Structural interventions — change the environment around the person, not the person's information

The thesis in one line

The interventions that produced the biggest sustained reductions in addictive behaviour over the last forty years did not work by changing what people knew or what they were told. They worked by changing the physical, social, economic, or regulatory environment in which the choice was made. This is the structural-intervention hypothesis. It is empirically well-supported, ideologically uncomfortable for the willpower frame, and operationally invisible to most political conversations about addiction policy.

The pages either side of this one make the constructive cases — individual-level treatments that work for the addicted minority, the five drugs that would close the largest share of the global disease burden. This page makes the third constructive case, the one that operates at the population level and the prevention end of the funnel: the things you do to the world that reduce who becomes addicted in the first place. The Countries page documents specific national experiments. This page extracts the pattern across them.

Iceland — Planet Youth, the headline case

The cleanest example in modern public health. Iceland in 1998 had among the highest adolescent substance use rates in Europe. By 2013, fifteen years later, it had among the lowest. The intervention contained no classroom drug-education content. It changed the environment around the kid.

The model, designed by Inga Dóra Sigfúsdóttir and colleagues at Reykjavík University and operationalised globally as Planet Youth, has five planks:

1. Parental engagement. Norms about parental presence, bedtimes, knowing where the child is and who they're with — turned from "uncool parenting" into a community-level expectation through municipal campaigns aimed at parents, not children.
2. Organised after-school activities. Massively expanded municipal funding for sport, music, art, and youth clubs — distributed through a "leisure card" voucher system that subsidised the activity at the family level so cost was not a barrier. The mechanism: occupy the after-school hours that would otherwise have been unsupervised and substance-curious.
3. Curfews. Legally enforced, age-banded, community-supported (parents patrol, not just police). Younger children off the streets earlier.
4. Community data feedback. Annual surveys of adolescent behaviour at the school level, fed back to municipal officials, parents, teachers — so the intervention is continuously re-calibrated against the actual local picture.
5. Long-term multi-sectoral coordination. Schools, municipalities, parents, sports clubs, health services — all on the same page, sustained across changes of government for two decades.

The results in Iceland are the largest sustained reduction in adolescent substance use ever recorded in a national population. Past-30-day drunkenness in 15-16 year olds fell from 42 per cent in 1998 to 5 per cent by 2016. Past-30-day cannabis use halved. Lifetime smoking rates fell from 23 per cent to 3 per cent ([Planet Youth](https://planetyouth.org/the-icelandic-prevention-model/); [Sigfúsdóttir et al., Health Promotion International, 2009](https://academic.oup.com/heapro/article/24/1/16/677129)).

The international replication story is more honest and more interesting. The model has been adopted in at least 24 communities across Canada, the US, Chile, Spain, Lithuania, the UK, the Netherlands, and Australia. Results are mixed. Lithuania's three largest cities saw real reductions in 10th-grader alcohol, cannabis and amphetamine use ([Tarragona / IPM scoping review](https://pmc.ncbi.nlm.nih.gov/articles/PMC10061134/)). Other replications produced smaller or no effects, primarily where the model was adopted in form but not in funding or in sustained multi-sector coordination.

The honest read: the IPM is not a magic formula. It is a particular synthesis of structural interventions that works when funded, sustained, and matched to local social architecture. The lesson that transfers is not "do these five things" but rather "the lever you are looking for is in the environment around the adolescent, not in the adolescent". The lesson that does not transfer easily is the Iceland-specific social cohesion that made the multi-sector coordination cheaper to sustain than it is in less cohesive societies.

Medellín — social urbanism removed the pathway

The developing-country counterpart, different shape, same thesis. Medellín in the 1990s was the most violent city on earth — homicide rate above 380 per 100,000 in 1991, gang recruitment from the city's poorest comunas an organised industry, child soldiers a documented feature of the cartel labour market. Twenty years later the homicide rate had fallen ninety per cent, gang recruitment had structurally collapsed in the neighbourhoods that received the most investment, and Medellín was being cited internationally as a model.

