Skip to content
SVERIGEFAKTA

Method, neutrality and evidence

How SVERIGEFAKTA works

Statistics about Sweden from official sources. We are not a public authority: we compile statistics and verifiable sources, separate observation from processing, model and analysis, report uncertainty and correct openly. Method and sources must be open to scrutiny.

Hold us to our word.

Something that does not add up? Have we missed a relevant source, used a definition incorrectly or drawn a conclusion that goes beyond the evidence? Email us.

We prioritise substantive objections that can be checked against a source, definition, method or data.

Report an error or objection

n824euiy@anonaddy.com

Principle 0 — overriding everything else

When a rule, an expectation or a narrative collides with what the evidence shows — follow the evidence.

The rules serve the truth, not the other way round.

1

What neutrality means here

Neutrality is a methodological rule, not a requirement to give every claim equal weight. The same evidence standard, the same language standard, the same definition requirements and the same uncertainty requirements apply regardless of which direction the result points in.

The same requirements in both directions

A result that confirms a common belief is scrutinised as hard as one that contradicts it. The same rule applies regardless of which direction the result points in.

Not false balance

Neutrality does not mean that a poorly supported claim should be given the same space as a well-supported one. Weight follows the evidence.

Neutrality principle.

SVERIGEFAKTA compiles public statistics, research and other verifiable sources. Official data, in-house calculations, models and analysis are labelled and kept clearly separate. Sources and methods must be open to scrutiny by the reader. The site does not recommend policy and does not attribute motives to authorities or other actors that are not directly supported by the source. Analysis and conclusions occur, but must never be stronger than the underlying evidence.

The symmetry test

Would we phrase and judge the same result in the same way if the outcome pointed the opposite way?

If the answer is no, the wording changes — not the result.

2

Evidence ladder and source hierarchy

The main rule is source fit: a source must be the right source for the specific claim being made. A secondary source can be entirely correct as the source for its own analysis, its own statement or its own synthesis — but original numerical data is taken from the original source whenever one is available.

  1. 1. Original source for the data point or definition

    The body that actually produces the figure or sets the definition. Where available, it is used.

  2. 2. Official statistics and the responsible authority

    The authority responsible for the statistics, or an official database with documented method and version.

  3. 3. Original research and the responsible research institute

    A published study or report where method and data can be reviewed.

  4. 4. Authoritative secondary synthesis

    Evidence reviews and compilations — when the synthesis itself is what the claim is about.

  5. 5. Editorial media

    Used for the outlet's own published information, quotes and accounts of events — not as a substitute for an available original source for a numerical data point.

  6. 6. Opinion material

    Used only to describe what an actor holds to be true, never as evidence for a matter of fact.

  7. Never: AI-generated material

    AI is never evidence and never the sole source for numerical facts.

3

Identity of a data point

A number without an identity is not information. The following must be known and stated for every figure we publish.

Metric
what is measured, with an unambiguous name
Definition
which definition applies, and who set it
Population
who is included
Denominator
what the share is calculated on
Period
which year or period the value refers to
Geography
country, county, municipality or area
Unit
count, share, SEK, per 100,000 inhabitants
Method
observed, derived, modelled — with the operation stated when it is not trivial
Data status
final, preliminary or revised
Source and version
publication, table, version and update time where available

Missing data is not zero

Years without a published figure are shown as gaps, never as zeros or interpolated guesses.

Series breaks are not hidden

When a definition or recording practice changes, the break is noted on the series instead of being smoothed away.

Incompatible series are not merged

Series with different definitions, populations or denominators are not added together or shown as a single curve.

4

Labelling the type of knowledge

Every item of information belongs to a type of knowledge. Neither text nor interface may make a model, a scenario or an interpretation look like observed statistics.

Observed
Published value from the responsible source, unchanged.
Preliminary
Published but not yet final at the source; may be revised.
Derived
A pure arithmetic operation on published figures — ratio, share, price conversion. No new information is added.
Model
An estimate of a value that is not published. Based on assumptions stated in the model card.
Forecast
The source's own assessment of future outcomes, with the source's uncertainty range.
Scenario / projection
A calculation under given assumptions. It carries no probability and is not a prediction.
Analysis / interpretation
Our reading of the evidence. Labelled as interpretation and never stronger than the evidence.

