How the war on drunk driving was won

My grandfather, an Essex doctor in general practice, was a pillar of his local community. An anxious and conscientious man, he was the kind of doctor who would visit unwell patients late into the night just to check their cases had not deteriorated into emergencies. His wife, my grandmother, was a magistrate.

He also would drink and drive on occasion, whether coming back from the opera or the golf club. After one night at Glyndebourne, he recalled watching other members of the audience, including other doctors, judges, professors, and journalists, stagger out to their cars, keys in hand, before clambering into the driver’s seats. I am guessing that his well-to-do friends seldom queried his capacity to drive.

My grandfather reduced, although did not quite eliminate, his drunk driving following one occasion when a police officer spotted him driving far too slowly home from one such event. He was pulled over and breathalyzed. He was certain he would be caught, but either the breathalyzer failed or he had metabolized just enough alcohol to be under the limit.

My grandfather dedicated his days (and many nights) to saving lives. But he also risked lives on his nights out.

Viewed from the 1960s it might have seemed like ending drunk driving would be impossible. Even in the 1980s, the movement seemed unlikely to succeed and many researchers questioned whether it constituted a social problem at all.

Yet things did change: in 1980, 1,450 fatalities were attributed to drunk driving accidents in the UK. In 2020, there were 220. Road deaths in general declined much more slowly, from around 6,000 in 1980 to 1,500 in 2020. Drunk driving fatalities dropped overall and as a percentage of all road deaths.

The same thing happened in the United States, though not to quite the same extent. In 1980, there were around 28,000 drunk driving deaths there, while in 2020, there were 11,654. Despite this progress, drunk driving remains a substantial public threat, comparable in scale to homicide (of which in 2020 there were 594 in Britain and 21,570 in America).

Of course, many things have happened in the last 40 years that contributed to this reduction. Vehicles are better designed to prioritize life preservation in the event of a collision. Emergency hospital care has improved so that people are more likely to survive serious injuries from car accidents. But, above all, driving while drunk has become stigmatized.

This stigma didn’t come from nowhere. Governments across the Western world, along with many civil society organizations, engaged in hard-hitting education campaigns about the risks of drunk driving. And they didn’t just talk. Tens of thousands of people faced criminal sanctions, and many were even put in jail.

Two underappreciated ideas stick out from this experience. First, deterrence works: incentives matter to offenders much more than many scholars found initially plausible. Second, the long-run impact that successful criminal justice interventions have is not primarily in rehabilitation, incapacitation, or even deterrence, but in altering the social norms around acceptable behavior.

The techniques used to drive down drunk driving help illustrate how many crimes were brought under control in the past, and how more might be in the future. Deterrence can jump-start a virtuous cycle: it reduces the rate of a crime, making the crime less normal. Less normal crimes bring higher community sanctions, or ‘moral costs’. This further reduces the rate at which people commit the crime, making it less normal still. This cycle cannot eliminate a crime entirely; some persistent offenders will continue to commit crimes unless they are incapacitated or rehabilitated successfully. But by using deterrence to create moral costs, we can make offenses rare, and focus enforcement efforts on the small group of those who commit crimes in spite of their costs.

Classical deterrence works

The classic economic explanation for how criminal justice interventions reduce crime is deterrence: criminals fear being caught, charged, and put in jail, or fined, and thus commit fewer crimes.

The theory of deterrence presumes potential offenders are rational actors, or at least make sensible cost-benefit calculations. It suggests potential offenders are balancing the expected costs of crime – how likely they are to be caught, multiplied by how badly they will be punished – against the benefits of offending.

According to the model, we can deter crime by raising the costs of offending, either by increasing detection – the likelihood of getting caught – or increasing the punishment for those caught. If deterrence works as intended, it is preventative: fewer people commit crimes and society rarely has to carry out punishments. Most people will not offend at all out of fear of the consequences, while habitual offenders will at least reduce their level of offending since there will be fewer instances where the crime is worth the risk.

One striking implication of this model is that you can make up for a relatively low rate of catching criminals by increasing the severity of punishment when they are caught, thus increasing the expected cost of crime. As we will see in a moment, this will not always work as theorized: criminals are not always aware of slightly longer minimum or maximum sentences, so simply increasing the probability a criminal gets caught is likely to work better.

