The default that nobody was bound by
Defaults are the most quoted result in behavioural science. They are in every introductory lecture, every consultancy deck, and every article explaining the field to people who have not encountered it before. The evidence behind them is real. Auto-enrolment into workplace pensions genuinely changed savings behaviour in the UK at a scale that few interventions of any kind have matched.
So it is worth looking closely at the case where the same logic was applied to something with much higher stakes, at national scale, and did not work.
What England did
The Organ Donation (Deemed Consent) Act 2019 came into force on 20 May 2020. It switched the default position for adults in England. Rather than requiring people to opt in by registering as donors, consent would now be presumed unless a person had opted out, either on the register or by telling someone close to them.
The reasoning was straightforward and behaviourally orthodox. Before the change, roughly 37 percent of the population in England had registered on the Organ Donor Register, the consent rate stood at about 63 percent, and the family refusal rate was around 36 percent, which was high next to comparable countries. Spain sat at about 13 percent. If most adults could be moved into the donor category by default, the reasoning went, consent rates should rise and more transplants should follow. NHS Blood and Transplant's own modelling predicted a post-legislation consent rate of 78 percent.
The observed rate was 61 percent.
It has since fallen further. NHS Blood and Transplant's activity report for 2024/25 puts overall consent and authorisation across the UK at 59 percent, describing it as static and as having halted a post-COVID decline. The target is 62.
The obvious objection, which is real
The law came into force during the first COVID lockdown, which is close to the worst possible timing for evaluating anything. Deceased donation numbers fell by around 25 percent during the pandemic period and had not fully recovered years later. Specialist nurses were redeployed. The evaluation had been designed before any of this happened.
When a health minister was asked in 2023 whether the government would review the impact of the Act, the answer was essentially that the confound made it impossible to say. That is a fair answer and I do not want to argue past it. Any strong claim about the size of the effect is not supportable from this data.
But the qualitative evaluation work does support a claim about mechanism, and mechanism is what I am interested in.
Where it actually broke
The evaluation carried out by the Policy Innovation and Evaluation Research Unit at LSHTM with Bangor University found something more specific than a null result. Deemed consent was not treated, in practice, as equivalent to a decision the person had explicitly recorded while alive. Families continued to believe that they were the decision makers. And the new system made the conversation at the bedside more ambiguous rather than less, because families were receiving mixed signals about whether their view was being sought or merely confirmed.
The numbers underneath this are stark, and the most recent ones are the starkest. In 2024/25, where a person had expressed their decision, the consent rate was 87 percent. Where consent was merely deemed, it was 48 percent, meaning more than half of families declined to support donation in exactly the cases the law was written to cover.
Sit with that gap for a moment. Same population, same hospitals, same conversations, same specialist nurses. The difference between a person having said yes and the state having said yes on their behalf is thirty-nine percentage points.
And even an explicit decision is not proof against it. In the same year, 173 families overruled their relative's recorded decision to donate. Research from Wales, where the law changed five years earlier, found families overriding a registered opt-in around 16 percent of the time.
So the default was applied to one person and the decision was made by another.
The heuristic
That is the part I think generalises, and it is a question I have started asking of any proposed default.
Does the default bind the person who holds the decision at the moment the decision is actually made?
For pension auto-enrolment, it does. The default applies to the employee, the employee is the person who would have to act to reverse it, and the moment of decision is an ordinary working day on which doing nothing is easy and socially unremarkable. The default and the decision sit in the same place.
For deemed consent, they come apart completely. The default applies to the deceased. The decision happens in a hospital, hours after a sudden death, and is made by a family member who is in acute grief, who often does not know what the person had registered, and who has not been relieved of the feeling that they are being asked. The default arrives at that conversation as information about someone's presumed preference, which is a much weaker thing than a decision, and it competes against a person's terror of getting it wrong on behalf of someone they loved.
A default only does its work if declining it requires effort. Here, declining required nothing at all, because the person the law bound was not in the room.
What this is not
I want to be careful about what I am claiming, because this argument is easy to overextend into a general dismissal of defaults, which would be wrong and would also be lazy.
Defaults work. The evidence is strong. What the organ donation case shows is a boundary condition rather than a refutation, and boundary conditions are how a field matures. The intervention was correctly identified and incorrectly located. Presumed consent addressed the registration behaviour of the general public when the binding constraint sat with a different group of people at a different moment.
There is also a version of this where the law was never really the intervention. The Act came with substantial funding for publicity and evaluation, and several accounts suggest that part of its purpose was to shift the cultural conversation about donation rather than to mechanically flip a switch. Judged as a culture change programme with a legislative centrepiece, the evaluation would look different, and possibly better. I am not sure how to weigh that fairly, and I notice that the people who supported the change on those grounds and the people who supported it on mechanical grounds were making quite different bets.
One more finding from the statistics deserves more attention than it usually receives. Consent rates have differed sharply by ethnicity, running at roughly 65 percent for white patients against 35 percent for patients from ethnic minority backgrounds. A default cannot address a gap of that size, because whatever is driving it is not about the effort of registering. Applying a uniform default across a population with a distribution that wide produces an average that hides the thing you most need to see.
There is also a signal in the opposite direction that I do not know what to make of. NHSBT reported opt-out registrations running at just over 52,000 in a single quarter of 2025, an increase of around 122 percent on the same quarter the previous year, which they attribute partly to negative media and social media narratives. A default that provokes people into actively registering against it is doing something, but it is not the thing it was designed to do.
The point
If I take one thing from this into any future work on defaults, it is to stop asking whether the default is set correctly and start asking who is standing at the decision when it is made.
Those are not the same question, and the second one is harder to answer from a desk.