Thursday, September 10, 2009

Why 'carrots trump sticks,' or, why 'winners don't punish'

Does punishment or reward work better at influencing human behavior? It's a question that's seemingly fundamental to what goes on in the justice system, but one that's seldom subjected to empirical analysis. So I found this discussion of public-goods games by Ed Yong at Not Exactly Rocket Science a fascinating train of thought. He writes that:

Many researchers have turned to punishment for an answer, testing the idea that societies are glued together by the ability to mete out penalties to freeloaders, often at some personal cost. Long -time readers of this blog may remember a set of articles on this topic.

The first, from Simon Gachter's group, showed that the ability to punish freeloaders stabilises cooperative behaviour, but a second study from Martin Nowak showed that this boost in cooperation carries a cost - by escalating conflicts, the ability to punish leaves groups with smaller rewards than those that shun punishments altogether. Gachter disagreed, and in a third study, his group suggested that in the long run, both groups and individuals are better off if punishment is an option. Now, [David] Rand, part of Nowak's group at Harvard University, is back with another take on the debate and this time, he has focused on punishment's cuddlier counterpart - reward.

According to Rand's research:
both rewards and punishment were equally good at promoting cooperation; when these options for interaction were available, the contributions stayed high. However, the two groups that could reward each other earned much higher payoffs than those that could only punish, or those that could do neither.


People also seemed to like playing the do-gooder; in both groups where rewards were available, players used them more and more as the games went on. Punishments, however, decayed over time. The cost of retaliations meant that players hardly ever doled out penalties by the experiment's end. In fact, when both options were on hand, the groups that largely rewarded each other ended up wealthier than those that favoured punishments.


A few questions remain. Rand's modified public goods game was designed to allow reputation to influence how people hand out reward and punishment. This reflects many of our most important interactions - with friends, family and colleagues - and it certainly reflects the world of our ancestors, who lived in tightly knit communities. Whether it applies to our globalised world, where we may only have one-off encounters with others and where online communication offers the boon of anonymity, remains to be seen.

But all in all, Rand's results suggest that when people repeatedly cross each other's paths, carrots are far better than sticks at fostering behaviour for the greater good.

These results strike me as having implications at a societal level for the criminal justice system. Punishment can be an effective way to foster cooperation with social norms, according to these results, but over time the cost of retaliation is excessive. Perhaps this finding may be analogized to the rise of mass incarceration in the United States over the last three decades? Even to the extent it's been effective at enforcing social norms, at this point society has reached a stage where we can no longer afford to use this tactic exclusively, or so this blog has repeatedly argued.

Yong questions whether Rand's analysis applies in a globalized world, but it seems to me that when considering incarceration as punishment, the model works as well in macro as in micro - at least in regard to punishment's negative impacts. The vast majority of offenders who go to prison later get out, so society cannot have a one-and-done attitude toward them without facing negative consequences.

Similarly, emphasizing punishment over reward reduces overall wealth, according to this theory, just as it's the case that prisons both cost society a lot to run and incur significant opportunity costs from the lost wages and productivity of the incarcerated. In the real world as in game theory, the benefit to public order derived from punishment comes at a severe and mounting economic price that reduces wealth for everyone.

For more on this research, see a related article from Nature published last year, "Winners Don't Punish" (pdf).


Karo said...

Eventually the mere lack of a reward is viewed as a punishment. When that happens we stop keeping score and just give all our little snowflakes a trophy.

Gritsforbreakfast said...

No need to "stop keeping score," Karo. According to this data, that would be wealth-maximizing behavior. ;)

PirateFriedman said...

Wow, talk about serious methodological flaws.

During this stage, one group could punish a specific player, paying 4 of their own units to deprive the victim of 12 of theirs. A second group could reward their peers with 12 units at the cost of 4 of their own.

This is such a joke. You start off defining a hypothetical world where punishment is a net loss for society of 8 and rewards are a net gain of 8. And you're surprised by outcome? Seriously. Imagine how this would work in the real world.

