My $0.02: there are several different functions person A can perform by punishing person B for some action C.
For example: (a) lowering B’s chances of performing C in the future (b) lowering the chances of observers performing C (c) encouraging observers to anti-identify with B (d) encouraging observers who anti-identify with B to support A (e) encouraging observers who identify with B to oppose A
IME, conversations about how prisons should work become really confused because people aren’t very clear about which of those functions they endorse.
Personally, it seems clear to me that (b) is by far the most valuable of these goals. That said, prison policy has almost no influence on (b); law enforcement and courts are far more relevant, and their current implementation pretty much screens off the effects of prison policy.
People who are interested in (a) and also value B’s continued existence will tend to be interested in punishment as a behavioral modification tool, and will happily set it aside in favor of more effective behavioral modification tools as science develops them.
People interested in (a) who don’t value B’s continued existence will be uninterested in punishment, since simply killing B is more efficient.
AFAICT, the folks who establish the policies that govern prisoner punishment (as distinct from prisoner restraint) are primarily motivated by the desire to obtain political support, which suggests minimizing (e) and maximizing (d), which does seem to be what most of our prison policies are designed to do. Maximizing (c) is one way to minimize (e), though there are many others.
Personally, it seems clear to me that (b) is by far the most valuable of these goals. That said, prison policy has almost no influence on (b); law enforcement and courts are far more relevant, and their current implementation pretty much screens off the effects of prison policy.
This isn’t obvious at all. In particular if prisons were extremely nice, their deterrent effect would be much less no matter how law enforcement and the courts worked. One could argue that the policies in the current Overton window aren’t significantly different from each other, but that argument would have to be made.
Agreed that if prisons were extremely nice, their deterrent effect due to the threat of punishment would be lower than it is now.
That said… when the mechanism that results in my being punished for an act is perceived as unreliable and capricious (including, but not limited to, cases where it is unreliable and capricious), the correlation between the severity of the punishment and the intensity of the deterrent effect is much, much lower than when the mechanism is perceived as fair and reliable.
So if law enforcement and courts were perceived as fair and reliable (that is, reliably assigning punishment to criminals and not assigning punishment to noncriminals), I expect making prisons equally unpleasant would create a much greater deterrent effect (to being a criminal) than it does now.
If my goal is to maximize deterrent effect, then, I expect that I would do better to invest my efforts in increasing the perception of law enforcement and courts as fair and reliable than to invest them in increasing the perception of prisons as unpleasant.
But, as I say, I don’t think many people involved in setting prison policies are primarily motivated by maximizing deterrent effect.
That said… when the mechanism that results in my being punished for an act is perceived as unreliable and capricious (including, but not limited to, cases where it is unreliable and capricious), the correlation between the severity of the punishment and the intensity of the deterrent effect is much, much lower than when the mechanism is perceived as fair and reliable.
Depending on what you mean by “unreliable and capricious”, I find this dubious. At the very least it seems to me that brutal dictatorships are much better at reducing crime (at least the crimes they care about) than democracies. For example, Mussolini’s successful campaign against the Sicilian mafia.
What I mean by enforcement being unreliable and capricious is, roughly. that agents believe that their performing the act is not well-correlated with their being punished.
It sounds from that wiki article like Mussolini created an environment where people believed that being a mafioso would reliably result in being punished.
I suspect they also believed that not being a mafioso stood a good chance of being punished, which has other consequences; when punishment occurs in the absence of a reliable and controllable cue, the result is learned helplessness. But if we care about deterring criminals and we don’t care about the effect on noncriminals, punishing 90% of criminals and 5% of noncriminals can work OK, even if only 5% of the people we punish are criminals.
Of course, if we care about things in addition to deterrence, that may not be a great policy, but that’s another conversation.
What I mean by enforcement being unreliable and capricious is, roughly. that agents believe that their performing the act is not well-correlated with their being punished.
So what you’re saying is that in modern developed states committing crimes is not well-correlated with being punished? I find this highly dubious.
At the very least, I’m saying that that’s the perception: most crimes go unpunished. But yes, I also suspect that perception is true.
I haven’t done any research on the matter, though, and attempts to find statistics via cursory Googling failed. If you have any cites handy, I’m happy to be corrected.
committing crimes is not well-correlated with being punished? I find this highly dubious.
TheOtherDave said:
the perception [is] most crimes go unpunished.
