Tackling The Cobra Effect and Other Perverse Incentives
A taxonomy of unintended consequences and some potential solutions
If you are like me, you can get frustrated when you hear of well-intentioned government programs and policies that are exploited or abused in unintended ways. There are many examples of naive or imprecise policy, but here I will be focusing on what economists call perverse incentives: what they are, where people go wrong, and how they might be prevented.
To set the stage, let’s hand the microphone to Mark Twain, aka Samuel Clemens. He wrote, “Once in Hartford the flies were so numerous for a time, and so troublesome, that Mrs. Clemens conceived the idea of paying George a bounty on all the flies he might kill. The children saw an opportunity here for the acquisition of sudden wealth... Any Government could have told her that the best way to increase wolves in America, rabbits in Australia, and snakes in India, is to pay a bounty on their scalps. Then every patriot goes to raising them.”
One of the things Twain is referring to is an incident in Colonial Delhi where British administrators thought there were too many cobras and offered a reward for each dead cobra people turned in. At first, this worked: people hunted the cobras, and the local cobra population decreased. But it soon came to light that people had been breeding cobras in their own homes because of the reward money. So the government, not interested in just giving money away, stopped the program. And because venomous snakes were the last thing people wanted in their homes, cobras were released into the wild, and then the problem was worse than before. Though the Delhi administrators wanted to incentivize fewer cobras, they wound up getting more of them, leading economists to label the incident as The Cobra Effect.
The Cobra Effect pattern happens when a government decides to reward quantity over quality. The fact that people respond to incentives and not what any policymaker thinks the public “should” do is not surprising; that’s a fact plain fact of the human condition. But given that human nature is hard to change, I submit that a specific cognitive error on the part of the policymaker gives rise to The Cobra Effect.
Specifically, on a technical level, the problem is that the person offering the reward thinks of each cobra as basically equivalent: no cobra is any more or less important than any other. Economists would describe this situation by saying that the cobras are assumed to be “fungible” or interchangeable. If you think of each cobra as an instance of the same class, you are stuck rewarding quantity over quality. However, should you have DNA sequencing technology, cobras are no longer fungible, and you can detect the difference between “wild-caught” cobras and “home-bred” cobras.
Indeed, once you spot the error — and assuming you have the technology — you can fix the incentive problem. Instead of offering money for each cobra unconditionally, a) record the DNA of each previously approved and rewarded cobra, and b) make the payout inversely proportional to the genetic similarity between the current cobra in question and the most similar previously caught cobra. In other words, let's say five hundred cobras have already been turned in to the government and sequenced. If the cobra #501 is suspiciously similar to the cobra #329, don’t pay any money. In one fell swoop, by essentially adding in a second incentive (dissimilarity) that is in tension with the first incentive (number of cobras), you’ve prevented the government from rewarding cobra breeding.
Just to be safe, you’d also want to ensure the public understands this scheme, make falsifying cobra DNA sequences a punishable offense, and just for good measure, make it illegal to order, transport, and receive cobras and cobra eggs. Finally, to prevent people from bringing Cobras in from out of town, you might also want to define a maximum dissimilarity beyond which you won’t pay anything.
I am well aware that genetic sequencing technology didn’t exist back in Colonial Delhi. So in a practical sense, the solution has two parts: a) recognizing the assumption of fungibility, and b) having some way of measuring or defining “quality” instead of quantity. But the theoretical case is important because a bevy of local governments have made (and are currently making) the exact same mistake.
In fact, had genetic sequencing technology existed, identifying the error of fungibility could have also prevented the Great Hanoi Rat Massacre. As Wikipedia summarizes it:
“The Great Hanoi Rat Massacre occurred in 1902, in Hanoi, Vietnam (then known as French Indochina), under French colonial rule, the colonial government created a bounty program that paid a reward for each rat killed. To collect the bounty, people would need to provide the severed tail of a rat. Colonial officials, however, began noticing rats in Hanoi with no tails. The Vietnamese rat catchers would capture rats, sever their tails, then release them back into the sewers so that they could produce more rats.”