Many criminologists take a dim view of deterrence theory: they claim most people are nonresponsive to both the severity of punishment and their chances of getting caught. Noncriminals, the bulk of the population, never even consider committing crimes, not because of a rational calculation but because they consider such acts to be morally wrong, the view goes. Those who do commit crimes don’t make rational calculations either, but instead follow what’s accepted in their community.

When looking beyond the individual to communities and countries, they argue, very little of the remarkably wide variance in crime victimization observed between countries, cities, and neighborhoods is explained by more severe punishment and higher rates of apprehension. Botswana retains the death penalty, and last executed a murderer (by hanging) in 2021, but has a homicide rate of over 10 per 100,000 – high by global standards. Sweden is more forgiving of murderers – its last execution for anything was in 1910 – but has a homicide rate of only around one per 100,000.

Instead of simply hiking penalties and putting more cops on the block, these deterrence skeptics suggest, we need to address social inequalities and enact real social change. Otherwise, we fail to prevent crime and enact senselessly harsh yet ineffective punishments on those who happen to get caught. For many reformers, prison especially is a blunt and cruel instrument that should be abolished and replaced with something else as a matter of urgency. Others argue that crime is only punished because it is different from normal behavior, and deterrence is an excuse, rather than a true empirical relationship.

To be fair to the deterrence skeptics, they are correct that the empirical evidence around the rational actor model is mixed. Some studies suggest deterrence works a bit here and there, especially when detection is increased, while others find the impact to be limited or even to have counterproductive side effects: a prison sentence can end up increasing offending on release.

While the empirical literature is mixed, some of the best well-documented studies of criminal justice do demonstrate that incentives effectively reduce crime in line with what deterrence would predict. The evidence is strongest for increases in the likelihood of detection. Two studies are particularly helpful, one on apprehension and the other on punishment.

The first study, from Mirko Draca, Stephen Machin and Robert Witt, has a particularly clear claim to identifying a causal connection between greater police presence – and by proxy, a greater probability of apprehension – on city streets and reduction in crime. The Metropolitan Police suddenly increased police patrols near London Tube stations in five central London boroughs in response to the infamous 7/7 suicide bombings in 2005, in a move called Operation Theseus. The policing intervention was not intended by policymakers to address street crime as such but to prevent further terrorism. It was not targeted selectively to achieve a crime reduction. Thus, it is what economists call a natural experiment, and it is useful for determining typical offender responses to increased police presence as opposed to the smaller-scale targeted interventions known as hot spot policing (for which the evidence is also strong).

In total, police activity went up 30 percent in central London. In the places where activity went up most, crime fell. For the six weeks of the policy, the affected areas saw about 12 percent lower street crime, or four fewer crimes per 10,000 people, while crimes that happen off the street, like burglary, saw no change. Because different affected locations got different increases in police presence, they could work out how much each extra dose of policing helped: they found that each ten percent increase in police activity caused a three to four percent decrease in crime.

The crimes prevented due to police presence in one area didn’t appear to cause more crime to take place at another time or place. There was simply less crime, meaning safer streets, presumably because offenders were worried about getting caught by all the police around.

The other key study, by Brian Bell, Laura Jaitman, and Stephen Machin – this one on severity of punishment – examined the English riots of 2011 that started, and were most intense, in London. Rioters looted shops, burned cars, and destroyed commercial and public buildings for three days before being quelled by the police. Between 13,000 and 15,000 people were involved in the riots in some form, and more than 5,000 offenses were recorded.

The police were initially overwhelmed, but the rioters left plenty of evidence, which later allowed several thousand rioters to be arrested and convicted of offenses including burglary, criminal damage, violent disorder, robbery, and theft. Although this was not expected in advance, the judiciary, in practice Magistrates’ Courts (a quicker, jury-less lower court in England and Wales usually used for more minor offenses), decided to treat participation in the riot as an aggravating factor. As a result, people convicted of riot-related offenses were three times more likely to be sentenced to prison than those who committed similar offenses unrelated to the riots. On top of that, prison sentences were around 2.5 times longer than comparable non-riot crimes. Around 3,000 people received a sanction related to their activity in the riots.