These aren’t actually in contradiction. If a criminal committing a “mid-size” offense has a 25% chance of being caught for each crime, then being a career criminal is likely to end you in jail, but most crimes will still be unpunished.
My sense is that most crimes (and most dollar-loss to crime) are small/midsize thefts; hundreds or thousands of dollars, not more. Thefts big enough to set you up for a lifetime are freakishly rare compared to the number of criminals. And that means to have a tolerable lifestyle as a criminal, you have to commit lots of offenses—so even a small chance of being caught for each mugging or burglary starts to add up.
Yeah, I agree with this. I’d be surprised if the chance was as high as .25, but the principle is the same; career criminals can count on eventually being arrested.
That said, the original context of this discussion was the behavior-modification effects of prison policy on the not-yet-arrested population, and from a behavior modification point of view a punishment that usually fails to kick in for the first several crimes doesn’t do much to deter those first few crimes.
And making the punishment more and more severe doesn’t help the deterrence factor all that much in that situation, which was my original point.
and from a behavior modification point of view a punishment that usually fails to kick in for the first several crimes doesn’t do much to deter those first few crimes.
Disagree. It deters the first crime. It’s deterrent power will decrease for subsequent crimes (until caught) unless the criminal has friends who have been caught.
Can you say more about the mechanism whereby increasing the severity of a punishment I am confident won’t apply to my first crime deters my first crime? That seems pretty implausible to me.
If committing a crime required playing Russian Roulette, a gun with a bullet in it would be more of a deterrent than a gun with a paintball in it. Yes?
The law-enforcement/courts system has significantly better first-time odds than Russian Roulette. For most crimes, the odds that I will be arrested and convicted and sentenced to significant jail time for a first crime are significantly lower than one in six.
“But Dave,” someone will now patiently explain to me, “that doesn’t matter. An N% chance of death is always going to be significantly worse than an N% chance of a paintball in the head, no matter how low N%. It’s scale-invariant!”
Except the decision to ignore the psychological effects of scale is precisely what I’m skeptical about here. Sure, if I make prisons bad enough (supposing I can do so), then everyone rational does an EV calculation and concludes that even a miniscule chance of going to prison is more disutility than the opportunity cost of foregoing a crime.
But I don’t think that’s what most people reliably do faced with small probabilities of large disutilities. Some people, faced with that situation, look at the magnitude of the disutility and ignore the probability (“Sure it’s unlikely, but if it happened it would be really awful, so let’s not take the risk!”). Some people look at the magnitude of the probability and ignore the disutility (“Sure, it would be awful, but it’s not going to happen, so who cares?”).
Very few look at the EV.
That said, if we restrict our domain of discourse to potential criminals who do perform EV calculations (which I think is a silly thing to do in the real world, but leaving that aside for now), then I agree that doubling the expected disutility-of-punishment (e.g., making prisons twice as unpleasant) halves their chance of performing the crime.
Of course, so does doubling the expected chance of being punished in the first place .
That is, if I start out with a P1 confidence that I will be arrested and convicted for commiting a crime, a P2 confidence that if convicted I will receive significant prison time, and a >.99 confidence that the disutility of significant prison time is D1, and you want to double my expected disutility of commiting that crime, you can double P1, or P2, or D1, or mix-and-match.
So a system primarily interested in maximizing deterrent effect among rational EV calculators asks which of those strategies gets the largest increase in expected disutility for a given cost.
It’s not at all clear to me that in the U.S. today, doubling D1 is the most cost-effective way to do that if I consider decreasing the QALYs of prison inmates to be a cost. So if someone insists on doubling D1, I infer that either...: ...(a) they value the QALYs of prison inmates less than I do, or ...(b) they have some reason to believe that doubling D1 is the most cost-effective way of buying deterrence, or ...(c) they aren’t exclusively interested in deterrence, or ...(d) something else I haven’t thought of.
In practice I usually assume some combination of (a) and (c), but I considered (b) potentially interesting enough to be worth exploring the question. At this point, though, my confidence that I can explore (b) in this conversation in an interesting way is low.
Some people look at the magnitude of the probability and ignore the disutility (“Sure, it would be awful, but it’s not going to happen, so who cares?”).
It seems rather difficult to actually affect those people, though. The difference between P1=.04 and P1=.08 would have dramatic effects on an EV-calculator, but very little effect on the sort of person who judges probabilities by ‘feel’.