Not mentioned in the Wikipedia article, but mentioned in other places is that officials also discovered at least one rat farm.
Once again, comparing the genetic similarity of any new rat tail to any previously cataloged one would have helped, and the reason that using such a measure of similarity can work is that it’s a computable way to determine the quality of the catch.
On one level, what I’m saying is “better incentives are good.” But to get more precise, given that any single incentive is going to be exploited, one solution is actually to have two incentives, both of which are in tension with each other.
Moreover, if we are to have any hope at combatting perverse incentives, we’ll need a taxonomic playbook of a) the different subcategories of perverse incentives, and b) the solution to each type. I also propose we think of all cases in which the similarity trick could work as being of the same subcategory of perverse incentive.
So without too much further ado, I will now work through a sample of the list of historical anecdotes on the Wikipedia list of Perverse Incentives and outline a) the more specific category to which each belongs and b) how each one could be solved. But before I do, I want to note that the list is biased toward public policy. It barely scratches the surface of perverse incentives in business. I also want to note that this categorization attempt isn’t perfect: some incidents and anecdotes could go into more than one category, and the boundaries are a little fuzzy. Note the proposed solutions are necessarily approximations. And lastly, if you want more links, sources and footnotes for each quoted incident, you’ll find them on the Wikipedia article.1
A. Cobra Effects
In addition to the incident of Delhi Cobras and the Great Hanoi Rat Massacre, here are two more examples of the same quantity over quality error, again with a similar solution:
“The 20th-century paleontologist G. H. R. von Koenigswald used to pay Javanese locals for each fragment of hominin skull that they produced. He later discovered that the people had been breaking up whole skulls into smaller pieces to maximise their payments.”
Say it with me: GENETICALLY TEST THE BONE FRAGMENTS. In addition:
“[In 2008,] Experiencing an issue with feral pigs, the U.S. Army post of Fort Benning in Georgia offered hunters a $40-bounty for every pigtail turned in. Predictably, however, people began to buy pigtails from butchers and slaughterhouses at wholesale prices then resold the tails to the Army at the higher bounty price.”
I did some research on this incident, and while people cheated by buying pigtails from butchers, the real reason the pigs increased is that people put out food for the pigs. People weren’t breeding them in their own homes or on farms, but they were providing the right conditions for the pigs to breed in the wild. Once again, it seems likely that genetic tracking/testing and not rewarding genetically similar pigs — along with clearly explaining to people what sorts of things would not get rewarded — could have prevented this.
Alas, this pattern recurs with some regularity. According to “Government Pest Control Only Breeds More Pests” (an article by the Foundation for Economic Education)2, as of 2017:
“… the city council in Petaling Jaya, Malaysia implemented a bounty on rats within the last year and a half. Similarly, Fayette County, Texas has a bounty on feral hog tails and ears, and unsurprisingly, there is suspicion of fraud. Caldwell County, Texas has a bounty on feral hog tails. Authorities in Australia’s Shire of Banana recently announced a bounty for feral cat scalps.3 Just this year, Louisiana’s Terrebonne Parish considered increasing the size of payments in its long-standing rat bounty program because the current program is failing to make a dent in the rodent’s population. Florida has a bounty on pythons.”
I take the prevalence of Cobra Effects in the 21st century to be evidence that the error of fungibility is easy to make, especially as it relates to pests. Paying a bounty for them seems like such a simple solution! Alas, as we have seen, it is not.
B. Uneven Incentive Gradients
Many of the other entries in the Wikipedia list of perverse incentives fall into a subcategory I call “Uneven Incentive Gradients.” The Cobra Effect happens with a single quantity-over-quality incentive in a decentralized setting. However, many other non-cobra-effect perverse incentives occur when there are competing incentives due to the fact that people will seek the one that pays better. To the extent that there is a cognitive error here, it is one of laziness and naïveté on the part of the policymakers:
“The Duplessis Orphans — Between 1945 and 1960, the federal Canadian government paid orphanages 70 cents per day, per orphan, and paid psychiatric hospitals $2.25 per day, per patient. Allegedly, up to 20,000 orphaned children were falsely certified as mentally ill so that the province of Quebec could receive the larger payment.”