Criminals were listening. The riots had been heavily concentrated geographically. During the riots, the main riot crimes (burglary, criminal damage, and violence against the person) went up 57 percent in 88 of the 4,760 sub-wards that make up London, and were flat in the others. But afterward, riot crimes fell three percent citywide, while non-riot crimes actually rose. Criminals appeared to be rationally shifting from activities they expected to be punished more seriously to ones they did not. Because the crime drop wasn’t concentrated to the most affected areas, from which the incapacitated criminals were drawn, and because other crimes rose, it probably wasn’t coming from incapacitating the offenders: rather potential offenders unrelated to the August 2011 riots were deterred from committing crimes that they now perceived to be associated with more severe sanctions.

Deterrence has its academic skeptics, but few in the real world. Criminals do seem to commit less crime when they are more likely to get caught. They do seem to be somewhat responsive to the severity of punishment as well, even if they are less sensitive to punishment than apprehension.

Can drunk driving be deterred?

Even if deterrence affects how many people commit robbery, drunk driving seems like a difficult case for it.

Many drunk drivers are thought to be alcoholics. That would suggest they are not paradigmatically good long-term decision-makers. This has led some researchers to claim, just as they do for other criminals, that prospective sanctions will not work on drunk drivers because of fundamental personality differences between them and the general public. What’s more, drunk driving has a very low detection rate. Can people really be worrying about getting caught when so few who drive under the influence are detected at all?

In fact, these researchers are too pessimistic. Drunk driving was deterred and reduced much like other crimes.

While being drunk in charge of a road vehicle had been illegal since the outset of motoring in the United Kingdom, it was only in 1967 that a formal legal limit (80 milligrams of alcohol per 100 milliliters of blood – or 35 micrograms per 100 milliliters of breath) was introduced.

Because of the new feasibility of enforcement, plus the emergence of a grassroots movement opposing it, many governments started campaigns to reduce drunk driving. The widespread deployment of mobile breathalyzers made detecting drunk driving more feasible and immediate, with British police carrying out 800,000 tests in 1998 when they peaked, up from 100,000 in 1971.

As drunk driving prevention became a greater policing priority, the sanctions for drunk driving, and especially causing death by dangerous driving, were increased substantially, eventually including extended prison sentences. For example, the UK’s 1988 Road Traffic Offenders Act had a maximum prison sentence of five years for causing death by dangerous driving. In 1993, it was raised to ten years. From 2022, the maximum penalty was life imprisonment. During this period, new offenses for causing death by careless driving (for example, by inattention to other road users or by tailgating) and causing serious injury by dangerous driving were introduced. Prosecutors charge people for these offenses with some regularity (between 100 and 250 prosecutions a year in Britain for most offense types).

One thing is immediately obvious, given the decline of drunk driving: while some significant subgroup of drunk drivers were alcoholics, most were not. A great many people who previously considered drunk driving, like my grandfather, became far more likely to refrain. This suggests an important conclusion: even if some offenders are truly irrational actors, many are surprisingly responsive to incentives even when their reasoning is impaired.

We know that fatalities from drunk driving dropped during the period that breathalyzer testing ramped up and sanctions increased.¹ But was this causation or coincidence?

In the United Kingdom, there are not many studies with research designs that credibly isolate the impact of these changes. Fortunately, parts of the United States adopted a similar range of policies over the same period.

As with other deterrence studies, the literature on drunk driving does suggest that increased detection lowered rates of drunk driving. It is slightly more skeptical about increased severity, though the totality of evidence does support the role of harsher punishments.

One study examined times when police highway patrols were reduced due to budget cuts. When there were fewer officers on the road, there was a closely associated increase in road fatalities. This suggests that police officers – and the risk of detection brought with them – not only deterred drunk driving but all sorts of reckless driving behaviors.

In terms of severity of punishment, the academic evidence on automatic license suspensions for drunk drivers is mixed: it seems to have worked in some states but not others. But one study from Benjamin Hansen makes it seem almost certain that sanction severity reduces drunk driving – in addition to the work detection is doing.