The intervention, branded "social urbanism" under mayors Sergio Fajardo (2004–07) and Alonso Salazar (2008–11), was not a drug-education programme or a counter-recruitment curriculum. It was a sustained pattern of physical infrastructure investment in the poorest areas. The Metrocable aerial tram connecting the hillside slums to the formal city. The library-parks (España, San Javier) as architectural anchors in neighbourhoods that had never had public investment. The escalators of Comuna 13 replacing 350 steps with a covered, lit climb. Schools, child-care centres, sports facilities, public squares — concentrated in the areas the cartels had previously owned as recruitment territory.

The mechanism is the structural-intervention thesis at urban scale. The cartels recruited because they offered the only available pathway to status, income, and social meaning for adolescent boys in the poorest comunas. The municipality changed the available pathways — same boys, different choice set. Recruitment did not stop because the boys were better educated about the dangers of gang life (they already knew). It stopped because the alternative pathways were now visible, accessible, and culturally legible. ([Globe and Mail](https://www.theglobeandmail.com/news/world/social-urbanism-experiment-breathes-new-life-into-colombias-medellin/article22185134/); [Americas Quarterly](https://americasquarterly.org/fulltextarticle/medellin-front-line-of-colombias-challenges/); [IPA Medellín gang-recruitment evaluation](https://poverty-action.org/understanding-how-reduce-adolescent-recruitment-gangs-evidence-medellin-colombia).)

Important caveats. Medellín's success is contested in detail — the homicide reduction is real but follows multiple causes (the Pablo Escobar death in 1993, the demobilisation of the AUC paramilitaries in the 2000s, urbanisation effects, exchange-rate effects on cocaine economics). The social-urbanism investments cannot claim sole credit. The city still has serious organised-crime problems. What is defensible is the narrower claim: physical and social infrastructure investments in the poorest comunas changed the recruitment economics for armed groups, and the effect is measurable in the IPA evaluations and in the longitudinal homicide and recruitment data. The intervention was structural. The mechanism was pathway-removal, not education.

Smoking — culture did most of the work, structural levers accelerated the late stages

The UK smoking story is the case most often cited for structural-intervention success. The numbers look spectacular: adult prevalence fell from ~45 per cent (men, 1974) to ~13 per cent (2023); per-capita cigarette consumption in 2024 is roughly one-fifth of its 1974 peak; lung-cancer mortality has followed with a 20-year lag. But the structural interventions were not the dominant cause of that decline, and overclaiming them as such is exactly the kind of post-hoc victory lap the rest of this site has been refusing to give the willpower frame.

The honest decomposition. Most of the 32-percentage-point decline since 1974 was already in the bank before the workplace ban (2007), display ban (2012/15), and plain packaging (2016). By 2000, UK adult prevalence was already around 27 per cent — meaning roughly 18 of the 32 percentage points (~55 per cent) had already fallen before any of the headline structural interventions arrived. The post-2007 phase added another ~14 percentage points, of which a meaningful but not majority share is attributable to the structural levers.

A rough per-lever attribution, with appropriate uncertainty. Numbers below are central estimates from the published evaluation literature combined with counterfactual decomposition; the bars are wide and the literature contests every row.

Adds to roughly 28–39 pp, comfortably bracketing the observed 32 pp decline. The cultural-and-informational baseline did about half the work; price did the next biggest chunk; the post-2007 bans contributed real but marginal acceleration. The "structural interventions were primarily responsible" framing collapses this distinction.

Two specific points that further weaken the structural case:

Youth uptake fell before the bans. The cohort of teenagers who would have become 2010s adult smokers were already largely not starting by the mid-1990s. The drivers were cultural and informational — smoking became uncool, parents stopped smoking in the car, anti-smoking PSHE landed in classrooms. The display bans and plain packaging (2012+) arrived after the youth-uptake curve had already bent. They reinforced the cultural shift; they did not initiate it.

The most addicted smokers barely moved. The structural interventions worked best on social smokers and the marginally-dependent — exactly the population already most responsive to cultural signals. The heaviest-dependence smokers (the ~10 per cent who account for the largest share of consumption and mortality) showed much smaller responses to bans, duty escalators, or display restrictions. The pattern is identical to the Scottish MUP case: the structural lever works at the level it was designed to operate (price-elastic moderate users), but the addicted minority — the population the policy ostensibly exists to protect — moves less than the headline implies. The actual lever for the heaviest smokers is pharmacotherapy.