In this example, the uneven gradient is between two competing policies, and I believe the solution to be a holistic combination of mitigations.4 For example…. First, put a barrier in the middle of the gradient by making the certification process not dependent on the judgment of any one person with the same agenda. Second, make abusing the system and falsifying records punishable with jail time. Third, track the rates of mental illness in local towns and look for any statistical anomalies, and so on.
This next example comes from the state of Oregon, which started to reward recycled containers at a higher rate than Washington. Guess who started to receive out-of-state bottles? (In addition, people started to steal large amounts of cans and bottles, but that seems like an incentive problem with any and all goods of value).
“Container-deposit legislation provides for a refundable deposit to be placed on beverage containers. When returned to an authorized redemption center the deposit is partly or fully refunded to the redeemer. Intended to encourage recycling and curb litter, these programs may result in containers, already bound for recycling, to be illegally collected from curbside bins or dumpsters. These containers may later be abandoned, resulting in more litter. Additionally, containers may be diverted to nearby regions with higher redemption values, as famously depicted in the Seinfeld episode The Bottle Deposit.”
In this example, the uneven incentive gradient is geographic in nature. In response to this, lawmakers made it so that stores could elect to stop rewarding people who they reasonably thought were bringing containers in from Washington. (Another way to erect a barrier in the middle of the gradient would be to demarcate which bottles were sold in Oregon, potentially by stamping them with a special number, and only reward those.) Notice that the incentive for people to steal bottles and turn them in — however undesirable this may be — doesn’t actually harm the goal of recycling more bottles.
For another uneven geographic incentive gradient:
“In the 2000s, Canada negotiated a "Safe Third Country Agreement"… Among the provisions was one that denied anyone entering Canada at an official port of entry from requesting asylum there, in theory limiting asylum applications to either those filed by refugees in camps abroad or those who could legally travel to Canada and do so at an immigration office. In the late 2010s, some migrants began entering Canada irregularly, between official border crossings, at places like Roxham Road between New York and Quebec, since once they were in Canada they were allowed to file applications with the full range of appeals available to them, a process that could take years. Canada wound up processing thousands more applications for asylum than it had planned to.”
Incidents like these really make you wonder, “how many unintended consequences could be prevented if policymakers were attuned to possible perverse incentives?” I don’t have time to attempt a good solution here and still get this post out on time, but I hope that naming the subcategory can spark further thought about how one would solve this problem.
Back to the non-geographic arena:
“Under the American Medicare program, doctors were reimbursed at a higher rate if they administer more expensive medications to treat a condition. This created a potential incentive for the physician to prescribe a more expensive drug when a less expensive one might do.”
Once again, this is an example of an uneven incentive gradient. With the caveat that I am not an expert on medical economics, it really makes me wonder: surely it doesn’t take that much more cognitive effort to prescribe one drug over another? If not, why pay doctors in any other way other than reimbursement at cost plus a flat fee?
This next one is controversial. I include it here because it is a) on the Wikipedia list, and b) though the degree to which it is a problem is disputed, it’s an interesting theoretical problem to work through.
“The ‘welfare trap’ theory describes perverse incentives that occur when money earned through part-time or minimum-wage employment results in a reduction in state benefits that would have been greater than the amount earned, thereby creating a barrier to low-income workers re-entering the workforce.. The welfare trap theory's accuracy is disputed, and some studies have shown the poor individuals who are given money tend to spend it on necessities, and continue working.”
If we take this to be purely an intellectual or theoretical problem, the issue is that the gradient is discreet and that the income limits are “hardcoded.” In other words, there is a large gap between desirable options. If there is no avoiding an incentive gradient, making said gradient smooth as opposed to a step function would probably be step in the right direction.5
C. Temporal Incentive Gradients
Some of the other entries in the Wikipedia list of Perverse Incentives aren’t so much the product of an uneven incentive surface in the present as they are the product of uneven incentive gradients through time. For this type of perverse incentive, the solution is adjacent to The Cobra Effect solution in that it involves similarity. But the thing to pay attention to isn’t so much similarity between cobras or rats as much as it is similarity to past behavior.