Washington State police kept over half a million individual records of drunk driving cases between 1995 and 2011. At that time, Washington had the same standards as Britain for drunk driving – 0.08 percent in the blood, or 80 milligrams per 100 milliliters.

Under this sanction regime, if you are caught with a blood alcohol concentration just above a threshold, you will be punished much more harshly than someone who was caught driving just under a threshold. Since you can’t target an exact percentage point of alcohol in the blood, those who were measured at .08 would be similar to those with a .079 level, except for the fact that they would be punished. It turned out those sanctioned for drunk driving did have a lower chance of being caught drunk driving again – ten percent for those sanctioned versus 12 percent for those who were under the limit, or about a sixth lower. This represents a substantial reduction especially when considering this is a policy regime applied to every driver who drinks in Washington State, rather than a subgroup of offenders, and that many of the costs (such as a fine and driving license suspension) fall on the offender. It compares favorably to rehabilitation programs focused on repeat offenders that can show larger reductions but for smaller populations.

There were other ways to raise the severity of punishment besides jail time. Prosecutors and judges used a wide range of punishments – fines, imprisonment, and driving bans – as well as rehabilitative requirements such as compulsory attendance of Alcoholics Anonymous meetings. One powerful intermediate sanction was requiring an offender to fit their car with an interlock device (at their own expense) that prevents the engine from starting unless the driver passes an automated breathalyzer test. Encouragingly, in this study those who were banned from driving continued to commit fewer driving offenses after their driving ban elapsed.

All of these estimates for deterrence are almost certainly a lower bound. For example, the Washington study only compares drivers who had lucky near misses with those who were caught driving just above the limit. Many drivers, similar to my grandfather, might have altered their behavior after experiencing near misses where they were stopped by police and marginally passed a breathalyzer test. The Washington study measures the effectiveness of the sanctions by how much more it reduced recidivism than a near miss – but if near misses also reduce drunk driving then the impact was undercounted. On top of that, many drivers, deterred by the prospect of being tested and sanctioned, could have avoided driving anywhere close to the limit at all.

Contrary to the hard-core skeptics, deterrence is important even for drunk driving. At the same time, deterrence cannot actually explain most of the decline in drunk driving: drunk driving fell by five times over this period, vastly more than what would be expected from studies of deterrence. This is where the skeptical criminologists have a point: social norms are important too.

Moral costs

When skeptical criminologists write about the role of norms, it is usually in contrast to the incentives faced by rational actors. In fact, they are deeply intertwined.

Social norms function as a type of cost – call them moral costs – which can be treated within a rational actor model. And these norms are created, in part, by the severity of punishment for a crime.

When we incorporate norms as moral costs into the cost-benefit analysis that potential offenders theoretically perform when considering whether to commit crime, it is clear that, unlike standard legal costs, these moral costs are experienced differently by different people. A potential offender in a low-crime milieu might view them as quite severe, as it would be a significant break from their community. A potential offender in a high-crime milieu might see them as quite low, as they would simply be acting like everyone else.

Still, how do we explain these moral costs? One aspect is whether committing the crime is normal in a community. Individuals want to know how other people will (or would hypothetically) judge their behavior. If crime is normal in their community, an offender will correctly judge that people will still think they are normal if they commit the occasional crime.

Interestingly, criminal justice can influence this perception of normality. Being caught and processed, even without any further penalty, sends a powerful message to many people: this is not normal. The process of police observation and detection helps to highlight that the community takes bad conduct seriously without necessarily requiring a heavy sanction.

Publicizing punishments can undergird this process. Governments frequently disseminate anti–drunk driving messages in print, on television, on roadside signs, and increasingly via text and social media, highlighting the health risks to oneself and others, as well as the legal consequences if caught. One reason why they might not always work is that media exposure can unintentionally raise the salience of drunk driving, making it perversely seem more normal. This is precisely what you want to avoid when dealing with occasional suspects who are more worried about what is normal than what is harmful. Nevertheless, anti–drunk driving campaigns that focus both on enhanced enforcement and the legal consequences of getting caught have been found to reduce drunk driving in a variety of circumstances.

