Does it work?

In short

What this page finds

• Few single measures have both their own cost and a causal study. Those that do number 13, for €111.8bn or 6.1% of public money, €1,845bn in 2024. Of that cost, €77.6bn is the general cuts in employer contributions, whose studies cover their introduction in the 1990s and the CICE's conversion in 2019, not the current scheme; without that one measure the share is 1.9%. Of the 155 official goals set for these lines, a causal study measures 50 and touches 11 more only indirectly. Counting a broad line, such as old-age pensions or primary education, in full as soon as one study touches one of its rules gives an upper bound: 82.3% of the money; defence counts because of studies of the end of conscription. What the rules studied inside spending lines cost is not published, so the share of money evaluated lies somewhere between the two.
• Whether effects move or last is rarely known. Where leakage is looked for, most studies find it. The forms found are listed in the table, landlords' share of housing benefit among them; who ends up with the money among those the law targets, workers through wages or parents through smaller transfers, is recorded apart, as a fact. Permanence is asked only where the aim is a lasting change, not of insurance, pensions or minimum incomes; some study looks at it in broad lines holding 17.5% of the money, and in single measures holding 0.1%.
• Measured against their own aims, the ten schemes evaluated here show two descriptive changes and two results that exclude zero and pass their design's checks; other results exclude zero but fail a check or have none, and most outcomes move by amounts the designs cannot tell from zero. In January and February 2026, when the hiring aid for higher-education apprentices had lapsed until the year's finance law and decree, their contracts moved by −38.1%, estimated relative to secondary ones, while large firms lost it at every level; since the 2023 pension reform, employment at 60-64 has risen by 5.5 pts, estimated more than at 55-59, which this site's comparison cannot attribute to the reform alone. Both are descriptive; neither design measures the scheme's effect. A result counts here when its 95% interval excludes zero and it passes its design's checks: no gap before the change, and a placebo that finds nothing. Two do. By regression discontinuity at the cutoff of the fraction cible of the dotation de solidarité rurale, €324m, estimated in 2019, where the grant comes and goes from year to year, communes saved the first-year euro in 2016-2019: their self-financing rose by 1.36, estimated euros per euro. A commune must balance its operating budget and may borrow only to invest, and a receipt that may not return the next year is one it can set aside. By regression discontinuity at the DETR cap, equipment spending rose by €50, estimated per inhabitant in the first year against each commune's own past, where the placebo finds +€0.5, estimated; in level the jump is €54, estimated, half of which appears where no rule changes, and over three years the design cannot tell the State's leverage from zero. Other results exclude zero but fail a check or have none. In the communes that entered the rural tax zone in 2017, private jobs moved by −7.4%, estimated against matched communes, about as much as the pairs' jobs had drifted apart before 2016; population moved by −1.4%, estimated between the 2015 and 2021 censuses and net migration by −11.1, estimated per 1,000 inhabitants, falls their unclassified neighbours share, which points to the area rather than the zone; income per household fell in the placebo before 2017 too. In the communes of Territoires d'industrie, industrial jobs moved by −6%, estimated over the same censuses, a difference no placebo can check. Action cœur de ville's towns lost employers, shops and services, but fewer than 10 of 100 matching draws balance them, and the services around the small towns of Petites villes de demain were already falling before the programme. By event study, where Pinel was withdrawn from zone B2 in 2018, the median price of all dwellings sold, mostly existing ones, moved by −2.4%, estimated against zone B1, with one year before 2017 to test the trend and a fall in zone A, where no rule changed, that cannot be told from it; some estimators find fewer flats authorised in B2, but they find gaps of the same size where no rule changed. Communes under the DETR cap borrowed more in level, but not against their own past. France Relance winners published accounts less often than their matches, in a design whose placebo fails in 2018. In the communes added to the vacancy tax in 2024, vacancy fell against never-taxed communes no more than in the communes whose status did not change, and their recorded vacant dwellings rose as the files they are counted from changed. For the research tax credit, researchers per thousand in employment rose 9.4%, estimated above the synthetic France in 2008-2012, a result that cannot be told from zero (p 0.10, the smallest ten countries allow), and neither can business research. Nor can the vacancy rate of the communes added to the vacancy tax in 2024, which moved by 0.13, estimated per 100 dwellings against the communes taxed since 2013, whose rates rose in 2023, or the flats authorised where the zoning moved communes up. In the small towns of Petites villes de demain, private jobs moved by 0.010%, estimated. France Relance grant winners' turnover moved by 2.3%, estimated and the CO2 their facilities declare by 3.3%, estimated. For collèges that left priority education, the social position index of pupils' families moved by 1.4, estimated points. The data cannot tell a loss in their brevet pass rate from none, and the comparison at entry fails its placebo. Teachers' turnover by collège and firms' imports, which would measure two more aims, are not published school by school or firm by firm. A result that cannot be told from zero may also mean the data are too coarse to show an effect.
• Against countries that spend as much per head, France's life expectancy is above the fitted line and its learning scores below it. Life expectancy is 82.4 years against 81.5, estimated predicted, learning outcomes 507.4 against 512.3, estimated predicted.
• At the State's discount rate, an effect that lasts five years instead of ten keeps 53.9% of its value; a statistical life is worth €3,000,000 in 2010 euros, a tonne of CO2 avoided €256.