For example:
“In 2002 British officials in Afghanistan offered Afghan poppy farmers $700 an acre in return for destroying their poppy crops. This ignited a poppy-growing frenzy among Afghan farmers who sought to plant as many poppies as they could in order to collect payouts from the cash-for-poppies program. Some farmers harvested the sap before destroying the plants, getting paid twice for the same crop.”
It is stupid to do something like this without having some way of comparing the similarity of current acreage to past acreage. If you wanted to do it better, the first thing you’d want to do is set up a record-keeping system for a few years without telling anyone what you are doing. Then, you could announce that the payout will be capped to a per-farmer extrapolation based on the average year-to-year rate of growth of poppy acreage applied to each farmer. I don’t know if this the only solution, but it’s the first one that springs to mind.
The same solution of tracking historical use and calculating similarity can be applied to Northern Ireland’s “Renewable Heat Incentive Scandal,” which has also been called “Cash for Ash”:
“Renewable Heat Incentive scandal — Introduced by the devolved government in Northern Ireland, the Renewable Heat Incentive (RHI) was a 20-year scheme intended to encourage businesses to reduce energy usage and promote switching to green sources. However, the subsidy for the renewable energy was greater than its cost, which allowed businesses to make a profit by switching to green sources and then increasing their energy use rather than reducing it. In some cases, an income was obtained simply by heating empty buildings…”
It can also be applied to:
“In 2005 the UN Intergovernmental Panel on Climate Change began an incentive scheme to cut down on greenhouse gases. Companies disposing of polluting gases were rewarded with carbon credits, which could eventually get converted into cash. The program set prices according to how serious the damage the pollutant could do to the environment was, and attributed one of the highest bounties for destroying HFC-23, a byproduct of a common coolant, HCFC-22. As a result, companies began to produce more of this coolant in order to destroy more of the byproduct waste gas, and collect millions of dollars in credits. This increased production also caused the price of the refrigerant to decrease significantly, motivating refrigeration companies to continue using it, despite the adverse environmental effects.”
As well as:
“[The] Wells Fargo account fraud scandal — Intending to increase the number of accounts sold, Wells Fargo in 2016 introduced and imposed overly ambitious sales goals to be met by their employees. In result, facing the threat of losing their careers if these quotas were not met, some employees began to open large numbers of unauthorized accounts.”
In each case, one can solve the temporal incentive gradient by looking at the respective per-person historical amount of poppy acres, energy usage, HFC-23 consumption, and bank-accounts. If any one person shows even a two fold increase, something is wrong, and the case should be inspected by an adjudicator (who is potentially given an incentive that is counter to the original one?).
Also, in addition to tracking historic similarity – and my proposed solution here assumes there is a verifiable trail – you’d also want to have steep fines and jail time for misreporting, falsifying records, lying about audits, etc.
D. Incentives in Need of Bidirectional Pressure
As I mentioned above, the solution via similarity works because it introduces a second incentive that is somewhat in tension with the original incentive. This can be generalized to tackling yet another category of Perverse Incentives: instead of problematically rewarding one type of behavior want, reward multiple types of behavior that a) can’t be done simultaneously and b) are what you want.
For example:
“Funding fire departments by the number of fire calls that are made is intended to reward fire departments that do the most work. However, it may discourage them from fire-prevention activities, leading to an increase in actual fires.”
My proposal for a solution to this type of perverse incentive is to reward two beneficial activities that can’t be done at the same time. If you are going to reward fire departments by the number of “fire calls that are made,” you should also reward fire departments – by the same amount – for the number of fire-prevention educational sessions they offer (for example.)
This next one is wickedly difficult:
“Paying medical professionals and reimbursing insured patients for treatment but not prevention encourages medical conditions to be ignored until treatment is required. Moreover, paying only for treatment effectively discourages prevention (which would reduce the demand for future treatments and would also improve quality of life for the patient). Payment for treatment also generates a perverse incentive for unnecessary treatments that could be harmful—for example, in the form of side effects of drugs and surgery. These side effects themselves can then trigger a demand for further treatments.”
