The deeper problem isn’t efficiency but incentive alignment. When punishment generates benefits for enforcers, the system begins optimizing for extraction rather than justice. That’s the mechanism behind civil forfeiture, colonial forced-labor sentences, and every historical case where “efficient” punishment quietly expanded the scope of criminalization.
A structural safeguard is simple: coercive institutions must never profit from the punishments they impose. Any cost imposed on an offender should flow to victims or neutral funds, never to the enforcers themselves. When enforcement is insulated from revenue, the incentive gradient that drives predatory behavior disappears.
In that sense, inefficiency is sometimes a feature, not a flaw. Friction in coercive systems prevents them from drifting into self-financing extraction. Efficient punishments are dangerous precisely because they lower the marginal cost of coercion.
A number of states set up civil forfeiture that way. Their law enforcement agencies, as I understand it, evaded the restriction by having the forfeiture adopted by federal law enforcement which then shared the money back with state or local law enforcement. I don't know if that is still happening.
You’re right about the workaround. Agencies found that even when states barred them from keeping forfeiture revenue, they could evade the restriction by routing the seizure through federal “adoption” and then receiving the money back through equitable-sharing. That’s exactly the kind of incentive-driven behavior your empirical work tends to highlight.
One thing I’ve noticed is that you and I usually approach these issues from opposite epistemic directions — you lean more toward empiricism, and I lean more toward rationalist, structural argument — yet we keep converging on the same conclusions. It’s a bit like predicting how many 7s will appear when rolling a pair of fair dice: one person derives it from mathematical probability, another confirms it through statistical sampling, and both methods land on the same answer. When our views align, it’s usually because the underlying structure forces the result.
As for how to keep the restriction from being evaded, the only practical solution is to eliminate the incentive at the architectural level rather than the statutory level. The rule has to be: no enforcement agency may receive any financial benefit from punitive actions — not directly, not indirectly, and not through federal pass-throughs. That means no revenue sharing, no grant offsets, no budget increases tied to forfeiture totals, and no federal adoption that returns money to the originating agency. All forfeiture proceeds would have to flow to victims, public defenders, or general funds that enforcement cannot access. Once punishment can’t finance the punisher, the incentive to find workarounds disappears.
And when we disagree, I usually take it as a cue to check whether my argument is unsound or uncogent, because your empirical instincts often reveal where a purely structural model needs friction added back in.
I think the right generalization here is that we can’t consider the efficiency of punishment in isolation from the incentives that creates for and around enforcement mechanisms, any more than we can make sense of optimal criminal or civil codes, without taking into account the costs and distortionary effects of detection and enforcement.
Agreed. Any analysis of punishment has to include the incentive effects on detection and enforcement. Treating efficiency in isolation misses the structural feedback loops.
Maybe a pedantic point, but in the US at least, execution is much more expensive than life imprisonment. And also delayed many years, reducing deterrence.
“You are driving somewhere far from home, carrying with you fifty thousand dollars in cash that you intend to use to make a down payment on a house.”
A good reason for holding stablecoin.
I agree with and accept all your arguments as good ones - *except* the one about the $50K cash.
Because I do think in today’s world it is *extremely* likely that amount of cash is for or as a result of illegal activity.
And so more “efficient” for the police to confiscate the cash rather than get their revenue through higher taxation of generally moral, law-abiding citizens.
You obviously don't interact with a broad section of society. Many, many people are paranoid about almost anything tech and therefore keep large amounts of cash and wouldn't dream of using any virtual currency
You are absolutely correct that I don’t interact with the parts of society who are that paranoid on the one hand *and also* have $50K they carry around with them *in their cars*. *And* drive through small towns far from home.
I’d love to hear your stories of such examples, however.
For the record, I’ve never owned bitcoin and wasn’t suggesting it as advisable or necessary for most honest people. It was merely a snarky joke.
I said no more and no less that such cases in the absence of criminality are rare.
Nick tried to claim that many people drive around with that much cash, and through unfamiliar small towns at that.
That was what I was calling him on, not a mere existence proof, which I did not doubt.
If - as I surmise likely - something like 99% of the time people are driving around with that much cash when they are stopped the money was for or resulted from some illegal activity, then I indeed might support the action on the “efficiency” grounds I noted above.