There may also be scope to use the promotion of certain social norms to reduce drunk driving, not just the deterrence itself. One quasi-experimental study looked at a mass media campaign that took place from 2002 to 2003 aimed at young adults in Montana, premised on precisely telling drivers what was normal behavior rather than what was harmful. The study split Montana up into three regions: the western part was subject to a TV and radio campaign supplemented by cinema, newspaper, and poster advertising. A representative slogan was ‘Most Montana young adults (4 out of 5) don’t drink and drive’. The eastern region was treated as a control, surveyed and studied without any intervention, while the central part was treated as a buffer zone where the media intervention was expected to produce some spillover.

Critically, the information disseminated through the campaign was an accurate summary of what Montanans reported doing. Drunk driving is deviant behavior in Montana but young people were not aware of quite how deviant it is. A series of random phone surveys before, during, and after the intervention measured the impact of the campaign. The advertising caused young people in the intervention counties in western Montana to update their perception relative to the control counties.

A significant shift was to norms around having a designated driver. Over 40 percent of young Montanans surveyed said they personally always had a designated sober driver when planning to go out drinking while only 30 percent believed that others engaged in this practice. After the intervention, the treated group estimated that just under 40 percent of their peer group, young Montanans, used designated drivers while the control group had remained roughly the same. In other words, the treated group’s estimate of designated driver use was higher and closer to the reality based on the same survey’s self-reported use. The treated group was significantly more likely to report using a designated driver themselves than the control group. The study also found a small but statistically significant reduction in road fatalities related to alcohol use compared to the control region.

This study and others like it suggest that the public perception of norms can be inaccurate and media interventions can play a role in updating them. Other advertising campaigns around the world at the same time emphasized the shamefulness of drunk driving, and are widely seen as having shifted social norms around it. Nevertheless, these public campaigns worked because they were backed by a credible threat of detection and sanction.

This need for publicity to create common knowledge of social stigma helps explain why detection is commonly found to be a stronger mechanism of deterrence than severity of punishment: more instances of detection can raise the moral cost of a crime more than a single long sentence. But it also helps explain why severe sentences can work too in some contexts: a more severe sentence reflects a more serious social judgment.

Taken as a whole, the evidence suggests an interesting model: moral costs add to the overall costs of committing a crime, but these moral costs are, at least in part, driven by the severity of punishment.

Conditional offenders: more common than you think

Are other criminals like drunk drivers? It depends on the crime.

We can call people who may or may not commit crimes based on the degree of moral costs conditional offenders. Where crime is low, they will sense that offending is wrong, and most crimes will be committed by hard cases who are resistant to social considerations. However, in a high-crime environment, the much larger group of conditional offenders will commit a great many offenses when added together. Because crime is normalized, and indeed they face the prospect of being victimized themselves, they will not feel as much guilt in participating.

This may have been the secret sauce in the campaign against drunk driving: most offenders were actually conditional offenders, who normally did not commit crimes but saw little moral sanction against drunk driving. Detection increased, which played some role in discouraging drunk driving. But most important was making society’s dislike of the crime credible with penalties and common knowledge created by stories of those punishments and bolstered by advertising.

People came to understand that to drive drunk would be to deviate from group norms. This created the virtuous cycle where drunk driving became increasingly infrequent, and therefore increasingly deviant (i.e., decreasingly normal). Today, most of us wouldn’t dare to drive drunk – but we might not have felt the same way in 1970.

Can this experience be generalized? An objection to this story is that it’s plausible for something like drunk driving, but not so much for other major crime categories like violence or theft. At least concerning theft, the logic can be readily extended. My record is not spotless: I illegally downloaded and shared an enormous volume of copyrighted music, films, and TV shows in my youth before platforms like Spotify and Netflix rendered on-demand listening and watching cheap and convenient. People I know still see torrenting as normal, and acceptable, and would be very unlikely to turn their nose up at it. Here I am a conditional offender. I, and many others, were undeterred by the famously aggressive and ubiquitous ‘You wouldn’t steal a purse’ anti-piracy notices disseminated during my youth. This was probably because mere consumers of pirated material were seldom sanctioned and this was well-known. Without enforcement suggesting those crimes were truly morally blameworthy, I felt little moral cost.