The honest summary: UK smoking decline is overdetermined — culture, information, price, bans, cessation drugs, and vaping all contributed. The structural-intervention thesis can claim a real but minority share of the credit. The "textbook case of structural-intervention success" framing this page started with was overconfident; the truer version is that structural interventions accelerated a decline that was already two decades old when they arrived. The cumulative effect is documented in ASH 2024 and in Public Health England's evaluation series — both of which, read carefully, support the partial-credit reading rather than the dominant-cause reading.

Sugar tax — industry reformulated, consumer effects contested

The most instructive recent UK example for the structural-intervention case is the Soft Drinks Industry Levy, introduced April 2018. The mechanism is structural in an interesting way: the tax was levied on producers of sugar-sweetened beverages above a sugar threshold, not on consumers. The producers had a year's notice before implementation. Most of them reformulated their products below the threshold to avoid the tax rather than passing the tax onto consumers.

The industry-response claim is solid. Total sugar content of soft drinks sold in the UK fell by approximately 30 per cent in the first year via reformulation rather than consumer switching, documented in BMJ 2020. The structural mechanism worked at the level it was designed to operate: producer behaviour changed because the tax incentive changed.

The downstream health claim is much weaker and worth flagging. Chris Snowdon at the IEA has been the most consistent critic and his core point lands: the 30 per cent figure is sugar per product, not sugar per consumer. People drink more of a reformulated product, substitute calories elsewhere, switch to non-taxed drinks, or top up sugar from non-soft-drink sources. UK household-purchase data on total sugar intake from soft drinks moved much less than the per-product reformulation figure implies. The widely-cited JAMA Pediatrics 2024 paper reports an 8 per cent relative reduction in year 6 obesity prevalence — in girls only, no effect in boys — which is the classic multiple-comparison flag that a methodologist would treat with suspicion. If a structural intervention worked at the population level, you would expect both sexes to move, not one. The defensible inference is that the SDIL worked as a producer-reformulation lever and may have had small downstream effects in some subgroups; the inference that it bent the obesity curve is not yet earned.

The honest version of the structural-intervention case from the SDIL is therefore narrower: regulatory pressure on producers can change product composition at scale without educating the consumer. Whether the composition change translates to population health outcomes depends on substitution behaviour, calorie balance elsewhere, and per-product consumption changes — none of which the SDIL evidence base has yet pinned down cleanly. The eatwell-plate campaigns of the 2000s were the willpower frame in operation and the obesity rate rose against them; the SDIL is a different kind of intervention and the jury on its population effect is still out.

Minimum unit pricing — the Scottish trial, confounded by COVID

Scotland's Minimum Unit Pricing of alcohol (50p per unit, May 2018) is the most-cited recent UK example of a structural price intervention. The intervention does not change the price of premium spirits; it changes the floor price of the cheapest, strongest alcohol — the white cider, the budget vodka, the products that are disproportionately consumed by the heaviest drinkers.

What the evaluation says, and the load-bearing caveat. Public Health Scotland's final report (June 2023) and the associated Wyper et al. Lancet 2023 paper report a 13.4 per cent reduction in alcohol-attributable deaths and a 4.1 per cent reduction in alcohol-attributable hospital admissions, concentrated in the most deprived quintile. The reduction is a synthetic-control counterfactual, not an observed reduction.

That distinction matters. Scottish alcohol-specific deaths actually rose in absolute terms across the post-MUP period, peaking at 1,245 in 2021 — the highest figure since 2008 (Scottish National Records). The MUP-attributable effect is the gap between the observed deaths and what the synthetic-control model says deaths would have been without MUP. The post-MUP period overlaps almost entirely with COVID lockdowns, during which home-drinking by heavy drinkers surged across the UK, MUP-jurisdictions and not. Disentangling the MUP effect from the COVID effect using synthetic controls is methodologically defensible but it is not the same kind of evidence as a clean before-and-after observed reduction. The "MUP saved lives" headline is doing more work than the underlying analysis supports.