Do not mistake this for my endorsement of all the civil forfeiture techniques used by governments, in particular the more aggressive ones.
$10,000, which is the old $7500 before Biden’s inflation, I accept.
$50K is a *lot* more.
But if you in fact have “lots of” legit reasons that people *actually* drive with $50K+ through small towns far from home - not just theoretical ones - then just as I said to Nick, I’d love to hear them.
An even worse case than the civil forfeiture: When China started selling the organs and made some government officials jobs dependent on the trade, it led to the needless killing of perhaps several hundred thousand Falun Gong members and Tibetan Buddhists.
The cost to the criminal of execution is his life, and the cost to us his execution is the cost of the execution process. The execution process can be evaluated using basic accounting methods and given a dollar value. However, the cost of the criminal's life is not something bought or sold - it is not the cost of anyone's life as estimated using actuarial methods, since the criminal loses not a life in general but his own. So there is no dollarized cost to the criminal's life. This is not a problem of inaccurate approximation.
Consequently, the ratio of the cost to the criminal and the cost to "us", is not something that can be computed. Efficiency is not properly measured by ratios where a quantity can't be assigned to each side of the division.
What we can say about different forms of punishment is that people prefer punishments that are less harmful to their ends than others. We can approximate that death is at or near the least preferred punishment. As long as there is some 'bottom' to the list, so that a person does not distinguish between one really bad punishment and a worse one, we can say that there is no marginal deterrent effect of punishments after that one. This gives a good reason not to make all punishments equal to or nearly as bad as the death penalty, because they would not yield a gradient of deterrence for crimes that are gradiently costly to us.
"it is not the cost of anyone's life as estimated using actuarial methods, since the criminal loses not a life in general but his own. So there is no dollarized cost to the criminal's life. This is not a problem of inaccurate approximation." This misses the point. At the time when we pass our criminal laws, we do not know which of us will be (rightly or wrongly) executed in the future, so it's absolutely appropriate to treat the lost life as statistical.
Think of it this way: If you live in a world of 100 people, where each execution (via its deterrent value) saves you $1,000,000 in losses. But each execution also increases the probability that you yourself will be executed by 1%. To say that the value of a life is $10,000,000 is to say that you are willing to accept this tradeoff, and would in fact be willing to accept it even if the probability of your being executed were as much as 10%. That's the tradeoff you want, so it's the tradeoff that the law should give you.
In the initial setup, Friedman describes the cost used in the denominator as as the "cost to the criminal" and at the same time the "deterrence". The notion is that your willingness to do an action is inversely proportional to how much that action costs you. For the cost in the denominator to correctly index for the deterrence, it must be the specific cost to the criminal when they consider whether or not to commit a crime, because this is the cost which has a marginal effect on the choices of the criminal. If not, then the cost in the denominator does not index for deterrence, but something else. But then you're doing an analysis different from the one Friedman seems to have intended.
If we're considering the question at the time we set laws for punishment, we would include the statistical cost of life in the numerator, because this would express part of the cost to us of implementing that policy, i.e., the risk to our lives. That risk we take in passing a law is not the risk a criminal engages in when he commits a crime. The function of the ratio is to be able to express the factor of cost-to-us vs cost-to-the-criminal, with the notion that this ratio indexes for what we pay per unit of deterrence, i.e., the efficiency of a punishment measured by its deterrent effect. Putting the statistical cost of life in the denominator doesn't achieve that goal. So I don't take your point.
The cost to the criminal is part of the numerator. It also affects the denominator, but we have no idea *how* it affects the denominator. If the cost of a year in prison to the criminal is $1000, that's a $1000 that belongs in the numerator. If the prospect of that cost deters exactly 7 crimes, then the cost to society of those 7 crimes belongs in the denominator. But how do we know whether it's 7 crimes or 13 or zero? That's a purely empirical question that I don't see addressed.
David, I receive a 403 Forbidden error when I try to follow the Embedded Economics link. Seems to be the case for all of davidfriedman.com. Thought you might want to knowm
I absolutely fail to understand your first couple of paragraphs. You talk about the ratio of cost to deterrence for various forms of punishment. But the value of the deterrence is an empirical question (how many crimes does an execution prevent? how many crimes does a five-year jail term prevent?) Yet you cited no empirical evidence. So how on earth were you able to assign (approximate) numerical ratios to cost/deterrence when you've given us no indication of how you estimated the denominator?