5

Measurement, analysis, valuation and policy

Four different kinds of claim that are often conflated. We keep them apart in both text and interface.

Measurement

What the figure is. Can be checked against the source.

Analysis

What the figure reasonably means, given the evidence. Labelled as interpretation and states its limitations.

Valuation

Whether the development is good or bad. That judgement is the reader's, not ours.

Policy

What ought to be done. External actors' positions may be reported with clear attribution, but never become a factual conclusion from us.

6

Causality

Two quantities moving together does not mean one causes the other. Causal language is used only when the evidence actually supports a causal relationship, and then with the uncertainty the source states. Otherwise we describe correlation as correlation.

7

Uncertainty and counter-evidence

Uncertainty is written out, not written away. Relevant counter-evidence must not be hidden — but presenting a well-supported result and a poorly supported one as equivalent is also an error.

  • Margins of error and confidence intervals where the source states them
  • Small populations and sample sizes
  • Short time series and single observations
  • Series breaks and changed definitions
  • Non-response and incomplete reporting
  • Definition differences between sources and countries
  • Revision risk in preliminary values
  • Sensitivity analysis and alternative reasonable interpretations

8

In-house calculations, models and projections

Anything that is not directly sourced official statistics must be traceable from data to operation to output. Every active item has a model card with sources, assumptions, limitations and reproducibility status.

Methodological rule: a population definition or volume parameter must never be multiplied through to crime, crime share, socioeconomic cost, net contribution or welfare dependency. The socioeconomic costs of crime are not included in the net-contribution calculations and are therefore not quantified here.

Derived values

A derived value is a pure arithmetic operation on published figures — ratio, share, price conversion or conversion to another time unit. No new information is added.

Own calculationDerivedReproducible — verified in test

KI 117 — publicerad beräkning och alternativ beräkning

Show the National Institute of Economic Research's (NIER) published net contribution for 2022 as the default, and an alternative published calculation as an explicitly labelled alternative.

Data → operation → output
NIER's published totals and subgroups for 2022; Sanandaji's published total for 2022 → No in-house arithmetic: each published figure is shown unchanged under its own label. Subvalues that have not been published are returned as null. → /, /historik, /jamforelser, /lonsam-for-vem, Faktamotorn
Primary sources
Assumptions
  • None. Only published figures are reproduced.
Observed vs modelled
Both modes are published third-party calculations. SVERIGEFAKTA does not model its own value and never mixes the modes.
Uncertainty
The difference between the modes is a methodological difference between two published calculations, not a margin of error.
Limitations
  • Only 2022 is available in the alternative mode.
  • Subgroups that have not been published in the alternative mode are missing and are shown as a missing value, never as zero.
Scope of validity
Sweden, reference year 2022.
Reproducibility
Deterministic tests reproduce the published figures for both modes and fail closed on missing subvalues.
Version
2.0 (Amendment 2B) · last verified 2026-09-13
Code/data reference
src/lib/ki117/calculation.ts

Open the model card as its own page →

Own calculationDerivedReproducible — verified in test

DeSO-skalning av stadsdelsandelar

Show neighbourhood-level figures for years other than Statistics Sweden's DeSO reference year.

Data → operation → output
Statistics Sweden's exact DeSO value for the reference year; The municipality's series for the current year → DeSO share(year) = exact DeSO share(reference year) × municipality value(year) / municipality value(reference year). → /karta
Primary sources
Assumptions
  • The neighbourhood's relationship to the municipality is assumed unchanged over time.
Observed vs modelled
Values are labelled as exact or scaled. Where there is no underlying data, no figure is shown — never zero.
Uncertainty
Scaled values can deviate sharply in neighbourhoods with rapid population turnover.
Limitations
  • Only Statistics Sweden's reference year is exact.
Scope of validity
Sweden's DeSO areas.
Reproducibility
The formula is a pure ratio of two published values and is tested deterministically.
Version
1.0 · last verified 2026-09-14
Code/data reference
src/lib/deso.ts

Open the model card as its own page →

Own calculationDerivedReproducible — verified in test

Statsskuldens kostnad per sekund

Make the annual interest cost understandable as a rate.