All public money

What is known of what it achieves

Public money here adds up, for 2024 and line by line, general government spending and the tax expenditures and exemptions from social contributions that serve public aims through the tax system, so that a euro spent and a euro of tax forgone are both counted; the tax credits paid out, which the national accounts already record as spending, are counted once, inside spending. Public money so defined comes to €1,845bn in 2024. For each line, the question is whether anyone has measured its effect against what would have happened without it. 82.3% of the money sits in broad lines that contain at least one causal study, an upper bound; 3.8% was reviewed without a causal design; for 13.9%, no causal study or review was found.

Each tax expenditure is costed by the Ministry of Finance with every other measure unchanged, and Voies et moyens, tome II, warns that their sum « est donc dépourvue de réelle signification »: the tax expenditures in this total are an order of magnitude, not an exact sum.

A line counts as studied as soon as one study touches it, and most studies cover one rule or one reform: defence counts because of four studies of the end of conscription, old-age pensions because of studies of the retirement-age rules, none of which measures what the pensions do for retirees' living standards, the first goal the law sets them. Counting only single measures with their own cost, tax expenditures, payable tax credits and exemption schemes, those with a causal study number 13 and hold 6.1% of the money; the rules studied inside spending lines have no separate cost, and that part is left unknown.

The largest lines are studied, but through one rule or one reform at a time. Four large lines have no causal study: interest on the debt (€58.9bn), general services of the administration (€53.8bn), research and development for the economy (€21.4bn) and basic research (€19.9bn). Some lines, such as debt interest, cannot be evaluated this way; others, such as research, could be in principle. The tax credits paid out sit inside spending lines the register does not identify, so their studies count among the single measures and not in these lines. Insee records the research tax credit as an investment grant or another transfer, and assigns a transfer to the function of the spending it funds where that is known, without saying which line holds this credit. If it is research and development for the economy, €7.8bn of that line is the research tax credit, which has causal studies, and the line's status here understates what is known of it.

The State's grants to local authorities, such as the DGF, the DETR and the DSR, weigh €0.1m in the register, on the line of transfers between levels of government: the accounts of general government are consolidated, so what one public body pays another disappears, and the money is counted where the communes spend it. The studies of these grants, this site's among them, sit on that line.

The register lists 338 causal studies, found by searching Crossref, OpenAlex, HAL and RePEc line by line; for 286 of them the quoted finding was checked word for word against the retrieved source. They are in studies.csv, the official goals of each line in objectives.csv, and the searches that found nothing in the gap notes beside them. Spending reviews are in evaluations.csv. A line with no causal study found may still have one the searches missed: some research departments' sites blocked automated access. 23 more rows rest on a structural or calibrated model's simulation: they stay in the register, ringed in the table below, and are not counted.