It sort of works without evidence, but not well. That is, you don't need evidence to know that the cost to the state of imposing a fine is close to 0. So it's basically infinitely efficient, which isn't the case for execution, and certainly isn't the case for incarceration.
However, that's where the need for evidence arises. While fines are basically infinitely efficient, that doesn't mean that they achieve the optimal results. Consider a car with amazing fuel efficiency that can only drive 0.1 miles per day. That is, every day, it can drive 0.1 miles, using hardly any fuel, at all.
That would be remarkable, and more efficient than any alternatives, but not very helpful in getting to work.
As far as the empirics, I'll add that it's widely reported that execution costs more incarceration, due to the lengthy legal expenses associated with trials for the former.
Additionally, there's a distinction between deterrence and incapacitation, with the article perhaps conflating the two. The effects of incapacitation are large, and of course fines, efficient though they may be, don't offer that benefit.
[Now, he could argue that the fines can simply be arbitrarily high, so as to afford sufficient benefits the losses relative to other forms of punishment, but that didn't seem to be his point.]
Request - please turn substack text to speech option on. Until recently if I brought up one of your posts on the Substack app I would see the triangular play icon which if clicked would start reading the text aloud. Now it's gone for your posts, but it's still available and works for the posts of other authors. I'm not certain, but as I understand it, there is an option for Substack authors to turn this feature on or off, so I suspect that option for you is now set to off. If so, I respectfully ask for you to turn it back on.
The text-to-speech icon has now reappeared for this post when reading in the mobile app. I tested it and it works. Perhaps the option was only temporarily disabled during the server problems that Substack was experiencing yesterday.
The deeper problem isn’t efficiency but incentive alignment. When punishment generates benefits for enforcers, the system begins optimizing for extraction rather than justice. That’s the mechanism behind civil forfeiture, colonial forced-labor sentences, and every historical case where “efficient” punishment quietly expanded the scope of criminalization.
A structural safeguard is simple: coercive institutions must never profit from the punishments they impose. Any cost imposed on an offender should flow to victims or neutral funds, never to the enforcers themselves. When enforcement is insulated from revenue, the incentive gradient that drives predatory behavior disappears.
In that sense, inefficiency is sometimes a feature, not a flaw. Friction in coercive systems prevents them from drifting into self-financing extraction. Efficient punishments are dangerous precisely because they lower the marginal cost of coercion.
A number of states set up civil forfeiture that way. Their law enforcement agencies, as I understand it, evaded the restriction by having the forfeiture adopted by federal law enforcement which then shared the money back with state or local law enforcement. I don't know if that is still happening.
You’re right about the workaround. Agencies found that even when states barred them from keeping forfeiture revenue, they could evade the restriction by routing the seizure through federal “adoption” and then receiving the money back through equitable-sharing. That’s exactly the kind of incentive-driven behavior your empirical work tends to highlight.
One thing I’ve noticed is that you and I usually approach these issues from opposite epistemic directions — you lean more toward empiricism, and I lean more toward rationalist, structural argument — yet we keep converging on the same conclusions. It’s a bit like predicting how many 7s will appear when rolling a pair of fair dice: one person derives it from mathematical probability, another confirms it through statistical sampling, and both methods land on the same answer. When our views align, it’s usually because the underlying structure forces the result.
As for how to keep the restriction from being evaded, the only practical solution is to eliminate the incentive at the architectural level rather than the statutory level. The rule has to be: no enforcement agency may receive any financial benefit from punitive actions — not directly, not indirectly, and not through federal pass-throughs. That means no revenue sharing, no grant offsets, no budget increases tied to forfeiture totals, and no federal adoption that returns money to the originating agency. All forfeiture proceeds would have to flow to victims, public defenders, or general funds that enforcement cannot access. Once punishment can’t finance the punisher, the incentive to find workarounds disappears.
And when we disagree, I usually take it as a cue to check whether my argument is unsound or uncogent, because your empirical instincts often reveal where a purely structural model needs friction added back in.
I think the right generalization here is that we can’t consider the efficiency of punishment in isolation from the incentives that creates for and around enforcement mechanisms, any more than we can make sense of optimal criminal or civil codes, without taking into account the costs and distortionary effects of detection and enforcement.