Data → operation → output
The National Debt Office's latest published annual cost of the national debt → Cost per second = annual cost / 31,557,600 seconds (365.25 days). → /ekonomi/statsskuld, /embed/statsskuld
Primary sources
Assumptions
  • The cost is spread evenly over the year.
Observed vs modelled
The annual figure is observed; the per-second rate is a derivation and is labelled as a model.
Uncertainty
None. An arithmetic conversion of a published annual figure.
Limitations
  • Money does not move in real time — the rate is an illustration.
Scope of validity
Sweden, latest published year.
Reproducibility
A division of a published figure; tested.
Version
1.0 · last verified 2026-09-14
Code/data reference
src/lib/statsskuld/contracts.ts

Open the model card as its own page →

Own calculationDerivedReproducible — verified in test

Kostnadsklockan (KI:s årsnetto per sekund)

Show NIER's published annual net for the refugee and relatives group as a rate.

Data → operation → output
NIER Special Study 117: net cost for the refugee and relatives group, 2022 → Per second = annual value / 31,557,600. Conversion to current-year prices is done using the CPI. → /metod, kostnadsklockan
Primary sources
Assumptions
  • Even distribution over the year.
  • CPI conversion when the user selects current prices.
Observed vs modelled
The annual figure is published; the rate and the price conversion are derived.
Uncertainty
Applies to a single year and one subgroup.
Limitations
  • Other immigration groups had a positive net the same year and are not included in the clock.
Scope of validity
Sweden, reference year 2022.
Reproducibility
Division and CPI conversion of published figures.
Version
1.0 · last verified 2026-09-14
Code/data reference
src/routes/metod.tsx (#kostnadsklocka), src/data/figures.ts

Open the model card as its own page →

Own calculationDerivedReproducible — verified in test

Inflationsjustering av belopp

Convert amounts between years for comparability.

Data → operation → output
CPI series; Amount and base year → Adjusted amount = amount × CPI(target year) / CPI(base year). → Belopp med prisomräkning, bl.a. kostnadsklockan
Primary sources
Assumptions
  • For the current year, the Riksbank's published CPIF forecast is used as an anchor, which is clearly labelled.
Observed vs modelled
Historical CPI figures are observed; the current year's anchor is a published forecast.
Uncertainty
The forecast anchor for the current year may be revised.
Limitations
  • The CPI reflects the household basket, not the price development of public spending.
Scope of validity
Sweden.
Reproducibility
A ratio between two published index values; tested.
Version
1.0 · last verified 2026-09-14
Code/data reference
src/lib/inflation.ts

Open the model card as its own page →

Models

Models estimate values that are not published. They are tools for understanding sensitivity — not predictions of outcomes.

Model estimateModelledReproducible — verified in test

KI-metodsimulator (offentligfinansiellt netto)

Let the reader see how NIER's result changes when assumptions about the employment gap, discount rate and time horizon are changed.

Data → operation → output
User-selected assumptions (employment gap, discount rate, horizon, tax and transfer levels); NIER's published defaults for 2022 and 2025 → Annual net per person = tax paid(employment) − transfers − consumption of public services (age-weighted). Present value = the sum of annual net amounts discounted at the chosen rate over the chosen horizon. → Metodsimulatorn i KI-avsnitten
Primary sources
Assumptions
  • Domestic employment rate of 82% as a reference.
  • Constant tax and transfer levels over the horizon.
  • The age composition is summarised in a single factor rather than a full cohort model.
Observed vs modelled
The defaults reflect NIER's published results; anything the user changes is modelled and not published by NIER.
Calibration
The defaults are calibrated so that the NIER-2022 mode reproduces NIER's published order of magnitude.
Uncertainty
The result is very sensitive to the discount rate and the employment gap. No confidence interval can be given.
Limitations
  • Does not replace NIER's full register-based model.
  • Cannot be used for individual municipalities or countries of birth.
Scope of validity
Sweden, the whole public sector, long-run average per person.
Reproducibility
The formula sits in a pure function, `simulateKi`, and gives the same output for the same input in tests.
Version
1.0 · last verified 2026-09-14
Code/data reference
src/data/ki-metod-model.ts

Open the model card as its own page →

Model estimateModelledPartially reproducible

What-if-simulator (volym, arbetskraftsandel, integrationskrav)

Illustrate how the public-finance net and the municipal gap change with different assumptions about the volume and composition of immigration.