Suppose I pay $4 dollars for a parole officer to reward a man who's stayed clean with $12. Does that $12 just appear out of nowhere? No, the taxpayer paid $16 to give him a $12 reward.

Gritsforbreakfast said...

Of course, Pirate, punishment has hidden costs, too.

To take an example similar to your hypothetical about a P.O., look at the war in Iraq - Did the "surge" work because we brought in more troops or because we began paying (i.e., "rewarding") the people who used to shoot at us? Obviously the latter. And we did so because punishing them with military might was MORE expensive in blood and treasure. Taxpayers pay in both cases, but the overall cost is higher in the punishment-only scenario.

sunray's wench said...

Not all rewards have to be monetary though, particularly in a corrections setting. One of the issues with TDCJ at the moment is that there are vast levels of punishment but only very small levels of reward, and much of both are issued in a random manner. There are no real set guidelines for the rewards (if you achive X you will get Y). There are not enough trusty places for the inmates who on paper have earned the right to be there. There are not enough workshop places for those who can afford to be there.

Anonymous said...

Excellent point, SW. Not all rewards are monetary

Dave Rand said...

I thought I should weigh in on the issue of positively non-zero sum rewards. You have to remember that these economic games are metaphorical - they are not supposed to literally represent situations where you give up $4 and magically the other person gains $12. Instead, this type of interaction (formalized as the Prisoner's Dilemma) represents a wide range of situations where you can help someone at a cost to yourself - not necessarily an actual financial cost, could be time/effort/information etc.

Our reward treatment is exactly equivalent to having each person play a public goods game, and then play a Prisoner's Dilemma with each other group member (where their actions in the PD are influenced by the other's action in the public goods game). The existence of these Prisoner's Dilemma type interactions is generally accepted, and that is what we are capturing in reward our setup.

Also, one of the main novel results in our study is that rewards are in fact able to maintain public cooperation as well as punishments - previous studies, which limited reputation and repetition, had found rewards to be relatively ineffective. To quote from the paper:

"In the RN and RNP treatments [where reward is possible], there is the possibility of generating additional income during the targeted interactions. Thus, it follows naturally from [the ability of rewards to stabilize public goods game contributions] that the reward treatments, RN and RNP, generate larger absolute payoffs than the punishment-only treatment, PN. Groups that have the opportunity to reward do better than groups that can only punish. The point we want to make is this: If several
targeted interactions can promote cooperation in the public goods game, then those that generate additional positive payoff will result in the best outcomes."

Thanks for your interest in our work!

David Rand
Program for Evolutionary Dynamics
Harvard University

Soronel Haetir said...

A question I have about this work, do the results remain if you select a population that has already shown itself to be non-cooperators? IE people who are mentally ill, drug dependent, otherwise pervasively criminal?

I have seen plenty of work tending to show that the people who tend to commit crimes already have a skewed sense of reward-punishment.

Anonymous said...

Psychological studies have long recognized that rewards are much stronger than punishments in promoting desired behaviors and outcomes. This research appears consistent with psychological research.

It is also consistent with an increasing body of criminological research that demonstrate the limited impacts of deterrence on criminal behavior (i.e., the imposition of punishment to change offending behaviors - specific or general).

With relatively consistent findings coming from several disciplines -- perhaps we can begin to rethink how criminal justice and the larger contruct "justice" can be achieved.

Swift said...

As long as we are creating hypothetical scenarios, how about one that is slanted TO FAVOR of punishment:

Thee Strikes and You're (Parted) Out
Repeat offenders will be euthanized and their organs will be sold for transplant. There is precedent for this in China where executed prisoners are a "major source of organs for transplants." Also CNN recently reported that an Israeli man named Nick Rosen sold one of his kidneys for $20,000! Here is a partial list of transplantable organs: Heart , Lung, Kidney, Liver, Pancreas, Cornea, Skin, and Bone Marrow. I believe the State could easily clear $100000 profit per "donor" but we would need to change execution methods to avoid poisoning the assets. Perhaps a bullet to the back of the head or hanging would be more appropriate. By avoiding the use of lethal chemicals, the remaining remains could be used as a better tasting version of Vita Pro. (See also: Soylent Green)

This scenario would create a large financial incentive in favor of punishment. Reason would dictate that the organs of younger "donors" may be more valuable than those of older criminals so perhaps we should consider allowing juvenile records to count as "strikes" when determining when an offender is to be parted out. In years with State budget shortfalls, the program could be re-worked to require only Two strikes.