Agreed. Any analysis of punishment has to include the incentive effects on detection and enforcement. Treating efficiency in isolation misses the structural feedback loops.
It is not exactly ideal but at least the Colonial powers got the road built!
Maybe a pedantic point, but in the US at least, execution is much more expensive than life imprisonment. And also delayed many years, reducing deterrence.
There's a colorable argument that execution is made expensive in the US intentionally, to counter the natural incentive to create more criminals.
Did u mean create or cremate? 😏
There is little doubt that the anti-capital punishment crowd makes it more expensive deliberately, of course.
“You are driving somewhere far from home, carrying with you fifty thousand dollars in cash that you intend to use to make a down payment on a house.”
A good reason for holding stablecoin.
I agree with and accept all your arguments as good ones - *except* the one about the $50K cash.
Because I do think in today’s world it is *extremely* likely that amount of cash is for or as a result of illegal activity.
And so more “efficient” for the police to confiscate the cash rather than get their revenue through higher taxation of generally moral, law-abiding citizens.
You obviously don't interact with a broad section of society. Many, many people are paranoid about almost anything tech and therefore keep large amounts of cash and wouldn't dream of using any virtual currency
You are absolutely correct that I don’t interact with the parts of society who are that paranoid on the one hand *and also* have $50K they carry around with them *in their cars*. *And* drive through small towns far from home.
I’d love to hear your stories of such examples, however.
For the record, I’ve never owned bitcoin and wasn’t suggesting it as advisable or necessary for most honest people. It was merely a snarky joke.
The post linked to the real story I was thinking of. $87,000.
I didn’t claim zero such stories exist.
I said no more and no less that such cases in the absence of criminality are rare.
Nick tried to claim that many people drive around with that much cash, and through unfamiliar small towns at that.
That was what I was calling him on, not a mere existence proof, which I did not doubt.
If - as I surmise likely - something like 99% of the time people are driving around with that much cash when they are stopped the money was for or resulted from some illegal activity, then I indeed might support the action on the “efficiency” grounds I noted above.
Do not mistake this for my endorsement of all the civil forfeiture techniques used by governments, in particular the more aggressive ones.
Nah. A week ago I bought a used ATV on Facebook Marketplace. Seller would accept cash only. I went to the bank and got $10,000 in cash for him.
There are lots of legit reasons to have a lot of cash.
$10,000, which is the old $7500 before Biden’s inflation, I accept.
$50K is a *lot* more.
But if you in fact have “lots of” legit reasons that people *actually* drive with $50K+ through small towns far from home - not just theoretical ones - then just as I said to Nick, I’d love to hear them.
An even worse case than the civil forfeiture: When China started selling the organs and made some government officials jobs dependent on the trade, it led to the needless killing of perhaps several hundred thousand Falun Gong members and Tibetan Buddhists.
The cost to the criminal of execution is his life, and the cost to us his execution is the cost of the execution process. The execution process can be evaluated using basic accounting methods and given a dollar value. However, the cost of the criminal's life is not something bought or sold - it is not the cost of anyone's life as estimated using actuarial methods, since the criminal loses not a life in general but his own. So there is no dollarized cost to the criminal's life. This is not a problem of inaccurate approximation.
Consequently, the ratio of the cost to the criminal and the cost to "us", is not something that can be computed. Efficiency is not properly measured by ratios where a quantity can't be assigned to each side of the division.
What we can say about different forms of punishment is that people prefer punishments that are less harmful to their ends than others. We can approximate that death is at or near the least preferred punishment. As long as there is some 'bottom' to the list, so that a person does not distinguish between one really bad punishment and a worse one, we can say that there is no marginal deterrent effect of punishments after that one. This gives a good reason not to make all punishments equal to or nearly as bad as the death penalty, because they would not yield a gradient of deterrence for crimes that are gradiently costly to us.
"it is not the cost of anyone's life as estimated using actuarial methods, since the criminal loses not a life in general but his own. So there is no dollarized cost to the criminal's life. This is not a problem of inaccurate approximation." This misses the point. At the time when we pass our criminal laws, we do not know which of us will be (rightly or wrongly) executed in the future, so it's absolutely appropriate to treat the lost life as statistical.
