011 min
Why a nudge needs a criterion of justice
In 2012 the United Kingdom changed how workers join a workplace pension. Until then, an employee had to sign up. From then on, employers enrolled every eligible employee automatically, and anyone who wanted out had to say so. A rule that applies when a person does nothing is a default. Setting the default is one job of choice architecture, the way a set of options is laid out for the person choosing.
The Department for Work and Pensions (DWP), the government department behind the scheme, measured what happened. The share of eligible employees in a workplace pension rose from 55 per cent in 2012 to 87 per cent in 2018. On average, the change worked.
An average does not say who gained, and three people can read the same figure in three ways. The first calls the change fair because every eligible employee was enrolled in the same way. The second calls it fair only if the lowest-paid workers gained the most. The third asks first who was never enrolled at all. Each person applies a different standard to the same numbers.
A standard of this kind is a criterion of justice. A team should pick it before it reads the results. If it picks afterwards, it can choose whichever standard flatters the outcome.
022 min
What a criterion of justice asks
The three readers of the pension figures were asking one question in different ways: who should get how much of a benefit, and on what ground? The criterion is the ground.
Philosophers split the problem in two. The formal part says that similar cases should be treated in similar ways. Nobody disputes it, but it settles little, because it does not say which differences make two cases dissimilar. The material part does say, and it is where a criterion works. A criterion names the difference that is allowed to change a person's share.
The Stanford Encyclopedia of Philosophy shows how many choices this leaves open. Principles of fair distribution differ in what they share out, such as income, opportunities or welfare. They differ in who receives it, whether single people or whole groups. And they differ in the basis of the split, such as equality, total benefit, individual traits or free exchange.
Four standards are enough to start comparing:
| Criterion | The share goes by | The question it puts to a nudge |
|---|---|---|
| Equality | the same amount for everyone | Did every group gain the same? |
| Contribution or desert | what each person put in or has earned | Did people who put in more gain more? |
| Need | how much a person lacks | Did the people with least gain most? |
| Difference principle | the position of the worst-off group | Does any gap in the results raise the floor for those at the bottom? |
Each rule has a cost. Equal shares ignore that people start from different places. Shares by contribution favour those who had more to give. A need rule gives more to the people who have least, whatever they put in, and someone must then measure need. The difference principle asks the same of whoever defines the worst-off group. No rule avoids every cost, so the practical task is to choose one in the open and defend the choice.
032 min
Where the criterion of justice comes from
The best-known criterion comes from the philosopher John Rawls. In 1971 he published A Theory of Justice, written as an answer to utilitarianism, the view that the best arrangement is the one with the greatest total welfare. A large total can hide a minority that fares badly. Rawls wanted a test that would not let that pass.
His method was a thought experiment. Representatives behind a veil choose rules for citizens whose race, class and gender they cannot see. The veil also hides age, natural abilities and wealth. Any citizen could end up the poorest, so the representatives have a reason to choose rules that work from every position.
Two rules come out of that choice. The first guarantees every person the same basic liberties. The second allows inequalities only if two conditions hold. Jobs and positions must be open to all on fair terms, and the arrangement must leave the least-advantaged group better off than any alternative would. The second condition is called the difference principle.
The difference principle is a test, not a ban on inequality
Strict equality is one extreme, and the difference principle sits beside it rather than inside it. Rawls's rule can allow more inequality than strict equality does, if that inequality lifts the bottom. A pay gap passes when it raises the position of the worst-off group and fails when it does not.
Liberty comes before gains
Rawls ranked his two rules. The first has priority, so the arrangement may not trade basic liberties for a better position for the worst-off. For a designer this has a direct use: a nudge cannot buy a fairer result by removing a person's ability to refuse it.
The phrase "criterion of justice" is wider than Rawls. Philosophers and social scientists use it for any standard of fair shares, including equality, contribution and need.
042 min
Other criteria of justice and their objections
Rawls's test is one answer among several, and each rival points at something it leaves out.
- Luck egalitarians hold that people should not end up worse off because of circumstances they did not choose, although they may differ because of choices they made. An audit of a nudge in this spirit asks whether a group's worse result came from its own choices or from the way the options were laid out.
- Welfare-based principles, of which utilitarianism is the best known, judge a distribution by the welfare people end up with, not by their share of material goods.
- Libertarians, with Robert Nozick as the best-known advocate, deny that a fair result has to match any pattern. For them a distribution is just if it arose from fair acquisition and free exchange. Under this view, an audit of a default has no pattern to check. It can only ask whether the default respected free action.
- Corrective justice does a different job. It repairs a wrong that one person did to another, where a criterion of justice, in the sense used here, shares out a good.
Rawls also gave a plain answer to the question of who counts as worst-off. He measured how citizens are doing by their holdings of primary goods: basic rights and liberties, freedom of movement and choice of occupation, powers of office, income and wealth, and the social bases of self-respect. A team applying the difference principle to a nudge has to decide which of these the nudge changes.
052 min
How a criterion of justice tests a nudge
The pension default can now be tested against each standard, and the DWP's figures by pay band supply the evidence.
Workers earning between £10,000 and £20,000 a year reached 81 per cent participation in 2018, up 47 percentage points since 2012. The highest participation, 93 per cent, was among those earning £50,000 to £60,000. Because the lowest-paid group of eligible workers gained the most, the distance between pay bands shrank.