All public money

Impact, leakage, permanence

The three questions come from the method that certifies carbon offsets, which asks of a forest project whether its carbon would have been stored anyway, whether the logging it prevents moved elsewhere, and whether the trees will stay. Here they read each study against the goal it measures; they are not a verdict on insurance, pensions or transfers, whose aim is the income they pay. Impact: did the money change the outcome the study measures, against what would have happened anyway? Leakage: did the effect move elsewhere, through displaced activity, a windfall for those who would have acted anyway, or capture by someone the law does not target? Who ends up with the money among those the law targets, workers through wages or parents through smaller transfers, is recorded as a fact, not as leakage. Permanence: does the effect outlive the money? It is asked where the aim is a lasting change, such as investment, training, zones or parental leave. Counting a broad line in full whenever one study informs it, even about one of its measures, some study informs impact for lines holding 65.2% of the money, leakage for 42.9%, permanence for 17.5%. Counting only single measures with their own cost, the shares fall to 5.6%, 5% and 0.1%. Of the 155 official goals set for these lines, a causal study measures 50 and touches 11 more only indirectly.

The rule, for every row. A square takes a verdict colour only when the study measures a goal of the line, or the aim the law gives a tax expenditure, by a causal design. A study of another outcome, or one that only touches a goal, keeps its reading in the panel, draws a grey square and is not counted. Money that reaches people the law does not target, such as landlords under housing benefit or sellers under a tax incentive, is capture, a leakage; money that reaches those it targets, such as workers under a wage subsidy, tenants or the parents of grant holders, is who ends up with it, a fact. A study of a reform that cut the money is marked: its colour says what the cut did to the goal, not what the money does. A finding that rests on a structural or calibrated model's simulation is ringed and not counted. An effect counts when it is significant at 5%, the threshold this site applies to its own estimates; one study is read as no significant effect for that reason, the Contrat d'engagement jeune for young people out of contact with public services, whose 7 points of employment come at p = 0.09. One study gives one verdict on each question.

What the table says

Impact is often there, but rarely for the whole line. Most lines with studies show an effect somewhere: on family benefits, that childcare and parental-leave rules change mothers' work. These are effects of one rule, not of the line's whole budget. On pensions, most studies evaluate a reform that cut the line, and their squares carry the mark of a cut: most find that raising the retirement age kept people in work, the goal of lengthening working lives; two find that the cut lowered retirees' pension wealth or raised the debt of the least educated, against the first goal the law sets pensions, retirees' living standard. For unemployment insurance and minimum income, several studies find that recipients return to work later, adverse on the return to work that the State's employment programme sets unemployment insurance and on the incentive to work set for minimum incomes; others find no effect (see the table). None measures job seekers' income, the first goal the law sets unemployment insurance, which a study touches only through access to its rights, and four of the 27 on the line of minimum incomes measure poverty, purchasing power or homelessness, the first goal the law sets them. Sick-pay studies measure sick days: they touch the goal of guaranteeing workers' income and that of controlling sick-pay spending, and the table gives them no colour, since whether the extra absences were warranted is not measured. On housing benefit, the studies find the goal of lowering modest households' housing costs largely missed: « one additional euro of housing benefit leads to an increase of 78 cents in the rent paid by new benefit claimants, leaving only 22 cents available to reduce their net rent and increase their consumption » (Fack, 2006).

Few single measures have been studied for leakage; most studies that look for it find it. It takes four recurring forms. Windfall: for the zero-rate home loan, « 85% of the beneficiaries in a 4-year period would still have moved to own in the absence of the subsidy » (Gobillon and le Blanc, 2008). Displacement: in the urban enterprise zones, the job gains « résulte[nt] pour les deux tiers d'un déplacement d'activités économiques en provenance de zones non ciblées » (Rathelot and Sillard, 2008). Shifts between schemes: after the 2010 pension reform, « around one fifth in the decrease in public spending is offset by increasing expenses in other public insurance schemes » (Rabaté and Rochut, 2019). And capture by those the law does not target: for housing benefit, landlords, through rents (Fack, 2006, above); for the 2009 VAT cut on restaurants, « the reform mostly benefited owners of sit-down restaurants, who pocketed 41 percent of the tax cut » (Benzarti and Carloni, 2019). None of these offsets appears in the scheme's budget. Who ends up with the money among those the law targets is recorded apart, as a fact: for the CICE, « Wage gains load on incumbents in high-skill occupations » (Carbonnier, Malgouyres, Py and Urvoy, 2022); for student grants, « a substantial part of student scholarship benefits low-income parents » (Grobon and Wolff, 2022).