That is, if I start out with a P1 confidence that I will be arrested and convicted for commiting a crime, a P2 confidence that if convicted I will receive significant prison time, and a >.99 confidence that the disutility of significant prison time is D1, and you want to double my expected disutility of commiting that crime, you can double P1, or P2, or D1, or mix-and-match.
I would suppose the D1 advocates would argue that the hidden costs of increasing P1 are higher than you think, or possibly they just value them more (e.g. the right to privacy). I admit I’ve never heard a good argument that what the US needs is to greatly increase the likelihood of sentencing a convict to significant prison time.
The difference between P1=.04 and P1=.08 would have dramatic effects on an EV-calculator, but very little effect on the sort of person who judges probabilities by ‘feel’.
I would expect it depends a lot on the algorithms underlying “feel” and what aspects of the environment they depend on. It’s unlikely these people are choosing their behaviors or beliefs at random, after all.
More generally, if I actually want to manipulate the behavior of a group, I should expect that a good first step is to understand how their behavior depends on aspects of their environment, since often their environment is what I can actually manipulate.
Edit: I should add to this that I certainly agree that it’s possible in principle for a system to be in a state where the most cost-effective thing to do to achieve deterrence is increase D. I just don’t think it’s necessarily true, and am skeptical that the U.S. is currently in such a state.
the hidden costs of increasing P1 are higher than you think
Sure, that’s another possibility. Or of P2, come to that.
I admit I’ve never heard a good argument that what the US needs is to greatly increase the likelihood of sentencing a convict to significant prison time.
Is this not the rationale behind mandatory sentencing laws?
I admit I’ve never heard a good argument that what the US needs is to greatly increase the likelihood of sentencing a convict to significant prison time.
Is this not the rationale behind mandatory sentencing laws?
I can’t think of a response to this that isn’t threatening to devolve into a political argument, so I’ll bow out here. Sorry.
At the very least, I’m saying that that’s the perception: most crimes go unpunished.
But yes, I also suspect that perception is true. I haven’t done any research on the matter, though, and attempts to find statistics via cursory Googling failed.
If you have any cites handy, I’m happy to be corrected.
In that case, why aren’t you stealing money and donating to SIAI? ;)
But seriously, there are countries where your comment is actually true. You can tell the difference pretty easily.
To be honest, I’m not convinced that it isn’t true even in first-world countries. Solve rates for murders in the US appear to be around 66% as of 2007. I haven’t directly been able to dig up solve rates for crimes in general, but clearance rates (the rate of crimes prosecuted to crimes reported) are available, and are well under 50% for pretty much everything except murder. Most prosecuted crimes appear to result in convictions, but this still says to me that TheOtherDave’s got it right, at least in a US context and assuming that most reports aren’t frivolous.
YMMV for other nations.
ETA: Looking over these statistics again, I strongly suspect that the “solve” figures you find in various places are in fact identical to the clearance rates I refer to. So the reports-to-convictions ratio would be significantly lower—compare conviction rates for cases brought to court.
I infer that your intuitions differ from mine but you don’t have any cites handy either. Fair enough. Updated, to a degree proportional to my confidence in the reliability of your intuition on this matter, in your direction.
My $0.02: there are several different functions person A can perform by punishing person B for some action C.
For example:
(a) lowering B’s chances of performing C in the future
(b) lowering the chances of observers performing C
(c) encouraging observers to anti-identify with B
(d) encouraging observers who anti-identify with B to support A
(e) encouraging observers who identify with B to oppose A
IME, conversations about how prisons should work become really confused because people aren’t very clear about which of those functions they endorse.
Personally, it seems clear to me that (b) is by far the most valuable of these goals. That said, prison policy has almost no influence on (b); law enforcement and courts are far more relevant, and their current implementation pretty much screens off the effects of prison policy.
People who are interested in (a) and also value B’s continued existence will tend to be interested in punishment as a behavioral modification tool, and will happily set it aside in favor of more effective behavioral modification tools as science develops them.
People interested in (a) who don’t value B’s continued existence will be uninterested in punishment, since simply killing B is more efficient.
AFAICT, the folks who establish the policies that govern prisoner punishment (as distinct from prisoner restraint) are primarily motivated by the desire to obtain political support, which suggests minimizing (e) and maximizing (d), which does seem to be what most of our prison policies are designed to do. Maximizing (c) is one way to minimize (e), though there are many others.