The standards give three different verdicts on a nudge like this one:
- Equal results: not met, but closer. Pay bands still differ, 81 per cent against 93 per cent, and the gap narrowed.
- Need: met. Among enrolled workers, those with the lowest pay gained the most.
- Difference principle: met inside the scheme's reach. The default lifted the lowest group it covered, but a test built for the worst-off must also ask who sits outside that reach.
The same data support three answers, and none of the answers is false. Only the standard changed.
Choosing what counts as a gain
Participation counts who is enrolled. It does not count how much each person saves. On a second measure the picture differed. In 2018, private-sector employees contributing more than the minimum were more likely to be older, higher-paid and at larger employers, although that skew was shrinking. A team that checks only enrolment would never see it. The criterion works only after the measure is fixed, and a different measure can change the verdict.
062 min
Where a criterion of justice gets hard
Choosing a standard is easier than applying it. Three problems come up in practice.
An average can hide a loss
In 2021 the legal scholar Cass Sunstein wrote that a nudge can leave an identifiable group no better off, or worse off, even when it helps people overall. A fairness test that looks only at the mean cannot see this. The check has to compare groups. Sunstein's suggested remedy is more targeted, personalised nudging, so that each group gets the design that works for it. That asks the designer to know which group a person belongs to, and that knowledge raises its own fairness questions.
Equality can lower everyone's share
An equality rule can demand a result that is worse for every group, when the only way to close a gap is to take from the better-off without giving anything to anyone else. Philosophers call this the levelling-down objection. Researchers who study fairness in software that makes decisions about people raise it against metrics that require equal rates between groups. A criterion that looks only at the gap, and never at the level, can approve a worse outcome for everyone.
The default may not reach the people who need it most
The UK scheme enrols only workers who earn above a set trigger. The DWP calls workers on the lowest earnings entitled workers. They can join a workplace pension if they choose, but they are not enrolled automatically. A test built on the difference principle must ask about this group first, because the benefit did not arrive for them by default.
071 min
How people judge whether a nudge is just
A criterion measures outcomes. People also judge fairness by how they were treated, so a nudge can pass the test and still feel unjust to the person nudged.
A 2024 thesis at the University of Oklahoma by Jenna Holt measured this. Holt built a questionnaire for four kinds of perceived justice: distributive, procedural, informational and interpersonal. In words, these ask whether the share was fair, whether the process was fair, whether the person was told what they needed to know, and whether they were treated with respect.
The questionnaire was used to compare two ways of encouraging plastic recycling. A framing nudge, which words a message to stress gains or losses, did not reliably beat a control group that saw no message. Educational messages that were well built often raised perceived justice.
The result comes from one setting and one thesis, so it is a prompt rather than a rule. It suggests that a team should measure how fair a nudge feels as well as how well it works.
082 min
Running a justice check on a nudge
Picture a product team that wants to make annual billing the default for new customers. Sign-ups rise in the first test. Annual billing asks for one large payment, which a customer on a tight budget may find hard. Before shipping, the team runs a justice check.
Name the groups.
List the customers who differ in ways that matter here, such as those on tight budgets, those new to the product and those using assistive technology.
Fix the measure.
Decide what counts as a gain for the customer, such as a lower yearly price or fewer failed payments, and what counts as a cost. A sign-up is a gain for the company, not necessarily for the customer.
Pick the criterion first.
Write down whether the test is equal gains, gains by need or the difference principle, before any results arrive.
Compare groups, not the mean.
Report the outcome for each group. A group with a worse outcome gets a closer look.
Check the way out.
Leaving the default must cost every group the same effort. If leaving takes more steps for some customers, the default favours the people who can afford the effort.
Ask people how it felt.
Add questions on the four kinds of perceived justice to the follow-up survey.
A check like this costs little compared with a default that reaches only part of the audience.
091 min
Criterion of justice vs. nearby concepts
Three neighbouring ideas are easy to mix up with the criterion.
| Concept | What it covers | How it differs |
|---|---|---|
| PsychologyEquity theory | ||
| DesignLibertarian paternalism | ||
| Not in the library yetProcedural justice |
?4 questions
Questions people ask
Is a criterion of justice the same as a law or a legal test?
Does a nudge need a criterion of justice if it is easy to opt out of?
Who decides which criterion a team uses?
Can one design satisfy more than one criterion?
§6 sources
Sources
UK Department for Work and Pensions (2019), Automatic enrolment evaluation report 2019. Participation figures by pay band, and the groups outside automatic enrolment.
Stanford Encyclopedia of Philosophy, Distributive Justice. The ways distributive principles differ, the difference principle and its priority rules.
Stanford Encyclopedia of Philosophy, John Rawls. The original position and the veil of ignorance.
Cass R. Sunstein (2021), The distributional effects of nudges, Nature Human Behaviour 6, 9-10. Read through the Harvard Law School Library record, which gives the abstract. Why overall welfare effects are not enough.
Show all 6 sourcesShow fewer sources
Jenna Holt (2024), An exploration and evaluation of choice architecture interventions on perceptions of justice, University of Oklahoma. The four kinds of perceived justice and the recycling studies.
Baumann, Hertweck, Loi and Heitz (2022), Distributive Justice as the Foundational Premise of Fair ML, arXiv. The levelling-down objection applied to fairness metrics.
