Small on average – but powerful where it counts: what really shifts diets and cuts food waste

What people choose to eat – or throw away – has consequences far beyond the dinner table. The global food system accounts for up to one third of global greenhouse-gas emissions. Food that is produced but never eaten represents around one quarter of food-related emissions, or roughly 6–10% of total global emissions.

“Food was the category with the highest technical potential for reducing emissions through changes in consumption,” says behavioural economist Paul Lohmann of the University of Cambridge Judge Business School in the United Kingdom, referring to recent assessments by the Intergovernmental Panel on Climate Change (IPCC).

“Shifting diets and reducing food waste are among the most impactful actions we have.”

But despite growing interest in nudges, labels and other behavioural interventions, the evidence has remained scattered.

“There is very little systematic evidence, and much of it is fragmented,” Lohmann says.

In a new study published in Nature Food, Lohmann and colleagues combined results from 110 articles, covering more than 2.4 million observations, to test how well consumer-focused interventions shift diets in a more sustainable direction and reduce food waste – focusing only on what people actually do across many real-world settings.

Across all interventions, the effects are small – and highly uneven. One category consistently stands out across settings and study types: interventions that change the choice setting itself, such as making plant-based meals the default or making them easier to find.

“These decision-structure interventions consistently work on average,” Paul Lohmann says. “They have the largest effect sizes – and still show meaningful shifts in behaviour even after correcting for publication bias.”

The findings suggest that behavioural interventions can influence climate mitigation – but they also raise a more fundamental question: which approaches change behaviour in the real world – and why?

A huge climate lever – but weak evidence on what actually works

Efforts to transform the global food system now focus not only on how food is produced but also on how it is consumed. Diets rich in animal-based products and large amounts of food waste contribute substantially to greenhouse-gas emissions, land use and biodiversity loss.

Researchers and policy-makers are now exploring whether interventions targeting consumers can shift eating habits toward lower-impact foods and reduce the amount of edible food that ends up in the bin.

The shift in focus reflects a broader change in how climate solutions are understood.

“The IPCC report was the first to have a dedicated chapter on demand-side interventions,” notes Lohmann. The chapter helped motivate the new research – and senior author Lucia Reisch, El-Erian Professor of Behavioural Economics and Policy, Judge Business School, University of Cambridge, contributed to it.

“The IPCC report highlighted this huge technical potential for demand-side changes to bring about meaningful reductions in emissions.” Compared with sectors like mobility, he adds, food stands out in terms of demand-side potential.

Despite this potential, what actually works remains surprisingly unclear.

“The evidence base is difficult to navigate,” Paul Lohmann says. “We do not fully understand which interventions actually move the needle.”

Many interventions, little comparability – and unclear climate impact

A growing number of studies have tested interventions aimed at influencing food-related behaviour. These range from ecolabels and information campaigns to financial incentives and social norm messages. Others change the choice environment itself, subtly reshaping the setting in which decisions are made.

But comparing these approaches is not straightforward.

“Most studies focus on very specific interventions or particular settings,” Paul Lohmann explains. “That makes it difficult to compare results across the literature.”

Another challenge is that studies often measure very different kinds of outcomes.

“There are many studies, for example, on increasing vegetable consumption,” Paul Lohmann says. “But it is not clear what the substitution patterns are – whether people actually eat less meat. So we did not include those.”

The problem runs deeper: not all behavioural changes matter equally for climate outcomes – and much of the existing evidence is based on what people say, not what they actually do, making it difficult to compare results across studies.

From fragmented studies to systematically testing what works

To resolve this, the new meta-analysis brings together a large and diverse body of experimental research under a common framework so that different approaches can be compared directly across contexts and study designs.

“If you are designing policies, you want to know which approaches work most consistently,” Paul Lohmann says. “But also under which conditions they work best.”

The goal is not just to estimate an average effect across all interventions types but to understand variation:

“The average effect can be misleading. What matters is which interventions actually work and where they work.”

To answer this question, the researchers conducted a systematic review and meta-analysis assisted by machine learning, combining evidence across experiments, settings and populations.

“We wanted to synthesize the literature,” says Paul Lohmann.

The team searched major databases and grey literature, initially identifying more than 35,000 records. Using a combination of manual screening and machine-learning tools designed to identify the most relevant articles, the researchers narrowed the dataset to 110 articles that met their criteria for inclusion.

Narrowing the field to interventions that can actually reduce emissions

The final dataset included 139 unique experiments – covering 95 studies on food consumption, 43 on food waste and one addressing both topics – which together produced 306 effect sizes based on more than 2.4 million observations.

The researchers included only studies that measured what people actually did – not just what they said they would do. They also restricted the analysis to outcomes with clear climate relevance, ensuring that results could be compared across studies. To make sense of highly diverse literature, the researchers grouped interventions into six categories based on how they are thought to influence behaviour – including incentives, information, labelling and various ways of changing the choice environment.

They then translated the results from different study designs into a common measure so that they could be compared directly.

“At least what was consistent across all these studies was that they looked at an intervention versus a control,” Paul Lohmann says. “That gives us a common basis for comparing very different interventions.”

Accounting for bias and real-world complexity

The team also addressed a key challenge: combining studies conducted in very different settings and designs while testing whether the results held across that variation.