Data → operation → output
Annual volume; Labour-force share; Integration requirement; Municipal cost coverage; Horizon → Per person: labour-force share × (+SEK 30,000) + asylum share × (Ruist anchor × integration factor × congestion factor). Annual net = volume × per person. Municipal gap = asylum volume × SEK 12,000 when the flat rate applies. → /what-if
Primary sources
Assumptions
  • Ruist's observed net flows from 2015 are assumed to persist.
  • An integration requirement lowers the per-person cost by up to 30% — an assumed elasticity, not an estimated relationship.
  • High volumes raise the per-person cost by up to 15%.
Observed vs modelled
Only Ruist's range (SEK 53,000–94,000) is published. The elasticities are model assumptions.
Calibration
The anchor is set to Ruist's midpoint value of SEK 74,000 per person per year.
Uncertainty
Reported as a low/mid/high outcome from Ruist's published range.
Limitations
  • The socioeconomic costs of crime are not included in the calculation and are therefore not quantified here.
  • The composition index for the second generation has been removed: it lacked a verifiable weighting method.
  • Is an educational tool, not a forecast.
Scope of validity
Sweden, non-EU/EEA immigration, 1–20 year horizon.
Reproducibility
The arithmetic is deterministic and tested, but the elasticities cannot be reproduced from a published source.
Version
2.0 (P1: brottskostnad och generationsindex borttagna) · last verified 2026-09-14
Code/data reference
src/data/whatif-model.ts

Open the model card as its own page →

Model estimateModelledPartially reproducible

Ackumulerad kostnad per ankomstkohort

Show the order of magnitude of the accumulated public-finance net per person after 10 years, 20 years and over a lifecycle.

Data → operation → output
Ruist ESO 2018:3; NIER's cohort results; Aldén & Hammarstedt → Manually compiled table: published per-person values by cohort, projected linearly to 10/20-year accumulation where the source gives an annual net. → /kohorter, /kohorter-over-tid, /api/kohorter.json
Primary sources
Assumptions
  • The annual net is assumed constant within each interval.
Observed vs modelled
Rows based on a projection are labelled as an in-house calculation/forecast in the table; the rest are published values.
Uncertainty
No margin of error can be given — the spread between the sources is larger than the rounding.
Limitations
  • Applies to averages, not individuals or municipalities.
Scope of validity
Sweden, refugee and relative cohorts from the 2010s.
Reproducibility
The values are static and traceable to source, but the exact projection per row is not coded as a formula.
Version
1.0 · last verified 2026-09-14
Code/data reference
src/data/kohorter.ts

Open the model card as its own page →

Model estimateModelledReproducible — verified in test

Skattekalkylator (brutto till netto)

Show the approximate breakdown between municipal tax, regional tax, national income tax and employer contributions.

Data → operation → output
Monthly salary; Standard parameters for the current income year → A standard calculation using the average municipal and regional tax rate, the threshold for national income tax, and a standardised earned-income tax credit. → Skattekalkylatorn
Primary sources
Assumptions
  • Average municipal tax rate instead of the individual municipality's rate.
  • No deductions beyond the earned-income tax credit.
Observed vs modelled
The tax rates are published; the combination is simplified.
Uncertainty
Differs from the Swedish Tax Agency's exact calculation, especially with deductions and additional income.
Limitations
  • An educational tool, not a tax assessment notice.
Scope of validity
Sweden, the current income year.
Reproducibility
A pure function with documented parameters; tested.
Version
1.0 · last verified 2026-09-14
Code/data reference
src/lib/skatt.ts

Open the model card as its own page →

Forecasts and projections

Projections and scenarios carry no probability and are not forecasts of what will happen. Scenarios from different sources use different base years and definitions and must not be added together.