Soronel Haetir said...

Also, prison itself is not intended to be punishment, though various levels of incarceration within prison are.

Prison's main benefit is one of segregation and incapacitation.

PirateFriedman said...

Dr. Rand, thank you for the clarification. I can see how this experiment relates to the prisoners dillema,

I do not really see how it relates to a public crime prevention strategy. Perhaps Grits will clarify his vision on this sometime. As Karo pointed out, most criminals will see rewarding good behavior as a punishment for those who don't recieve the reward.

Gritsforbreakfast said...

Pirate, I'll respond further in another post, but there are many examples of how this can work.

E.g., part of Texas probation reforms in 2007 was to give offenders more chances to earn their way off probation through good behavior, incentivizing the behaviors and outcome we want instead of focusing solely on punishing aberrance.

The use of good time or earned time as incentive for good behavior in prison is another example of where this approach is already in place but could be utilized more robustly.

Or consider the situation of mentally ill, homeless "frequent flyers" in the jails. Maybe providing housing, case workers, medication, healthcare, etc., seems like a "reward" to some (though it also comes with more intensive state supervision while on probation). But if it keeps them stable and committing fewer crimes it's worth the investment.

But I also think this approach to security potentially works on many levels. See my comments above about paying Iraqi insurgents instead of fighting them. US Sen. Carl Levin just got back from Afghanistan and I heard him critizing the Obama Administration on NPR today for not utilizing the same pay-em-off strategy with the already divided factions of the Taliban.

We can't give up the stick; sometimes it's needed. But we could stand to rely much more on the carrot.

sunray's wench said...

Having watched District 9 yesterday, and reading Swift's hypothetical scenario, I feel like my warning bells are sounding. While I favour the carrot and stick approach, I can see where it could easily be reduced to the lowest denominator: if you do X because we tell you to, we wont kill you.

I know the power has to remain with those who make the rules, the representatives of the people etc, but when you can't trust those who make the rules (as is displayed regularly by dishonest DAs, Judges, Correctional staff, Chaplains, etc) in Texas, then you could just be handing them an open social-check book on how to "cleanse" Texas of anyone they don't want to be there. And then once the criminals have gone, who do they focus on next?

Anonymous said...

response to sunray's wench:

Yes, rewards ARE given in prison, from what I have been told. Take, for instance, the cellphone contraband issue. Besides the obvious point of making calls and texting, the cellphones can be used as torture devices, of which the prison gaurds are very well aware of. See article "Teen Torture Iphone App Discontinued Due To Public Alarm" at

It reminds me of the Steven King book called "Cell" ( see blurb at

Apart from that, the issue of rewards vs. punishment as discussed here overlooks the job creation "benefits" needed in a down economy, see

Anonymous said...

Anonymous, I don't know that the gaurds are aware that cell phones can be used to torture. Some can't even read or use computers. They peck at the keyboards as evidenced by the long 2 hour waits at the gates when one goes to visit their family member behind bars. Of course it is torture that you can't call your family when you need to or even have a cell phone. And it's torture to pretend to give rewards...such as the fabled "good time" that rewards you nothing although you have accrued years and years of it.

Anonymous said...

Of course the gaurds are aware of it, since even in the link I supplied ( ) it talks about how public saftey people use it to enforce curfews...

morat said...

Thanks for sharing this great article! That is very interesting Smile I love reading and I am always searching for informative information like this.
Kenali dan Kunjungi Objek Wisata di Pandeglang | Blog SEO | cah bagoes