This isn’t obvious at all. In particular if prisons were extremely nice, their deterrent effect would be much less no matter how law enforcement and the courts worked. One could argue that the policies in the current Overton window aren’t significantly different from each other, but that argument would have to be made.
Agreed that if prisons were extremely nice, their deterrent effect due to the threat of punishment would be lower than it is now.
That said… when the mechanism that results in my being punished for an act is perceived as unreliable and capricious (including, but not limited to, cases where it is unreliable and capricious), the correlation between the severity of the punishment and the intensity of the deterrent effect is much, much lower than when the mechanism is perceived as fair and reliable.
So if law enforcement and courts were perceived as fair and reliable (that is, reliably assigning punishment to criminals and not assigning punishment to noncriminals), I expect making prisons equally unpleasant would create a much greater deterrent effect (to being a criminal) than it does now.
If my goal is to maximize deterrent effect, then, I expect that I would do better to invest my efforts in increasing the perception of law enforcement and courts as fair and reliable than to invest them in increasing the perception of prisons as unpleasant.
But, as I say, I don’t think many people involved in setting prison policies are primarily motivated by maximizing deterrent effect.
Depending on what you mean by “unreliable and capricious”, I find this dubious. At the very least it seems to me that brutal dictatorships are much better at reducing crime (at least the crimes they care about) than democracies. For example, Mussolini’s successful campaign against the Sicilian mafia.
What I mean by enforcement being unreliable and capricious is, roughly. that agents believe that their performing the act is not well-correlated with their being punished.
It sounds from that wiki article like Mussolini created an environment where people believed that being a mafioso would reliably result in being punished.
I suspect they also believed that not being a mafioso stood a good chance of being punished, which has other consequences; when punishment occurs in the absence of a reliable and controllable cue, the result is learned helplessness. But if we care about deterring criminals and we don’t care about the effect on noncriminals, punishing 90% of criminals and 5% of noncriminals can work OK, even if only 5% of the people we punish are criminals.
Of course, if we care about things in addition to deterrence, that may not be a great policy, but that’s another conversation.
So what you’re saying is that in modern developed states committing crimes is not well-correlated with being punished? I find this highly dubious.
At the very least, I’m saying that that’s the perception: most crimes go unpunished.
But yes, I also suspect that perception is true. I haven’t done any research on the matter, though, and attempts to find statistics via cursory Googling failed.
If you have any cites handy, I’m happy to be corrected.
Eugine said:
TheOtherDave said:
These aren’t actually in contradiction. If a criminal committing a “mid-size” offense has a 25% chance of being caught for each crime, then being a career criminal is likely to end you in jail, but most crimes will still be unpunished.
My sense is that most crimes (and most dollar-loss to crime) are small/midsize thefts; hundreds or thousands of dollars, not more. Thefts big enough to set you up for a lifetime are freakishly rare compared to the number of criminals. And that means to have a tolerable lifestyle as a criminal, you have to commit lots of offenses—so even a small chance of being caught for each mugging or burglary starts to add up.
Yeah, I agree with this. I’d be surprised if the chance was as high as .25, but the principle is the same; career criminals can count on eventually being arrested.
That said, the original context of this discussion was the behavior-modification effects of prison policy on the not-yet-arrested population, and from a behavior modification point of view a punishment that usually fails to kick in for the first several crimes doesn’t do much to deter those first few crimes.
And making the punishment more and more severe doesn’t help the deterrence factor all that much in that situation, which was my original point.
Disagree. It deters the first crime. It’s deterrent power will decrease for subsequent crimes (until caught) unless the criminal has friends who have been caught.
Can you say more about the mechanism whereby increasing the severity of a punishment I am confident won’t apply to my first crime deters my first crime? That seems pretty implausible to me.
If committing a crime required playing Russian Roulette, a gun with a bullet in it would be more of a deterrent than a gun with a paintball in it. Yes?
The law-enforcement/courts system has significantly better first-time odds than Russian Roulette. For most crimes, the odds that I will be arrested and convicted and sentenced to significant jail time for a first crime are significantly lower than one in six.
“But Dave,” someone will now patiently explain to me, “that doesn’t matter. An N% chance of death is always going to be significantly worse than an N% chance of a paintball in the head, no matter how low N%. It’s scale-invariant!”