In fact, many cases of declines in crime are similar to the drunk driving story: increased enforcement and punishment created common knowledge that society is against some socially harmful activity, and thereby driven most conditional offenders out.

Approaches to tackling litter illustrate both the potential value of creating a common norm but also its limitations without enforcement. Littering was once thought to be widely permissible. In fact, it was widely legal. That began to change in the 1950s, as communities around the world united to oppose litter. In 1971, punishments in the UK got much stricter with the Dangerous Litter Act, which raised maximum fines from £10 to £100, almost £2,000 now. Common knowledge of the punishment driven by mass campaigns has made it clear that society condemns littering. Did this make littering more rare? It is hard to assess from a counterfactual, especially given the growth of key sources of litter such as fast-food outlets, but litter is still considered a significant problem in much of the UK. Rather than adapting their behavior to the public norm, many people appear to have adopted different interpretations of what constitutes littering. More recent campaigns have tried to tackle this by highlighting that so-called careful littering is still a crime.

This experience contrasts with Singapore, which has gone from suffering waste as a major public health hazard in the 1960s to remarkably low levels of litter today, and substantially less than any major city in the UK. Singapore, of course, has famously harsh criminal penalties for offenses such as drug trafficking. In the case of littering, however, Singapore is not quite so far out of line with the UK in terms of punishment severity. A litterer faces a S$300 (£176) fine for a first offense (and occasionally just a warning), compared to an £80 fixed penalty notice in the UK. The real difference is enforcement. Despite facing little visible litter, Singapore’s authorities issued around 20,000 penalty tickets last year (between 3 and 4 fines for every 1,000 Singaporean residents). Singapore’s relatively highly law-abiding population still faces constant reminders of the social sanction for littering. By contrast, 91,000 fines were issued in England and Wales (1.5 for every 1,000 residents) in 2021/2022. Enforcement is patchy, with some local authorities barely enforcing this area of the law at all.

Without enforcement, conditional offenders are able to justify their occasional littering to themselves if nobody else on the basis that it is harmless, unintended, or caused by an inconvenient lack of rubbish bins. In Singapore, there is less opportunity for the conditional offender to hide from the moral cost.

Remarkably, even murder may be like this. According to the best available evidence, homicide rates were 50 or 100 times higher in the 1300s, across Europe, than they are today. It was simply considered normal – even admirable – to prosecute feuds and escalate drunken fights over honor into potential killings. These norms steadily changed, in substantial part because governments gained the capacity and desire to punish perpetrators. Blood feuds and brawls turned into ritualized duels, and eventually self-control became normal and impulsiveness deviant. It is widely argued that the honor culture of the US South helps to explain why murder rates are higher there today.

When you know how to look, you start finding conditional offenders everywhere. I was surprised to find out that two well-to-do friends, one now an economist and the other a philosopher, stole food on occasion as undergraduates. A classic method is ‘mistakenly’ failing to scan concealable items at a self-checkout (one study from Emmeline Taylor suggests that up to a third of self-checkout users have committed theft). This might be justified out of necessity (stealing food when they were young or poor) or in pursuit of a bit of excitement and rebellion. Since the detection of shoplifting is low and the detection of theft at self-checkouts is likely even lower, the moral sanctions against committing that sort of theft are weak.

Five percent of the potential UK tax take, and about ten percent of the US equivalent, is lost to tax evasion, a huge fraction of which is driven by paying cash and not declaring the transaction. Even people who do this, who would never think of themselves as criminals, become criminals under the conditions of low enforcement and little moral cost.

But if the story of drunk driving is instructive, there might be a similar path forward here. Higher penalties and better enforcement can make a behavior less normal, and by becoming less normal, and more deviant, the actions carry greater moral costs for nearly everyone.

Criminals carry and spread bad social norms

I have introduced the notion of conditional offenders and explained how they are responsive to social norms when considering whether to commit an offense. What about their more well-known relatives, the people who commit crimes like murder, rape, and assault notwithstanding the strong condemnation associated with these crimes?