Snowdon at the IEA has made the COVID-confound case at length and the methodologists' response (the synthetic control matches on pre-COVID trajectories, the gap survives several sensitivity checks) is partly persuasive. Net read: the MUP probably did some structural work at the heavy-drinker margin where the price floor most bit; the magnitude of the population-health effect is contested and the headline is overconfident; the regressivity critique (Spiked, the IEA's Snowdon arithmetic on the Costs page) is also defensible and applies in parallel.

The honest version of the structural-intervention case from MUP is therefore: price-floor mechanics work at the level the policy was designed to operate (heaviest drinkers buying cheapest products do reduce volume); the population-mortality effect is plausible but confounded and probably smaller than the headline. The willpower frame has no language for either fact; the structural-intervention frame has language but has to be honest about the inferential gap between "sales fell in the targeted decile" and "deaths fell in the targeted population".

Choice architecture — the quieter lever

The smallest interventions are also structural and they accumulate. Some recent UK examples:

• Supermarket placement rules for HFSS foods (high fat, salt, sugar) — October 2022. Restricted location at checkouts, store entrances, and end-of-aisle displays. Removes the impulse-purchase environment for the most calorically problematic foods. Early evaluation shows measurable basket-composition shifts.
• Calorie labelling in chain restaurants — April 2022. The educational layer of this intervention is the one journalists wrote about. The structural layer — restaurant menu redesign in response to having to display the numbers — is the one that has produced the larger effects. Chains have reformulated portion sizes and menu composition in response to the labelling requirement.
• Bans on HFSS advertising before 9pm — 2025 implementation. Removes the cue-exposure environment for younger viewers. Cue exposure is the Berridge incentive-sensitization mechanism (see Pathway § Wanting); reducing the cue is structurally how you reduce the wanting.

The unifying point: the choice architecture around a decision matters more than the information available at the moment of decision. Behavioural economics has been making this case for two decades. Addiction policy is finally catching up.

Why structural beats educational at population scale

Three mechanisms.

One: the binding constraint on behaviour change is rarely information. The Policy page documents this exhaustively. People know smoking is bad. People know the calorie content of a Big Mac. People know what fentanyl does. The information gap closed decades ago. The binding constraints are price, availability, social pressure, default options, and the architecture of the decision — all of which are structural variables.

Two: structural interventions don't require sustained individual effort to keep working. A smoking ban is on every day. A sugar reformulation is in every can. A library-park is open every weekend. The intervention is in the environment and the behavioural change follows automatically. Educational interventions require the individual to retain the information, value the information, and apply the information — three independent failure points, all of which are heavily moderated by stress, mental health, and the very addictions the intervention is trying to address.

Three: structural interventions reach the people educational interventions don't. The educational frame works best for the highly-motivated, well-resourced, low-comorbidity middle. The structural frame works for everyone in the environment, regardless of motivation, resources, or comorbidity. Population-level interventions that depend on individual receptivity systematically miss the population that needs them most. Population-level interventions that operate on the environment do not.

Where structural fails

Three things, named honestly.

One: structural interventions can be politically extractive. Sin taxes are regressive — they extract revenue from a population (often the poorest) that has the highest exposure to the product. The Snowdon critique on the Costs page bites here. The fact that an intervention is structural does not make it fair, and the welfare loss to the population whose pleasure is being made more expensive is real and worth pricing.

Two: structural interventions can produce displacement, not reduction. Restrict alcohol access in one channel and consumers shift to another — illicit, cross-border, online. Restrict cigarette pack design and consumers buy generic vape devices. Restrict gambling advertising on TV and the same advertising migrates to social media. The intervention has to be designed to anticipate the displacement vector, or the headline reduction is offset by harm migrating to a less regulated surface.

Three: structural interventions do not treat the addicted. Iceland's Planet Youth model prevented adolescents from becoming addicted in the first place. It did not treat the adolescents who had already become so. Population-level prevention and individual-level treatment are complementary, not substitutes. A policy regime that does only structural prevention leaves the existing addicted population without a clinical pathway — which is the Untaken Lever argument from the other end of the funnel. Both are needed.