Permanence is almost never measured. Fewer than one study in ten where it applies says whether the effect outlives the money. Where they do, enterprise zones, job-placement help and speed cameras fade, while training for job seekers, intensive support for young people, early schooling and electronic monitoring last, and so does the wage loss of mothers after a longer parental leave (Lequien, 2012; Joseph, Pailhé, Recotillet and Solaz, 2013). It is not asked of insurance, pensions, housing benefits or minimum incomes, whose aim is the income they pay while they pay it.

Each square is one study, or one of this site's evaluations, coloured by what it found against a goal of the line; open a line to read the goal each study measures, the reform it studies and the phrase each verdict rests on. Studies are classified by this site from their quoted findings, in criteria.csv; a line counts a study tied to any of its measures or rules, so a verdict on one rule is not a verdict on the whole line. Goals counts how many of the line's official goals a study measures, from goals.csv: a goal is measured when a study has its own outcome, or its official indicator, as outcome, and touched when the studies measure only something it depends on or a related outcome, or only simulate it.

All public money

Spending and social outcomes, across countries

Studies measure what one scheme does. Another view sets whole functions against the outcomes they are for, across the 30 European countries in Eurostat's accounts: spending per head on health against life expectancy, on education against learning, on public order against homicides, and all spending against life satisfaction, from Our World in Data, and on social protection against the inequality taxes and transfers take off, from the OECD. France lives longer than its health spending predicts, 82.4 against 81.5, estimated, but its pupils score lower than its education spending predicts, 507.4 against 512.3, estimated, and its people are less satisfied with their lives than its total spending predicts, 6.6 against 7, estimated.

Reading the panels. Richer countries spend more on everything and do better on almost everything, so most of these slopes are wealth. Holding GDP per head constant, the link with life expectancy halves and is no longer significant; education and social protection keep a clear one: among countries equally rich, those spending more per pupil score higher, and those spending more on social protection take more off inequality through taxes and transfers, where France ranks 3 / 28. The OECD counts public pensions among transfers, so countries whose pensions are mostly public take more off; the level of inequality, which also depends on market incomes and on age, would not measure what the spending does. Homicides and air pollution barely follow the spending on them.

France. None of these comparisons says what any euro does; they show where France stands among countries that spend alike.

Inside France, by département

The same question across the 96 metropolitan départements, with what is published by territory: what communes spend per head, and the State's operating grant to them (DGF), against school results, job growth and income. The DGF goes where income is low, as its equalisation goal intends. Commune spending goes with job growth even at equal income. But what neither explains is itself clustered in space (Moran's I, last column): neighbouring départements do better or worse together, so regional factors, such as the growth of the south and west, may explain these associations, which do not measure the spending's effect.

Each panel puts spending per head on the function, on a log scale, against its outcome; the line is fitted across the 30 countries. Countries: 2015-2022 averages; spending per head is the share of GDP spent on the function (Eurostat, by COFOG, the UN classification of government spending by purpose) times GDP per head at purchasing-power parity (World Bank, via Our World in Data). Inequality taken off is the OECD's Gini of market income less its Gini of disposable income, total population (Income Distribution Database), for the 28 countries it covers, Cyprus and Malta aside. The correlation is with the log of spending per head; the slope given GDP per head holds national income constant, with its standard error. Départements: 2019-2022 averages of commune accounts (OFGL, the observatory of local public finance), income per tax household (DGFiP, the tax administration), brevet results 2016-2019 (DEPP, the education ministry's statistics office) and private jobs (Urssaf, which collects social contributions); Moran's I uses inverse-distance weights within 250 km, with p from 999 random reshufflings. These are associations, not effects. The code is analysis/outcomes.ml and the data are in results.csv and departements_results.csv.

All public money

What the State says social outcomes are worth

Much public spending aims at outcomes that have no market price, such as health or air quality. The State has its own price list for such outcomes, which every public investment over €20m must use in its socio-economic appraisal: a statistical life is worth €3,000,000, a year of life €115,000, an hour of commuting €10, and a tonne of CO2 avoided €256 in 2025, rising to €300 in 2030.