ProjectionModelledReproducible — verified in test

Kommunframskrivning av utländsk bakgrund (Modell 1)

Fill in years outside Statistics Sweden's observed series in the map's timeline (1924–2124).

Data → operation → output
Statistics Sweden's municipal series on foreign background; Latest observed year per municipality → Before the first observed year: an S-curve towards a historical base level. After the last observed year: a growth curve with a cap. Within the series: linear interpolation between observed points. → /karta, /lokal-framskrivning, /befolkningspyramid
Primary sources
Assumptions
  • The municipality's historical trend continues, dampened towards a cap.
Observed vs modelled
Every year is labelled in the interface as observed, modelled history, modelled future or modelled gap.
Calibration
The curve is anchored to the municipality's observed Statistics Sweden values.
Uncertainty
No probability distribution. The projection is deterministic and illustrative.
Limitations
  • Is not Statistics Sweden's official population projection.
  • Municipalities without observed values get a deterministic fallback that is labelled as modelled.
Scope of validity
290 municipalities, the years 1924–2124.
Reproducibility
A pure function with no randomness; the same municipality and year always give the same value in tests.
Version
1.0 · last verified 2026-09-14
Code/data reference
src/lib/demographics.ts

Open the model card as its own page →

ProjectionModelledReproducible — verified in test

Scenariotabeller 2050 och 2060

Reproduce published scenarios from Statistics Sweden, LU2023, the Swedish Fiscal Policy Council, and Pew.

Data → operation → output
Published scenarios from each respective source → No in-house calculation: published scenario values are reproduced unchanged with source and scenario name. → /makro-framtid-2060, /framskrivning-ursprung-2050, /framtidsscenarier
Primary sources
Assumptions
  • Each source's own assumptions apply and are reported per scenario.
Observed vs modelled
Everything is a modelled scenario from the source — none of it is an observed outcome.
Uncertainty
Scenarios are not forecasts and carry no probability.
Limitations
  • Scenarios from different sources have different base years and different definitions and should not be added together.
Scope of validity
Sweden and Europe respectively, horizon 2050/2060.
Reproducibility
The values can be checked row by row against the respective publication.
Version
1.0 · last verified 2026-09-14
Code/data reference
src/data/makro-framtid.ts, src/data/framskrivning-ursprung-2050.ts

Open the model card as its own page →

ProjectionModelledReproducible — verified in test

Kostnadsresan — linjär framskrivning 2023–2040

Extend the published net-contribution series for the foreign-born beyond the last outcome year as an arithmetic exercise, clearly separated from the actual outcome.

Data → operation → output
The last published outcome year in the net-contribution series (2022) → value(year) = round(value(2022) × (1 + 0.01 × (year − 2022)), 1) for the years 2023–2040. No other input is used. → Kostnadsresan (nettobidragets historia)
Primary sources
Assumptions
  • A flat mark-up of 1 percent per year. The mark-up does not follow from the source but is a chosen calculation parameter.
Observed vs modelled
1983–2022 is published outcome data. 2023–2040 is entirely modelled and drawn as a separate dashed series.
Uncertainty
The projection has no confidence interval and no probability. It says nothing about the actual development after 2022.
Limitations
  • A constant percentage change captures neither the business cycle, compositional change nor policy changes.
  • Must not be read as a forecast and must not be compared with published forecasts.
Scope of validity
Sweden, the foreign-born group, the years 2023–2040.
Reproducibility
A deterministic formula with no hidden parameters; reproduced in tests from the last outcome year.
Version
1.0 · last verified 2026-09-15
Code/data reference
src/components/cost-journey-timeline.tsx

Open the model card as its own page →

ProjectionModelledNot reproducible — HOLD

Lokal framskrivning — kommunandelar och skolsegregationsindex

Show municipality-level population shares to 2050 and a school-segregation index over time on the local-projection page.