"90 per cent of public policy fails" — steelman + teardown

The challenge that hangs over the whole structural-intervention case is sharper than the willpower critique. It is the empirical claim that most public policy does not produce its stated effect. If the base rate of policy success is 10 per cent, then every successful intervention I have cited on this page has to be re-examined with the suspicion that it is a survivorship-bias artefact and the failure cases are the norm. The position deserves the steelman treatment.

Most evaluated public policies, across most domains, across most decades, do not produce the effect they were designed to produce. The Education Endowment Foundation in the UK runs RCTs on education interventions and reports null effects on roughly four out of five. US federal programmes when properly evaluated show similar patterns. The Cochrane review of compulsory drug treatment found no consistent evidence of effectiveness. Pressman and Wildavsky's Implementation (1973) documented that even well-designed policies fail at the gap between intention and execution. Easterly on aid, Scott on legibility, Tetlock on expert prediction, the entire behavioural-economics literature on Goodhart's law — they all point the same way. The base rate of policy success is low. The bureaucratic apparatus persists failed programmes because budgets are easier to defend than discontinue (see the 28-day rehab industry). The few successes that do occur are massively overrepresented in advocacy literature because failures don't generate the same publication incentive. Smoking decline is cited as a policy win, but as the section above just demonstrated, most of the work was done by cultural shift and information, not by the structural levers the policy lobby claims credit for. The honest base rate is somewhere between 70 and 90 per cent of public policy failing on stated objectives. Build a worldview on that floor, not on the cherry-picked wins.

Where the position is right, named in detail.

The base rate is genuinely low. The Education Endowment Foundation's data is real: roughly 80 per cent of UK school-based interventions evaluated by RCT show null effects on attainment. The OECD-wide pattern is similar. Most evaluated US federal social programmes either fail to demonstrate the intended effect or produce effect sizes so small that the cost-effectiveness collapses. The What Works Network in the UK, the J-PAL evaluations in development economics, the Coalition for Evidence-Based Policy meta-analyses — every serious effort to evaluate policy at scale finds that "no measurable effect on the dependent variable" is the modal outcome.

Implementation gaps are real. Pressman & Wildavsky's argument lands repeatedly. Policy designed in Westminster, Whitehall or Washington meets local administrative reality, professional capture, perverse incentive structures, and capacity constraints. The DARE programme had federal funding and was rolled out at scale and was evaluated to failure and continued running anyway. The same shape recurs across most policy domains. The mechanism the policy lever was supposed to engage either does not exist as designed (Goodhart) or is captured by the bureaucracy that runs it.

Bureaucratic persistence is real. The Industry page documents this for addiction services specifically — and Michael Shellenberger's San Fransicko (2021) is the most-cited recent diagnostic of exactly this failure mode in the US West Coast harm-reduction apparatus. The mechanism Shellenberger names — county-contracted residential rehab funded by bed-night billing producing poor outcomes but persisting because the funding model rewards beds not recoveries; NGO grant pipelines that depend on the problem continuing; political class that cannot publicly endorse evidence-based accountability without losing primaries — is the policy-failure base rate operating at the level of a specific industry. The 28-day rehab industry, the abstinence-only sex-ed curriculum, DARE, the various incarnations of the eatwell-plate campaign — all evaluated to failure, all still running or running in lightly-rebranded form. The bureaucratic-rent hypothesis is empirically supported. Full Shellenberger steelman + teardown on the Industry page.

The smoking case demonstrates the modesty of policy contribution honestly. The honest decomposition on this page just gave the post-2007 structural interventions roughly a third of the credit for the post-1974 smoking decline. Most of the work was done by culture, information, and price — and "culture" is not really a policy lever in the sense the policy literature uses the term. The willpower-critic argument that smoking decline is mostly cultural-and-informational is closer to the data than the structural-intervention-victory-lap version this page initially tried to claim.

Where the position is wrong, also named in detail.