With these values, an effect measured by a causal study becomes euros of social value, to set against what the line costs, such as lives saved by a health programme or emissions avoided by a subsidy. Health products are assessed in cost per quality-adjusted life year, but the Haute Autorité de santé sets no threshold:

En l’absence de valeur(s) de référence définie(s) pour la France, les résultats sont présentés graphiquement dans un plan bénéfice net / dispositions à payer.HAS, Choix méthodologiques pour l'évaluation économique (2020)

What a temporary effect is worth

At the State's discount rate of 3.2% a year, a benefit that lasts T years is worth 1 − (1 + r)^−T of the same benefit kept for ever. Five years are worth 14.6%, ten years 27%, twenty 46.7%, so an effect that lasts five years instead of ten keeps 53.9% of its value. This answers the third of the three questions, permanence, in value: with an impact measured and its decay known, the value of a scheme is its effect, priced at the values above, times the share its duration keeps. Of the studies in the register, only those classed as lasting or fading measure duration at all.

Wellbeing has no French value. The UK Treasury values one point of life satisfaction for one person for one year at £13,000. In France, life satisfaction is one of ten indicators the law asks the government to report beside GDP each year, with no price attached:

Values for lives and time are in 2010 euros and grow with GDP per head; the carbon values are the 2025 revision, in 2023 euros. The indicators are Insee's tables for the 2024 report of loi n° 2015-411; the list of ten is the one of the Government's first report, of October 2015. Sources, pages and quotes are in SOURCES.md.

Method

The unit follows the money

To know what a euro of public money achieves, compare what it paid for with what would have happened without it. The comparison needs units that received the money and similar units that did not, and the unit is whatever the money is assigned to: a commune for a zoned tax break, a country for a national tax credit, a level of diploma for a hiring aid, a school for teaching resources.

Every design carries a test that should find nothing (a placebo), and the page reports when it fails. For matched designs the intervals are 95% confidence intervals with clustered standard errors; the spread across matching draws, which the method borrowed from carbon-offset evaluation reports, is given beside them. Synthetic controls are ranked among placebo countries, and the apprenticeship bands come from the noise between years, as each section says.

Evidence

Evaluated by this site

This site evaluates ten schemes, each on its own page against a comparison group, plus one related study of mayors' party.

A scheme is evaluated here when open data exist at the unit the money is assigned to, and a rule or a dated change sets some units apart from similar ones: a threshold, a zone, a cut or an extension. The first seven studies were chosen on that test in early October 2026. The next came from a ranked list of twenty-six candidates, scored on the money at stake, the credibility of the design and the readiness of the data: the studies of the State's grants to communes, the housing zones and the tax on vacant dwellings, run in October 2026, cover its first five ranks and its eighth. Schemes with no such data or rule are not evaluated here.

• ZRR

  One tax break, evaluated here

  Whether exempting firms from profit tax in the rural revitalisation zones, replaced by France Ruralités Revitalisation in 2024, created jobs in the low-density, low-income communes the zones targeted.

• Research tax credit

  The research tax credit, against a synthetic France

  Whether the research tax credit, the largest tax expenditure, raised business research in France.

• Apprenticeships

  The apprenticeship hiring aid, by level of diploma

  How apprenticeship contracts moved when the hiring aid changed, by level of diploma.

• Priority education

  Priority education, collège by collège

  Whether the extra teachers and smaller classes of priority education raise collèges' results.

• France Relance

  France Relance grants, firm by firm

  Whether France Relance industrial grants changed the size or profitability of the firms that won them.

• Towns

  Programmes for towns and territories

  Whether the State's programmes for towns and territories changed jobs and income in the communes they chose.

• Vacant dwellings

  The tax on vacant dwellings, extended in 2023

  Whether taxing vacant dwellings in the communes added to the tax from 2024 brought empty dwellings back into use.

• Housing zones

  The housing zones behind Pinel and the PTZ

  Whether the housing zone a commune is classed in, which opens or closes the Pinel tax reduction and the zero-rate loan for new homes, changes how many flats it authorises.

• Local grants

  The State's grants to communes, at their thresholds

  Whether the State's grants to communes, the DETR for investment and the DSR and DSU for rural and urban solidarity, change what communes invest, spend and save, measured where a published rule decides who receives them.

• Retirement age

  Retirement age, by age band

  How employment at 60 to 64 moved as the minimum retirement age rose under the 2023 reform, now partly suspended.

• A related study: who runs the town

  Mayors' party and commune finances

  Whether a left or a right mayor changes a commune's spending, debt and jobs.