Data → operation → output
Municipal population and share values for 2024 and 2050; School-segregation index per measurement year → No in-house arithmetic in the code: the values sit as static table rows and are reproduced unchanged. → /lokal-framskrivning
Primary sources
Assumptions
  • That the static table values correspond to each authority's published main scenario.
Observed vs modelled
2024 is stated as outcome and 2050 as a projection by the source. SVERIGEFAKTA does not model its own value.
Uncertainty
Municipality-level projections carry greater uncertainty than the national level and are reported here without an interval.
Limitations
  • The selection of municipalities is editorial and is not a random or comprehensive sample.
  • The municipal shares refer to the foreign-born and must not be read as foreign background.
Scope of validity
A selection of Swedish municipalities, the years 2024 and 2050.
Reproducibility
HOLD: the individual table values are not tied to a named table or extract from Statistics Sweden or the National Agency for Education in the repository, so an external reviewer cannot follow data → operation → result. The values must be tied to an exact source table or be unpublished.
Version
1.0 · last verified 2026-09-15
Code/data reference
src/data/lokal-framskrivning.ts

Open the model card as its own page →

ProjectionModelledPartially reproducible

Avslutande extrapolationer på WTF-sidorna

Answer the closing question 'if the trend continues' on the theme pages with a mechanical projection of the series shown.

Data → operation → output
The series shown on each theme page, with a stated start and end year → A linear extrapolation of the slope of the most recent trend period to the target year. The result is given as a range and is labelled in the text as a mechanical extrapolation. → Temasidornas avslutande avsnitt
Primary sources
Assumptions
  • That the slope of the most recent period continues unchanged — an assumption with no source support.
Observed vs modelled
All years up to the series' last published year are observed. The target-year value is entirely modelled.
Uncertainty
No probability and no confidence interval. The range expresses the sensitivity of the slope, not uncertainty in a statistical sense.
Limitations
  • A linear slope does not capture turning points; where later outcomes point in another direction, this is noted in the same paragraph.
  • The extrapolations are free text per page and are not automatically recalculated when the series is updated.
Scope of validity
Each theme page's series, geography and period.
Reproducibility
The slope can be recalculated from the published series, but the value is written as text and is not generated by code, so there is no deterministic test that ties the text to the series.
Version
1.0 · last verified 2026-09-15
Code/data reference
src/data/wtf-pages.ts, src/data/wtf-pages-extra.ts

Open the model card as its own page →

9

Reproducibility

Status is set by whether the model's output is actually reproduced from documented sources and operations in a deterministic test. A fetch returning HTTP 200, or a file that parses, is not the same as verified data.

VERIFIED · 10

Output is reproduced from documented sources and operations in testing.

PARTIAL · 3

Parts of the chain are verified, but not the whole way from source to output.

HOLD · 1

Not reproducible today. HOLD is not a passing state and is never shown as green.

Off 14 active items in the model registry, 10 are verified, 3 partially reproducible and 1 on HOLD.

10

From source to published page

Two flows: an automated one for data ingestion and an editorial one for manual content.

Automated data flow

  1. Step 1

    Retrieval

    Data is retrieved from the source's published interface or file.

  2. Step 2

    Parsing

    The content is parsed into a structured format.

  3. Step 3

    Schema validation

    Fields, types and required keys are checked.

  4. Step 4

    Semantic validation

    Definition, population, period and unit are checked against the contract.

  5. Step 5

    Sanity check

    Anomalies and implausible jumps are flagged instead of published.

  6. Step 6

    Diff and staging

    The difference from the previous version is reviewed before publication.

  7. Step 7

    Publication

    The new version is published with source and update time.

  8. Step 8

    Post-verification

    The published value is checked against the source after publication.

If the chain fails, no new figure is published. The system keeps the last known good value and flags it for manual review.

Editorial flow for manual content

  1. Step 1

    Source

    The right source for the claim is identified using the evidence ladder.

  2. Step 2

    Definition check

    Definition, population, period and unit are verified against the original document.

  3. Step 3

    Processing

    In-house calculations are documented with formula, assumptions and sensitivity.

  4. Step 4

    Counter-check

    Counter-evidence and alternative interpretations are actively sought before publication.

  5. Step 5

    Publication

    The page is published with source, date and clear labelling of the type of knowledge.

  6. Step 6

    Version history

    Every change is dated and described in the changelog.

11

Correction, revision and update

Three different things that are often given the same name. We keep them apart, and all three are dated in the changelog.

Correction

We were wrong. The information did not match its source, or the conclusion went beyond the evidence. The error is described openly.

