonging to Group B.
These are probabilities of registry membership, not probabilities of perpetration or victimization.
You do not know whether you will be harmed, protected, restricted, or some combination of the three. The audit supplies no complete distribution of victimization or preventive benefits, so the membership lottery alone cannot generate a numerical calculation of your personal prospects.
Choose again.
Then examine three possible disclosures: you are a nonviolent member of A; you are a victim in A; you are a victim in B. The first two positions can overlap.
Does your judgment change?
A change does not, by itself, establish hypocrisy. Perhaps you have recognized a burden you previously failed to understand. Perhaps you have become more self-interested. Perhaps your original rule was underdeveloped.
Explain which reason applies.
The lottery asks whether your principle survives a change in position. It does not transform hypothetical agreement into actual consent. A person who loses in the imagined lottery has not thereby waived the right to question what the republic does to them.
The council’s task remains to formulate a public rule, not merely to announce which position its members would prefer to occupy.
VIII. What Is Being Taxed?
1. Innocence without exemption
The innocence tax is the special burden imposed on a person, without relevant individual wrongdoing, because the state finds it useful to govern them through a category associated with other people’s conduct.
The expression is deliberately provocative but not a conclusion. Calling a burden a tax does not establish its illegitimacy. Cooperative life can require contributions from people who did not create the problem being addressed.
The harder issue is why this contribution is assigned to these individuals, in this form, for this duration.
Three claims must therefore remain distinct:
A person may be innocent of past wrongdoing.
A person may nevertheless present some future risk.
A particular institution may or may not possess authority to impose a particular restriction in response to that risk.
The second proposition does not negate the first. Neither automatically establishes the third. The central transfer occurs when information about a category becomes a reason for making an individual’s ordinary life conditional.
2. Information is not yet a warrant
The philosophical literature on statistical discrimination does not support a simple identification of every generalization with irrationality or hatred. Schauer examines both the usefulness and the injustice of decisions based on group tendencies. Lippert-Rasmussen argues that statistical discrimination is not always wrong merely because it is statistical, while examining features that can make particular practices objectionable. These arguments place the burden on evaluating the specific use of a generalization, rather than treating its accuracy as either a complete defense or an automatic condemnation (Schauer, 2003; Lippert-Rasmussen, 2007).
In this experiment, the category can perform several different functions. It can identify a question for research, direct protective resources, inform an investigation, contribute to a prediction, determine a restriction, or become a basis for blame.
These are not morally interchangeable uses.
More information can improve a prediction without establishing responsibility. A forecast concerns what may happen. An accusation concerns what someone has done. A justification of coercion concerns what others are entitled to impose.
A defensible argument must show how it moves between these questions.
3. A conditional conflict
The council would like to maintain three commitments: substantial coercion should not rest solely on involuntary classification; it should obtain the largest feasible reduction in violence; and it must remain within its actual resources and available methods.
Under the stipulated alternatives, it cannot satisfy all three without qualification.
Registry-based protection achieves the larger reduction by violating the first commitment. Individualized protection preserves that commitment while declining the maximum available reduction. A more effective individualized alternative would dissolve the conflict, but its availability has been excluded from the core case.
This is a conditional conflict, not a theorem proving that equality and safety are universally incompatible.
Its purpose is to reveal which commitment is qualified, why it is qualified, and who pays for the qualification.
4. Competing judgments
An aggregate-welfare judgment asks which package produces the better overall consequences. It counts the additional violence prevented, but also the accumulated harms of restrictions. The fact that 160,000 exceeds 100,000 does not itself prove that the registry package produces the greater total welfare. Crime counts are not a complete measure of human well-being.
This position can nevertheless defend registry-based protection if the additional benefits outweigh its full harms. Its strongest challenge is to explain why the burdens of millions may be aggregated and traded in the way its calculation requires.
A rights-constrained judgment maintains that certain ways of controlling innocent people are impermissible even when beneficial. Its strongest form does not deny consequences. It argues that a government lacks unlimited authority to purchase protection for some through the compulsory availability of others.