To address this, they used multilevel statistical models and explored variation across intervention types, settings, populations and study designs. They also tested for publication bias, which can inflate reported effects, to check whether the overall pattern held under more conservative assumptions.

The underlying data have also been made available so that others can explore specific interventions, settings or questions in more detail.

When the researchers finally pooled results across all studies, the overall picture was clear – but more modest than many individual experiments might suggest. Across studies, demand-side interventions can shift food behaviour, but the effects are generally small: on average, they lead to modest changes rather than dramatic shifts (corresponding to effect sizes of about 0.3–0.4).

“The overall average effect is about 0.3, which is considered small,” says Paul Lohmann.

But this does not make the interventions irrelevant: small changes, repeated across millions of decisions, can still add up – and cost–effectiveness matters as much as effect size. The average, however, is only part of the picture.

“This average effect hides a great deal of variation across studies and contexts,” Lohmann says. “Some interventions work quite well, whereas others have almost no effect – and much of that variation remains unexplained.”

The average hides what actually works

The variation is so large that outcomes can differ dramatically depending on context. This suggests that most studies do not capture many factors influencing behaviour. As a result, looking only at the overall average can be misleading.

“We caution against interpreting the overall average as the main finding,” Paul Lohmann says. “Looking at the different types of interventions is more useful.”

When the researchers break down the results, a clear pattern appears across the evidence: the most effective interventions are those that change the choice environment – the setting in which people make decisions.

“It was very clear that the category that worked best was what we call decision-structure interventions,” Lohmann says.

These include interventions such as making plant-based meals the default option, increasing their availability or reducing portion sizes. One reason is that these interventions reduce the effort required to make a sustainable choice.

“If the default is already the plant-based option, people do not have to actively choose it,” Lohmann says. “And many people simply stick with the default.”

Changing the environment beats changing minds

By contrast, interventions that rely on information – such as educational campaigns or labels – show much weaker effects. The evidence on financial incentives remains limited, although Lohmann notes that they appear to more strongly influence food waste than diet change.

“That is not to say that information plays no role – it may raise awareness in society. But in terms of changing behaviour in real-world settings, it does not seem to do much.”

The analysis also highlights publication bias.

“When we correct for publication bias, many intervention types move closer to zero,” Lohmann says. “So the literature may overestimate some of these effects.”

Even so, the overall pattern remains: interventions that change the decision structure are consistently the most promising across analyses. Across methods and robustness checks, the same conclusion emerges: behavioural interventions can shift diets and reduce food waste – but they are unlikely to drive large changes on their own.

“We should not expect very large effect sizes from very small interventions,” Lohmann says.

The literature may overstate effects – but the pattern holds

Rather than dismissing these approaches, Lohmann argues that their value lies in how they are used – and combined. Even modest shifts can matter when applied across millions of decisions. Small changes in individual behaviour translate into substantial reductions in emissions when scaled across entire populations.

Not all interventions are equally effective.

“Providing information alone does not do much to change behaviour,” Lohmann says. “It may raise awareness – but it is unlikely to shift real-world choices.”

Instead, the strongest and most consistent effects come from interventions that reshape the decision environment itself.

“What works best is changing the environment in which people make decisions,” Lohmann says. “Making the sustainable option the easy or default choice.”

These approaches are already being tested in practice – from plant-based defaults in canteens to redesigned menus and portion sizes in restaurants. But scaling them raises new questions, particularly about feasibility and incentives.

“I think one challenge is how to align incentives so that companies actually want to implement these changes,” Lohmann says. “Because many existing choice environments are shaped by commercial priorities.”

Make the sustainable choice the easy choice

Digital platforms may be especially well suited for testing these ideas.

“Online environments are arguably the easiest place to change the choice environment at scale and low cost,” Lohmann says. “You can reorder menus, change defaults and adjust what is shown most prominently.”

The study also points to a crucial insight: the current food environment is not neutral. Many of these effects can be seen as the social cost of commercially driven choice environments that already steer people toward certain, often less sustainable, options.

Rather than adding something new, many interventions counteract a system that is already shaping behaviour.

Scaling change requires aligning incentives

The results also highlight a fundamental challenge: what works in one context may not work in another. Interventions often perform better in universities or schools than in more complex environments like grocery stores – yet many of the most consequential choices are made in grocery stores.

Part of the reason is practical: researchers have better access to educational settings. But it also mirrors the messy reality of everyday decision-making, in which consumers face many competing influences at once.

Even after accounting for known variables, much of the variation in outcomes remains unexplained.

“There is still a lot of heterogeneity that we cannot explain,” Lohmann says. “This suggests that many important factors are simply not being measured.”

For Lohmann, this highlights a clear priority for future research.

“We need more systematic, large-scale experiments that help us understand not just whether interventions work but when and why they work,” he says.

What works depends on context – and much remains unexplained

At a deeper level, the findings also raise questions about how choices are shaped.

“Whether it is still the right decision depends on the perspective,” Lohmann says.

The implication is that behaviour is shaped not just by preferences but by the environments in which those preferences are turned into choices.

Ultimately, the study reframes these approaches as part of a broader toolbox that should complement – not replace – supply-side policies.

“Demand-side interventions are a genuine lever for reducing food-related emissions,” Lohmann says.

But using them well means moving beyond averages and focusing on mechanisms:

“If we want to design better policies, we need to understand which interventions actually work, in which contexts – and why.”