Revision

The source has changed a previously published value. We follow the source and note that the value has been revised.

Update

A new period or version has been published. No earlier value was wrong.

We never overwrite earlier content without a trace. Everything is in the the changelog.

12

Independence

Facts are not adjusted to suit a client, donor, sponsor, search engine optimisation or revenue target. External funding confers no right to influence selection, wording or conclusions. Only information that has been checked and holds today is stated below.

Who runs the site
The site is run privately by a small editorial team that currently does not publish its names, for security and privacy reasons. The team belongs to no political party, public agency or media group.(checked 2026-09-14)
Funding
Hosting and development are currently paid for by the editorial team itself and by voluntary donations. The site carries no advertising, no paid content, and accepts no sponsorship or public grants.(checked 2026-09-14)
Donations
Voluntary contributions are currently accepted in cryptocurrency. We keep no donor register. A donation grants no access, influence or insight into the content.(checked 2026-09-14)
Advertising, membership and API
There is currently no advertising, no membership and no paid API access on the site. If any of this is introduced it will be documented here before it goes live.(checked 2026-09-14)
Editorial independence
No external party reviews, approves or commissions the content. Selection and wording are made by the editorial team, and every figure is published with its source so the reader can check it.(checked 2026-09-14)
Corrections policy
Verified errors are corrected in the open. The change is dated and described in the changelog, and earlier content is not overwritten without a trace. Method clarifications and versions are handled in the same log.(checked 2026-09-14)
Method principles
Official statistics, own calculations, models and analysis are labelled and kept apart. The method portal documents data, operation and output for every active calculation of our own.(checked 2026-09-14)
Privacy and analytics
Analytics are cookie-free and aggregated. The public domain loads two external analytics scripts, Fathom Analytics and Ahrefs Analytics; they set no cookies and do not identify individual visitors. No visitor data is sold or shared.(checked 2026-09-14)

13

Sensitive topics

Politically charged topics call for higher methodological discipline — not lower evidence standards, and not silence. Group-level figures say nothing about an individual, and correlation is not cause. Definition, population and denominator are always stated, because that is where such figures most often go wrong.

14

What we never do

  • We do not fabricate statistics or sources.
  • We do not replace a missing value with zero or a guess.
  • We do not hide series breaks or merge incompatible series.
  • We do not present a model or a scenario as observed statistics.
  • We do not pick the period, comparison year or population to produce a desired narrative.
  • We do not write a headline stronger than the evidence behind it.
  • We do not use AI as the sole factual source for numerical information.
  • We do not mix policy recommendations with the reporting of facts.

15

Technical method sections

Details on official statistics, definitions, data quality, version history, code and data.

Official statistics

The national time series are national aggregates taken directly from the responsible authority: the Swedish Police Authority, the National Council for Crime Prevention, the National Agency for Education and OECD/PISA, SALAR and IVO, Statistics Sweden's labour force survey and the Swedish Public Employment Service, the National Board of Health and Welfare, and the Migration Agency and Statistics Sweden. Such values are classified as observed and require no model card.

No interpolation: years without a published figure are shown as empty points, not guesses. A missing value is never zero. Definition changes (e.g. the Police Authority's registration practice from 2017, or the National Council for Crime Prevention's revised rape definition in 2005 and 2013) are noted under the relevant chart.

Definitions

The definitions of foreign background in Sweden, Denmark and Germany are handled by a canonical definition engine with the official definition, the responsible authority, a version, comparability flags, and the calculations that are permitted and explicitly not permitted for each definition. Read the full definition comparison. A difference in definition may only affect population counts and population shares — never behavioural or economic outcomes.

Uncertainty in the lifecycle calculations

Lifecycle results are sensitive to assumptions about future employment, income growth and welfare consumption. Ruist 2018 reports a range from SEK 53,000 to SEK 94,000 per refugee per year depending on integration potential. NIER 2025 reports sensitivity analyses for, among other things, the discount rate.

Key assumption: Ruist's SEK 74,000/year rests on the assumption that the observed net flows in 2015 persist over the entire lifecycle. It is an accounting extrapolation, not a forecast.