Except the decision to ignore the psychological effects of scale is precisely what I’m skeptical about here. Sure, if I make prisons bad enough (supposing I can do so), then everyone rational does an EV calculation and concludes that even a miniscule chance of going to prison is more disutility than the opportunity cost of foregoing a crime.
But I don’t think that’s what most people reliably do faced with small probabilities of large disutilities. Some people, faced with that situation, look at the magnitude of the disutility and ignore the probability (“Sure it’s unlikely, but if it happened it would be really awful, so let’s not take the risk!”). Some people look at the magnitude of the probability and ignore the disutility (“Sure, it would be awful, but it’s not going to happen, so who cares?”).
Very few look at the EV.
That said, if we restrict our domain of discourse to potential criminals who do perform EV calculations (which I think is a silly thing to do in the real world, but leaving that aside for now), then I agree that doubling the expected disutility-of-punishment (e.g., making prisons twice as unpleasant) halves their chance of performing the crime.
Of course, so does doubling the expected chance of being punished in the first place .
That is, if I start out with a P1 confidence that I will be arrested and convicted for commiting a crime, a P2 confidence that if convicted I will receive significant prison time, and a >.99 confidence that the disutility of significant prison time is D1, and you want to double my expected disutility of commiting that crime, you can double P1, or P2, or D1, or mix-and-match.
So a system primarily interested in maximizing deterrent effect among rational EV calculators asks which of those strategies gets the largest increase in expected disutility for a given cost.
It’s not at all clear to me that in the U.S. today, doubling D1 is the most cost-effective way to do that if I consider decreasing the QALYs of prison inmates to be a cost. So if someone insists on doubling D1, I infer that either...:
...(a) they value the QALYs of prison inmates less than I do, or
...(b) they have some reason to believe that doubling D1 is the most cost-effective way of buying deterrence, or
...(c) they aren’t exclusively interested in deterrence, or
...(d) something else I haven’t thought of.
In practice I usually assume some combination of (a) and (c), but I considered (b) potentially interesting enough to be worth exploring the question. At this point, though, my confidence that I can explore (b) in this conversation in an interesting way is low.
It seems rather difficult to actually affect those people, though. The difference between P1=.04 and P1=.08 would have dramatic effects on an EV-calculator, but very little effect on the sort of person who judges probabilities by ‘feel’.
I would suppose the D1 advocates would argue that the hidden costs of increasing P1 are higher than you think, or possibly they just value them more (e.g. the right to privacy). I admit I’ve never heard a good argument that what the US needs is to greatly increase the likelihood of sentencing a convict to significant prison time.
I would expect it depends a lot on the algorithms underlying “feel” and what aspects of the environment they depend on. It’s unlikely these people are choosing their behaviors or beliefs at random, after all.
More generally, if I actually want to manipulate the behavior of a group, I should expect that a good first step is to understand how their behavior depends on aspects of their environment, since often their environment is what I can actually manipulate.
Edit: I should add to this that I certainly agree that it’s possible in principle for a system to be in a state where the most cost-effective thing to do to achieve deterrence is increase D. I just don’t think it’s necessarily true, and am skeptical that the U.S. is currently in such a state.
Sure, that’s another possibility. Or of P2, come to that.
Is this not the rationale behind mandatory sentencing laws?
I can’t think of a response to this that isn’t threatening to devolve into a political argument, so I’ll bow out here. Sorry.
In that case, why aren’t you stealing money and donating to SIAI? ;)
But seriously, there are countries where your comment is actually true. You can tell the difference pretty easily.
To be honest, I’m not convinced that it isn’t true even in first-world countries. Solve rates for murders in the US appear to be around 66% as of 2007. I haven’t directly been able to dig up solve rates for crimes in general, but clearance rates (the rate of crimes prosecuted to crimes reported) are available, and are well under 50% for pretty much everything except murder. Most prosecuted crimes appear to result in convictions, but this still says to me that TheOtherDave’s got it right, at least in a US context and assuming that most reports aren’t frivolous.
YMMV for other nations.
ETA: Looking over these statistics again, I strongly suspect that the “solve” figures you find in various places are in fact identical to the clearance rates I refer to. So the reports-to-convictions ratio would be significantly lower—compare conviction rates for cases brought to court.
I infer that your intuitions differ from mine but you don’t have any cites handy either.
Fair enough.
Updated, to a degree proportional to my confidence in the reliability of your intuition on this matter, in your direction.