Information quality, punishments and income inequality🍋

When information is unreliable, punishment often misses its mark

Our laboratory experiment shows that when enforcement relies on noisy or unreliable signals, even well-intentioned punishment risks doing more harm than good – by reducing cooperation and widening inequality.

The spirit level

Richard Wilkinson and Kate Pickett published The Spirit Level in 2009. This book reframed inequality as a driver of public health and social dysfunction. Their argument was simple: countries that are more unequal fared worse across a range of social indicators, from mental health and education to social trust and violence.

Source: World Inequality Database

One striking observation from Wilkinson and Pickett was that more unequal countries treat lawbreakers more punitively. While their work was confined to OECD countries, we have replicated this finding globally. Income inequality is closely tied to higher rates of incarceration (per 100,000 people). The relationship is particularly pronounced for high-income nations.

The chart below presents the relationship between income inequality (on the x-axis) as measured by the Gini Index, the standard measure for inequality¹; and the incarceration rate (on the y-axis), which indicates the number of people imprisoned per 100,000 population.² Each dot represents a country.

Take the United States: it imprisons over 650 people per 100,000, the highest in the developed world. It also has one of the highest Gini coefficients among wealthy countries. South Africa, similarly, combines high inequality with a high prison rate of approximately 280 per 100,000. By contrast, countries with lower inequality see far fewer people behind bars; Norway and Germany, for example, have just 56 and 68 per 100,000, respectively.

Wilkinson and Pickett argued that inequality leads societies to punish more harshly. Our recent work suggests that the direction of causality may be in the opposite direction; punishment, especially when based on noisy information, can lead to greater inequality.

In recent times, New Zealand’s coalition government has been arguing strongly in favour of more punitive measures for lawbreakers, such as limiting how much judges are allowed to reduce sentences and reinstating “three strikes” sentencing rules. We argue that in the presence of unreliable information, punitiveness exacerbates inequalities.

Doubtful Sound, Fiordland. Dave Heatley

Who gets punished, and how?

Consider the following cases:

• Caroline Ellison, a 30-year-old Stanford graduate and daughter of eminent scholars and professors at MIT, was found guilty of helping her then boyfriend, Sam Bankman-Fried, misappropriate around US $8 billion of investor money through the cryptocurrency trading platform FTX. In October, she was sentenced to 2 years in a minimum-security prison. The decision was surprising to many, who expected Ellison to walk free in return for cooperating with the prosecution and testifying against Bankman-Fried.³

• By contrast, 21-year-old Ta’Kiya Young, an African American woman pregnant with her third child was shot dead in Columbus, Ohio after an altercation with police officers who had accused Young of shoplifting. As she tried to drive away, one officer shot and killed her.

• David Coulson was locked up for life for stealing $14 under one of the more extreme “tough on crime” laws in the US.

While blue-collar crimes are often more directly violent than white-collar ones, the harm caused by large-scale financial misconduct is far from negligible. Billions in lost investments can ruin thousands of lives and wipe out life savings, directly and via destabilising entire markets. The harm is real, even if no weapon is drawn.

In a perfect world, authorities would punish every crime accurately and proportionally. But real-world monitoring is uneven. Lower-level crimes are more visible, easier to prosecute, and harder to defend against. White-collar offences, by contrast, are often obscured by complexity, resources, and legal firepower. Our system doesn’t just punish wrongdoing; it punishes visibility.

A recent investigation by the New York Times highlights how offenders accumulate significant, often overwhelming debts during their time in the prison system. Bail fees, and a host of other charges such as court costs and administration fees, quickly add up.

Proponents of tougher sentencing argue that they target real offenders – true positives. Any practical criminal justice system will have both false positives (non-offenders get punished) and false negatives (offenders escape punishment). Reasonable people can take different views as to how to trade off these risks. Generally, however, harsher punishments make false positives more costly.