"90 per cent" is hyperbolic; the actual fail rate depends on the bar. A more defensible figure is that roughly 60 to 75 per cent of evaluated policies fail to achieve their stated objectives. The 90 per cent number comes from the strictest "did the policy produce the headline effect at the magnitude promised" bar. If you relax to "did the policy produce a measurable effect in the right direction at any magnitude", the fail rate is lower — maybe 50 to 60 per cent. The 90 per cent figure is rhetorically powerful and analytically imprecise.

A small number of policies massively succeed and become invisible. Seatbelt laws roughly halved road-traffic fatalities and are now so culturally absorbed that nobody calls them a "policy success" any more. Childhood vaccination programmes eliminated diseases that killed millions per year. Water fluoridation reduced childhood caries in entire populations. The smallpox eradication campaign was a policy. The CFC ban under the Montreal Protocol prevented an ozone-layer catastrophe. These are not boutique exceptions; they are the strongest evidence base in public policy. They become invisible because successful structural interventions blend into the background of normal life. The 90 per cent figure systematically underweights them via survivorship bias of the opposite sign — the failures are visible because they keep failing, the successes are invisible because they stopped being problems.

The frame can become an excuse for nihilism that prevents the next attempt. If the policy maker absorbs the 90-per-cent-fails frame too completely, the response is paralysis or pure libertarianism. Both are bad answers for the addiction case specifically. The Untaken Lever argument requires policy action — opioid agonist therapy in primary care does not deploy itself. The right inference from a high base rate of policy failure is not "do nothing" but rather "design more carefully, evaluate more honestly, kill faster, scale less, target better". Tetlock's response to his own expert-prediction findings was not "abandon prediction" but "use better methods and track calibration". The same applies here.

Smoking decline really did include a structural component. The honest decomposition above gives the post-2007 structural interventions roughly 4 to 7 percentage points of contribution. That is not zero. Plus the duty escalator over 30 years probably 6 to 8 pp. Together that is roughly 10 to 15 pp of the 32 pp decline — about a third. Not the majority the structural-intervention thesis at full strength wanted, but not the rounding error the 90-per-cent-fails frame would predict. The policy contribution is real and quantifiable; what the page should not have done is claim it as the dominant cause.

The synthesis that survives both critiques.

The base rate of public-policy success is low — probably 25 to 40 per cent depending on the bar — and the addiction-policy field is no exception. Most evaluated addiction interventions fail; the Policy page documents that explicitly. Some interventions, however, do produce measurable structural effects, and those interventions cluster around a recognisable pattern: they operate on price, availability, environment, or the choice architecture around the decision, rather than on the information set or moral resolve of the agent. The smoking story, fairly told, illustrates both halves: cultural-and-informational shifts did most of the work; structural levers (price, bans, pharmacotherapy access) added meaningful acceleration. The structural-intervention thesis is not the dominant explanation of the smoking decline. It is the part of the explanation the willpower frame has no language for, and it is the part that policy makers can in principle deliberately deploy. That is a smaller claim than "structural interventions work" and a bigger claim than "all public policy fails".

The meta-pattern

The pattern across Iceland, Medellín, the UK smoking story, the sugar tax, Scottish MUP, and choice-architecture work is the same. The intervention that moves the dependent variable changes the environment around the person rather than the person's information set. The educational layer continues to exist — warning labels, school programmes, public-information campaigns — and it does the upstream work of making the diagnosis legible and the help-seeking acceptable. But it is not what does the heavy lifting at population scale.

The willpower frame on the Steelman page has no language for this. Its operational implication is always "tell people more clearly" or "raise the moral standard". The empirical case across the structural interventions on this page is that the route to population behaviour change is upstream of where the willpower frame is looking. The smoker quit because the workplace went smoke-free and the price went up and the pack went plain. The kid in Reykjavík did not become a drinker because the youth club was free and the streetlights were on at 10pm and the parents knew where the kid was. The boy in Comuna 13 did not get recruited because the library-park gave him somewhere to be and the Metrocable gave him somewhere to go.

None of these are educational interventions. All of them are structural. All of them worked.

The next page documents the specific country experiments in detail. The Industry page after that documents the political economy that resists structural intervention because the failure of the willpower frame is the addiction industrial complex's revenue base.