We can call these people persistent offenders, in that they commit offenses with some frequency despite well-established social norms. They do not feel the burden of ‘moral costs’ to the same extent as the public generally. Unsurprisingly, they tend to be well-known to police and can become the model of what a typical offender looks like for professionals dealing with crime.

Although small in number, they can play an outsize role in the burden of victimization placed on communities. The disproportionate influence of a small number of very prolific offenders on local experience of crime was powerfully illustrated in Leinster, Ireland, in 2021 when three members of a notorious burglary gang died in a car crash after fleeing police. Burglaries in the area they had habitually targeted halved in the following months. In the City of London, catching and sentencing a small organized gang reduced bike theft in the area by 90 percent.

When the UK government applied austerity measures to the criminal justice system during the 2010s, officials justified shaving time off prison sentences on the premise that each offender cost the public more while incapacitated in prison than finishing their sentences in the community. In recent months, the UK government has told judges to delay imposing new custodial sentences and announced extensions of early-release schemes because prisons are full. However, it is easy to underestimate the cost of leaving criminals out in society partly because most crime is unmeasured, and also because of the damage prolific offenders do to norms.

How do we know most crime is unmeasured? The United Kingdom’s Ministry of Justice occasionally conducts self-report surveys of criminality among representative samples of current prisoners. In the most recent, 74 percent of offenders acknowledged they had committed an offense for which they had not been caught in the year prior to imprisonment. They had committed a median of three offenses in just four weeks before they were incarcerated (a reasonably high frequency in itself) but the mean was 44, implying that a subset of offenders go through sustained periods of committing low-level crime (such as shoplifting) at remorseless volumes. A vanishing minority of these offenses results in detections and convictions.

Similarly, a Home Office analysis of lifetime criminal careers found that the ratio between official criminal convictions and self-reported crime was 39 to 1 (for shoplifting, the ratio was nearly 77). Many offenders may cost the government more when incarcerated but cost the public more when they are in the community, even including how much prisons cost to the taxpayer.

You can think of flagrant crime as the opposite of deterrence: while deterrence makes a crime more rare, and therefore more morally condemned, flagrant crime makes a crime more common, and therefore less morally condemned. Each crime lowers the cost of doing further crimes for everyone, where each crime prevented raises it. Each dropped piece of litter and stolen candy bar makes you a little more likely to do the same.

Yet usually criminal justice policies treat offenders as isolated actors rather than carriers of antisocial norms. Or, to put it another way, crime, especially crime for which people are not being caught, processed, and punished, has negative externalities that are typically not taken into account. Such approaches neglect the high costs of a high-crime equilibrium and the difficulty of escaping one once established.

This phenomenon can arguably be seen in a microcosm in the case of the aforementioned London riots. The summer 2011 riots were initially triggered by a protest in North London around the police killing of a local drug dealer. Normally such an event would fizzle out, but every now and then it can ignite something much larger. In this case, the riots rapidly spread to other boroughs of London and then on to other cities across the UK.

Those who participated in the looting were less likely to have prior convictions than typical. The media reported startling cases of young people with great prospects in sports or higher education getting caught up in the action and potentially ruining their futures. One prominent case was Chelsea Ives, a former Olympic ambassador who was sentenced to prison for two years at the age of 18 for looting a phone shop and vandalizing a police car. By her own account, her behavior was out of character and indeed it likely was. Such a person would generally not have put her future at risk were it not for the outbreak of riots.

This is evidence that there are a surprisingly large number of conditional law abiders in the general population; people, especially young people, who will normally behave well (or at least well enough to stay under the radar of the formal criminal justice system) but will commit crimes of some severity if given collective permission or peer encouragement.

The moral economy of sanctions

Social norms can also explain why different aspects of deterrence (certainty of apprehension, celerity – i.e., swiftness – of sanction, and severity of punishment) cannot always be traded off against one another as the standard economic model would predict. As we’ve seen, apprehension seems to work better.