I am not a healthcare economist, and paying doctors in a capitalistic system is a very tricky problem. I will note, however, that the money we would be paying for preventative care instead goes to health insurance. Then again, what if, like the firefighting example, doctors were reimbursed for preventative care at the same approximate rate as what they’d make for providing treatment? It seems to me that part of the issue here is definitional and stems from the fact that “treatments” are more concrete than “prevention” — it is easier to name instances of the former than it is the latter.
“In late 2004, Fannie Mae was subject to an investigation regarding its accounting practices. It was discovered that, by providing company executives with bonuses for reporting higher earnings, executives at Fannie Mae and other large corporations were encouraged to artificially inflate earnings statements and make decisions targeting short-term gains at the expense of long-term profitability.”
Though I don’t have the ideal solution here, this one makes you wonder if something like a partially deferred salary / bonus system / pension plan that paid out over the long term in proportion to long term profitability might provide the “bidirectional pressure” necessary to prevent the original perverse incentive. It would need to be coupled with having fixed leadership terms. If you can only be the executive for seven years, it’s a good bet that you’ll still be interested in the long-term incentive after you’ve left the office.
Phew, that was a long list. There are a few other edge-cases on the Wikipedia list, but this is already a lengthy post, and so I’m going to stop things here.
One high-level conclusion from looking at these mini case studies is that in addition to the Government Accountability Office, perhaps all legislation should be analyzed by a “Cynical Office of Incentive Identification” or something similar.
You’d also think there’d be a book version of this post that was composed of case studies not just from government but from business and the non-profit world. And maybe, just maybe, if that were published, it could be “required reading” before going into lawmaking. Now isn’t that a utopian idea?
Also, because I’ve grouped the list by subcategory, it represent a re-ordering from the entries in the Wikipedia list. In addition, note that I’ve removed the citational links from the quotes; as the last thing this post needs is a second set of footnotes.
It is with reluctance that I quote from an article from the “Foundation for Economic Education.” I appreciate having many Cobra Effect links in one place, but if you read the article, it is contains language along the lines of “private enterprise doesn’t have this problem because businesses, ‘disciplined by profit and loss,’ learn from their mistakes.” This is under the heading “The Wrong Incentives.” Setting aside the fact that the author’s description of the exact nature of the Delhi cobra bounty appears to be factually incorrect, this line of reasoning is disappointing and troubling.
While it is obviously true that businesses have a greater incentive to spend money in a way they think is wise when compared with governments, the only reason for this is because businesses use profit as a proxy to measure the quality of their decisions. Moreover, I object to the ludicrously obtuse, disingenuously vacuous, and downright stupid implication that perverse incentives are worse in governments than in private enterprises. How can you earnestly say that “firms are disciplined by the lure of profit and the threat of loss” and not say anything else? Firms, by definition, are profit-seeking; they don’t need discipline to do what they are defined to do. They are actually disciplined by regulations that, for example, prevent anti-freeze from being mixed with wine. What would you rather have, Perdue Pharma and the opioid crisis, or a little waste and inefficiency for pest-control bounties?
I’m all for education about incentives. In fact, we probably don’t do it enough. But if you want to have any credibility at all, you’ve got to point it out wherever it occurs, and not use the issue as a prop to further “corporations never harmed anyone” propaganda. /end rant.
I looked this up, and no joke, the place is actually called “Banana Shire,” which I like to think has based its entire economy in selling tropical smoothies, bananas flambé, and cream pies to hobbits.
I’m fond of thinking that progress happens in three percent chunks, and thus one always needs a combination of things / policies / practices to reach a goal. You heard it here first: success happens three percent at a time.
I sometimes wonder if policy would be better if it were written by people who had been taught how to think algorithmically and how to code. Hard coding a single numerical cutoff is a somewhat lazy move, especially when compared with using a parameterized function. For example, parameterized functions could be used to have certain policies set to automatically track local inflation within limits, establish local minimum wages, etc. In fact, this makes me further wonder if we’ll one day have “policy engineers” responsible for translating lawmaker intentions into more precise and better policy.