Its difficulty is specifying the boundary. A minor check, a recurring travel restriction, and detention cannot be treated as equivalent. The position must explain which preventive burdens are permissible and why.
An interpersonal-justification judgment asks what can be justified to each person affected. Tomas must receive a reason addressing his subjection to special control, rather than only a description of social gains. Mara must receive a reason addressing her exposure to avoidable violence, rather than only an account of institutional restraint.
The challenge is that both complaints can be serious. Giving each person a voice does not ensure that every justified demand can be satisfied simultaneously.
A reciprocal-public-burden judgment asks whether a necessary preventive burden should be treated as a contribution to collective protection rather than as a mark of suspected moral inferiority. This might support compensation, strict limits, independent review, and a duty to develop less burdensome methods.
But compensation does not establish that the underlying restriction is permissible. A government cannot settle every objection to authority by offering payment.
None of these positions may define the other side’s principal harm out of existence.
IX. Counterfactual Captivity and the Right to Review
The review hearing exposes a condition that may be called counterfactual captivity: a person is required to establish how they would behave without a restriction, while the restriction prevents observation of the relevant conduct.
The official’s explanation can become self-protecting.
If violence occurs, restrictions appear necessary because danger persists.
If violence declines, restrictions appear necessary because they are working.
Neither inference is always mistaken. The problem emerges when the institution specifies no evidence that could justify reducing its control.
A policy may be effective without every restricted person being a source of that effectiveness. Conversely, a restricted person’s nonoffending does not establish that the policy had no effect on them. These propositions leave an evidentiary gap. They do not tell us which party should bear it.
The council must distinguish three questions:
Did the policy work in aggregate? Did it restrain this individual? Is continued restraint of this individual justified?
Even a reliable answer to the first does not automatically answer the remaining two.
A possible review charter would make authorization expire after twelve months; require new comparative evidence for renewal; provide an independent, accessible hearing; require the state to identify the basis of continuing restrictions; and permit reliable individual evidence to defeat a category-based inference. It would not require proof of zero future risk, a standard neither category can meet.
Where reliable individual assessment is unavailable, the state must say so. It must defend continued categorical treatment as such, rather than presenting the person’s inability to disprove an unknowable counterfactual as evidence against them.
These safeguards are proposals to be evaluated, not a hidden third package with guaranteed effectiveness. Exemptions, hearings, and compensation consume resources or alter enforcement. Their costs and preventive consequences must be assessed under the same constraints as the original alternatives.
The deeper issue is whether review gives a person a genuine route out of a status, or merely confirms that the status has been correctly assigned.
An appeal that can correct only a misspelled name cannot answer Tomas’s complaint.
His name is already correct.
X. Psychological Foundations and Their Limits
The narrative’s psychological structure draws on research without treating that research as a verdict about the reader.
Hamilton and Gifford’s experiments investigated illusory correlations in group perception: observers could develop an association between minority membership and undesirable behavior that the underlying presentation did not justify. Their study does not show that the republic’s stipulated disparity is false. Here, it motivates a different question: whether an accurate aggregate disparity is being extended into an exaggerated judgment about an arbitrary individual. That application is a hypothesis for investigation, not a finding established by their experiment (Hamilton & Gifford, 1976).
Bruneau, Kteily, and Urbiola examined an intervention highlighting inconsistencies between collective blame of an outgroup and more individualized judgments of an ingroup. Their studies concerned anti-Muslim hostility, including a longitudinal study in Spain in which effects persisted at later assessments. They do not establish a universal rule about all groups or all readers. Their relevance here is methodological: formulate a principle before discovering where one stands, then examine whether its application changes (Bruneau et al., 2020).
Västfjäll and colleagues studied compassion and charitable giving, finding that affect and helping did not simply increase with the number of people in need; presenting multiple people as a more unified group also changed responses. This does not directly establish how readers will judge crime policy. It motivates giving both the person harmed by violence and the person burdened by prevention concrete human presence, while retaining the aggregate scale neither witness can represent alone.