Increasing severity alone is impractical as well as ineffective. Social norms also help determine what people think is an appropriate punishment. A community will resist the imposition of harsh sanctions for conduct that is considered relatively ‘normal’ even if it is widely acknowledged to be harmful. In some cases, juries are unwilling to convict and victims are unwilling to bear witness if they are concerned the resulting punishment for an offender will be too high.

The starkest example of this jury nullification comes from a historical study comparing eighteenth- and nineteenth-century juries sitting in London’s Old Bailey, England and Wales’s Central Criminal Court. Over the period, the death penalty was progressively abolished for most of the 220 offenses it covered, starting with fraud in 1813 and ending with arson in 1856, leaving only murder, treason, ship burning, espionage, piracy, and various military offenses. The study found that death penalty abolition for a given offense was associated with an average 16 percentage point increase in the likelihood of conviction. Jurors were not willing to hang the accused for offenses like stealing a handkerchief, and when a conviction obliged them to, they refused to convict at all.

A smaller but still significant increase in convictions was found when the British government was forced to halt transportation to colonies (after losing the American Revolutionary War) and introduce less punitive prison sentences. If juries (and the public more generally) find severe punishment repugnant, then increased detection combined with relatively minor sanctions (such as fines) could be more likely to push us toward a low-crime equilibrium than occasional harsh sanctions.

Observe the success of drunk driving reduction through this lens. In a society where a large proportion of people, including doctors, judges, and even police, drinks and drives, then the public and criminal justice system will struggle to impose long prison sentences when people accidentally kill other road users. But such a society will be much more amenable to the rolling out of breathalyzer tests that sanction the same individuals before their reckless behavior leads to casualties. A stinging fine or temporary driving license ban is easier for courts to apply with public support.

Those sanctions then reduce the prevalence of offending especially among conditional offenders who have a great deal more to lose. Since many of these conditional offenders are responsive to social norms, the overall impact on drunk driving will be larger than those directly experiencing a sanction.

Once conditional law abiders more consistently avoid drunk driving, a smaller, more resistant population is responsible for most drunk driving. Then, stiffer penalties both for drunk driving itself and causing death or injury while drunk in charge of a vehicle become more feasible from a prison cost perspective and in the eyes of the public. The expressive element of harsh sanctions, widely communicated, authorizes bystanders or third parties to prevent people from committing an offense, perhaps in the very interests of the prospective offender. This might take the form of a friend taking away someone’s keys, or a bouncer at the pub calling a customer a taxi.

Progress in crime reduction  

Crime imposes enormous costs on society. Yet academic criminologists and policymakers alike have despaired at the prospects of effective
crime prevention.

Robert Martinson, a famous American sociologist and former civil rights activist, after decades of studying prisoner rehabilitation, famously proclaimed, ‘nothing works’. More recently, Megan Stevenson, a prominent professor in both law and economics at the University of Virginia, systematically reviewed randomized controlled trials in criminal justice available in the academic literature, finding very limited instances of interventions focused on individual behavioral change reducing offending.

Social critics attribute crime to deep structural factors and early childhood experiences that criminal justice policy cannot influence. Meanwhile, policing and prisons are expensive, while their benefits are not immediately apparent, so they are easy prey for policymakers during cost-cutting parts of a political cycle.

The staggering reduction in drunk driving shows the social critics to be wrong. Criminal justice has a unique ability to do the very thing academic criminologists want: to change social norms by creating credible common knowledge about what society dislikes, thus raising the moral costs of doing crime.

Radical critics of punitive criminal justice are correct that life-changing sanctions like prison sentences are always expensive and sometimes cruel. Nevertheless, some punishment is necessary. Nearly everyone wants to see a lower prison population. Some want to see prison abolished altogether. But I seldom hear anyone specifically suggest, with the benefit of hindsight, that the penalties for causing death by drunk driving should be reduced.

We all benefit from being in a relatively lower drunk driving equilibrium, where drunk driving is highly stigmatized such that many would not even dream of doing it. The best way to have fewer people in prison is to have a generally low-crime society where norms of good behavior are substantially self-enforcing. In the meantime, criminal sanctions, including the prospect of prison, are part of a wider process that keeps respectable people, just like my grandfather, the celebrated GP, away from endangering others.