Simulating society in the lab

To explore how imperfect information affects cooperation and punishment, we designed a laboratory experiment as part of ongoing, unpublished research. The study incorporates a classic social dilemma problem, where there is tension between cooperating for the common good and free-riding to maximise individual self-interest. Cooperation is measured by how participants contribute to a public pot. Everyone contributing maximises the public good returns. But self-interest suggests zero contribution.

Our game was dynamic with earnings accumulating over multiple rounds – mimicking how real-world wealth builds over time. Just like in real life, where a small advantage early on can compound over time, early cooperation allowed participants to build up wealth that they could reinvest in later rounds. Groups that contributed generously from the outset saw their wealth grow. Those that didn’t began to fall behind. Before long, inequalities began to emerge, both within and between groups.

Then we added two features: punishment and imperfect monitoring.

Instead of modelling a state-sanctioned centralised punishment mechanism, we implemented decentralised peer-to-peer punishment. This situation fits well in the absence of a powerful state actor or where the social dilemma is of a localised nature limiting the degree of state intervention. Peer enforcement differs from judicial sanctioning, but it helps us isolate key mechanisms that underlie real-world dynamics: what happens when punishment decisions are made under imperfect information?

In some groups, participants could punish others at a small cost, a proxy for social sanctions. Participant contributions could be clearly seen by the potential punishers, i.e. information was perfect.

In other groups, monitoring was imperfect. Participants observe others’ contribution imperfectly, as the information about those contributions was displayed with introduced noise. Sometimes cooperative members appeared to hold back under these conditions, raising the risk of false positives.

Finally, in a third treatment, we combined punishment with imperfect information, a situation that reflects the reality of our justice system, where decisions are made under uncertainty.

What did we find?

We found that peer punishments under perfect monitoring did boost average contributions. When participants knew they could be sanctioned, they were more likely to cooperate. But the averages tell only part of the story. Beneath the surface, there was significant gap between groups. A few groups cooperated early, stayed on track, and built substantial wealth – enough to pull the average up. But many groups struggled. Cooperation broke down, and by the end, they were left with little to nothing.

Imperfect monitoring on its own had little effect. But when combined with punishment, the picture changed drastically. Contributions fell. Wealth declined. And inequality within groups rose sharply. The reason? When you can’t reliably tell who’s cooperating, people start punishing the wrong targets. This kind of “antisocial punishment” – directed at those who were actually playing fairly – became common. As misdirected punishment increased, cooperation collapsed.

While our experiment relied on peer punishment – not formal, state-led sanctions – it still points to a familiar risk: when information is unreliable, punishment often misses its mark. Even well-intentioned enforcement can deepen inequality and erode cooperative norms. As seen in real-world justice systems, imperfect monitoring and unequal visibility often distort who gets punished and how.

What does this mean for the justice system?

The prevailing view, advanced by Wilkinson and Pickett, is that unequal societies punish more harshly. But our findings suggest the relationship may also run the other way: when punishment is applied under conditions of imperfect information, it can itself become a source of inequality. Our experiment showed that when enforcement relies on noisy or unreliable signals, even well-intentioned punishment risks doing more harm than good – reducing cooperation and widening inequality.

Injustice often stems not from intent, but from imperfect monitoring and unequal visibility, which shape who is punished and how. Blanket calls for tougher sentencing tend to overlook these subtleties. Yet it is precisely such details that determine whether justice policies reduce harm – or quietly entrench it.

By Sanket Sen & Ananish Chaudhuri

Kaipo River, Fiordland National Park. Dave Heatley

1. The Gini Index ranges between 0 and 100; higher values denote higher income inequality.

2. Incarceration rates are a measure of prison population, most of whom are serving long sentences for offences committed a long time ago. This makes them a indicator of crime rates and sentencing harshness in the past, whereas the Gini Index of income inequality is a relatively recent snapshot. An extension to our work would be to unravel this.

3. Bankman-Fried received a sentence of 25 years imprisonment and ordered to pay US$11bn in repatriations.