Think of it this way: If you live in a world of 100 people, where each execution (via its deterrent value) saves you $1,000,000 in losses. But each execution also increases the probability that you yourself will be executed by 1%. To say that the value of a life is $10,000,000 is to say that you are willing to accept this tradeoff, and would in fact be willing to accept it even if the probability of your being executed were as much as 10%. That's the tradeoff you want, so it's the tradeoff that the law should give you.
In the initial setup, Friedman describes the cost used in the denominator as as the "cost to the criminal" and at the same time the "deterrence". The notion is that your willingness to do an action is inversely proportional to how much that action costs you. For the cost in the denominator to correctly index for the deterrence, it must be the specific cost to the criminal when they consider whether or not to commit a crime, because this is the cost which has a marginal effect on the choices of the criminal. If not, then the cost in the denominator does not index for deterrence, but something else. But then you're doing an analysis different from the one Friedman seems to have intended.
If we're considering the question at the time we set laws for punishment, we would include the statistical cost of life in the numerator, because this would express part of the cost to us of implementing that policy, i.e., the risk to our lives. That risk we take in passing a law is not the risk a criminal engages in when he commits a crime. The function of the ratio is to be able to express the factor of cost-to-us vs cost-to-the-criminal, with the notion that this ratio indexes for what we pay per unit of deterrence, i.e., the efficiency of a punishment measured by its deterrent effect. Putting the statistical cost of life in the denominator doesn't achieve that goal. So I don't take your point.
The ratio is cost/deterrence.
The cost to the criminal is part of the numerator. It also affects the denominator, but we have no idea *how* it affects the denominator. If the cost of a year in prison to the criminal is $1000, that's a $1000 that belongs in the numerator. If the prospect of that cost deters exactly 7 crimes, then the cost to society of those 7 crimes belongs in the denominator. But how do we know whether it's 7 crimes or 13 or zero? That's a purely empirical question that I don't see addressed.
David, I receive a 403 Forbidden error when I try to follow the Embedded Economics link. Seems to be the case for all of davidfriedman.com. Thought you might want to knowm
It worked for me the second time I tried it. The first time it brought me to a different page. I don't know why.
David, I receive a 403 Forbidden error when I try to follow the Embedded Economics link. Seems to be the case for all of davidfriedman.com
I absolutely fail to understand your first couple of paragraphs. You talk about the ratio of cost to deterrence for various forms of punishment. But the value of the deterrence is an empirical question (how many crimes does an execution prevent? how many crimes does a five-year jail term prevent?) Yet you cited no empirical evidence. So how on earth were you able to assign (approximate) numerical ratios to cost/deterrence when you've given us no indication of how you estimated the denominator?
It sort of works without evidence, but not well. That is, you don't need evidence to know that the cost to the state of imposing a fine is close to 0. So it's basically infinitely efficient, which isn't the case for execution, and certainly isn't the case for incarceration.
However, that's where the need for evidence arises. While fines are basically infinitely efficient, that doesn't mean that they achieve the optimal results. Consider a car with amazing fuel efficiency that can only drive 0.1 miles per day. That is, every day, it can drive 0.1 miles, using hardly any fuel, at all.
That would be remarkable, and more efficient than any alternatives, but not very helpful in getting to work.
As far as the empirics, I'll add that it's widely reported that execution costs more incarceration, due to the lengthy legal expenses associated with trials for the former.
Additionally, there's a distinction between deterrence and incapacitation, with the article perhaps conflating the two. The effects of incapacitation are large, and of course fines, efficient though they may be, don't offer that benefit.
[Now, he could argue that the fines can simply be arbitrarily high, so as to afford sufficient benefits the losses relative to other forms of punishment, but that didn't seem to be his point.]
Request - please turn substack text to speech option on. Until recently if I brought up one of your posts on the Substack app I would see the triangular play icon which if clicked would start reading the text aloud. Now it's gone for your posts, but it's still available and works for the posts of other authors. I'm not certain, but as I understand it, there is an option for Substack authors to turn this feature on or off, so I suspect that option for you is now set to off. If so, I respectfully ask for you to turn it back on.
I have not figured out how to turn it on. Some sources say it is only available for selected Substacks, others imply for all in English.
The text-to-speech icon has now reappeared for this post when reading in the mobile app. I tested it and it works. Perhaps the option was only temporarily disabled during the server problems that Substack was experiencing yesterday.