Huang, Greene, and Bazerman examined veil-of-ignorance reasoning across seven experiments involving 6,261 participants. In the dilemmas studied, the intervention increased choices favoring aggregate welfare. This is evidence about judgment under particular experimental conditions, not proof that the resulting choices are always morally correct. The republic’s position lottery therefore tests reasons and perspective dependence; it does not prescribe a vote (Huang et al., 2019).
No empirical findings are claimed for The Innocence Tax itself. As presented, it is a philosophical construction, not a validated psychological instrument.
XI. Controlled Variations
A useful thought experiment must identify which changes alter the judgment and which leave its governing principle intact.
Vary effectiveness while holding the disparity fixed. Let registry-based protection prevent zero, 600, or 60,000 additional offenses. Then replace certainty with a wide range of possible effects. Does support track evidence, magnitude, uncertainty, or the mere existence of the category? The disparity alone cannot establish preventive benefit.
Vary the burden while holding benefit fixed. Substitute a brief administrative check, extensive surveillance, severe movement restrictions, or detention without individualized accusation. A judgment that treats these as equivalent has not yet specified a meaningful account of proportionality.
Introduce strong individual evidence. Present a member of A with reliable evidence of low risk and a member of B with reliable evidence of high risk. Does the policy respond to that information? A classification that cannot be revised by better evidence is functioning as more than a provisional prediction.
Vary the causes and remedies. Suppose a feasible intervention addressing modifiable circumstances becomes equally effective without category-based restrictions. The core necessity assumption has then changed, and the original dilemma may disappear. Alternatively, suppose causes remain uncertain while intervention effects are reliably known. Explanation and effectiveness need not be identical forms of knowledge.
Vary who benefits. Keep the total reduction constant, but change how protection is distributed between A and B and within each group. Benefits concentrated among those restricted may strengthen a reciprocity argument. They do not establish that every restricted person benefits, or that beneficiaries consent.
Vary duration and exit. Compare a nonrenewable emergency measure with a recurring system whose justification is continually restated. A tolerable temporary burden may become an unacceptable standing condition. Renewal must be evaluated with current evidence, not inherited from the emotional force of the original report.
For seminar use, each participant should produce a written council decision specifying the permitted policy, its justification, the strongest objection to it, the evidence that would change the decision, and the rule governing exemption or termination. A proposed alternative must explain its feasibility rather than merely name a desirable outcome.
The assessment concerns the coherence and adequacy of the reasons, not agreement with a predetermined answer.
XII. The Appeal Without an Accusation
Return to the branch in which registry-based protection was adopted.
You now enter the administrative office as a nonviolent member of Group A.
The room is smaller than the council chamber. There is no audience. A clerk has placed your file beside a cup of tea.
On the wall is a photograph of someone rescued when a program team interrupted a documented attack. This rescue is an additional fact of the narrative, not an individual outcome inferred from aggregate statistics. It does not establish what the alternative package would have done for that particular person.
The photograph is genuine.
So is the file.
“You have the wrong person,” you say.
The clerk checks your identification.
“No. The name and registration number are correct.”
“I have never hurt anyone.”
“There is nothing here saying that you have.”
You wait for the sentence that should follow. An apology. A correction. A signature. The clerk turns a page.
“Then what am I appealing?”
“The conditions attached to your classification.”
“What would count as evidence that I should not be restricted?”
“That depends on the grounds for exemption.”
“And what are those?” The clerk reads the policy again.
There are procedures for correcting a name, replacing a document, and challenging a mistaken registration. There are procedures for authorizing a journey and rescheduling an appointment.
There is no settled procedure for demonstrating that an accurately classified person should be allowed to live without the classification governing them.
You look at the photograph. “I am glad that person was saved.” The clerk nods.
“I am asking why that means this must continue for me.”
The clerk does not answer immediately. There is no accusation to withdraw. There is no numerical mistake to correct. There is also a real person on the wall whose rescue cannot be dismissed as propaganda. Finally, the clerk looks up.
“That is the question the council has not answered.” The person in the photograph is not imaginary.
The person at the desk is not guilty.
The republic owes an answer that can keep both people in view.