Cross-section vs lifecycle

Cross-section: a single year's net for everyone in the group that year — strongly affected by age profile and the business cycle. Lifecycle (NPV): the discounted present value of an average person's net over their whole life; NIER uses 1.5% as its base assumption and 2022 prices.

Data quality and data gaps

NIER 2025 and Ruist 2018 measure direct public-finance inflows and outflows. Below are items that the research identifies as relevant but that are not quantified in the underlying studies.

Important context

What is not included in the figures?

NIER 2025 and Ruist 2018 measure direct fiscal inflows and outflows. Several items that research and public debate treat as relevant are not quantified in the underlying studies. The direction of the net effect varies between items — see each card below.

  • Crime and the justice system

    Police, prosecutors, courts, corrections and victim costs are not allocated by group in either NIER or Ruist. Recurring reports from the National Council for Crime Prevention show that the foreign-born are overrepresented as suspects for several offence types — but no fiscal cost allocation by country of birth exists in the underlying studies.

  • Second generation

    Children of foreign-born parents count as native-born in NIER's register data. Their net contribution — positive or negative — does not appear in the 'foreign-born' figure but is hidden in the reference group.

  • Effects on native-born wages and employment

    Labour market competition, possible wage effects and changes in native-born employment are not included. These are dynamic effects; the NIER and Ruist analyses are static.

  • Housing market and overcrowding

    Housing allowance is included, but not indirect costs of residential segregation, rent regulation, public housing investment or the health and schooling effects of overcrowding.

  • Schools, segregation and learning outcomes

    The school cost per pupil is included, but not composition effects on results, teacher density or long-run effects on human capital across the whole cohort.

  • Shadow economy and undeclared work

    Unregistered income and undeclared work are by definition absent from register data. The effect can run in both directions, but the net effect is not quantified.

  • Growth, innovation and entrepreneurship

    Positive spillovers to GDP growth, business formation and innovation are not included — nor are any negative effects on productivity.

  • Defence, security and preparedness

    Some public expenditure (defence, general administration) is treated as a collective good and is not allocated per individual.

Sources for individual items are given in the methodology section. sverigefakta.com makes no estimates of the size of these items — we only flag that they are not part of the +SEK 6bn figure for 2022.

Dataluckor — vad webbplatsen inte mäter

Sex konkreta poster som varken Ruist 2018 eller KI 2025 kvantifierar. Färgmarkeringen visar i vilken riktning utelämnandet sannolikt påverkar huvudsiffran.

Risk: underskattar kostnaden

Brottskostnad ingår inte

Polis, rättsväsende, kriminalvård och brottslighetens samhällsekonomiska kostnader fördelas inte per grupp i Ruist eller KI. Givet Brå 2021:9:s överrisksiffror skulle inkluderingen sannolikt sänka flyktinggruppens nettobidrag.

Risk: underskattar kostnaden

Kommunala merkostnader saknar nationell aggregering

Kommunernas faktiska kostnader för försörjningsstöd, socialtjänst, SFI och tolkning utöver statlig ersättning aggregeras inte nationellt. Statskontoret 2021:6 visar att den statliga ersättningen mer än halverats 2018–2020.

Riktning okänd

Trångboddhet, skola, vårdköer — ej individkopplade

Konsekvenser som ökad trångboddhet, sjunkande skolresultat i vissa kommuner och längre vårdköer dokumenteras av Boverket, Skolverket och regionerna — men kopplas inte individuellt till migrationsbakgrund i nettokalkylerna.

Riktning okänd

Undanträngning på arbets- och bostadsmarknad

Eventuella effekter på inrikes föddas löner, sysselsättning och bostadstillgång ingår inte. Forskningen är delad — IFAU och internationella studier hittar små men icke-noll-effekter, främst för låginkomstgrupper.

Risk: underskattar kostnaden

Andra generationen klassas som inrikes födda

Barn till två utrikes födda räknas som inrikes födda i KI:s data. Deras nettobidrag göms därför i 'inrikes födda'-siffran. Brå 2021:9 visar 3,2x överrisk för brottsmisstanke i denna grupp — högre än bland utrikes födda själva.

Risk: överskattar kostnaden

Dynamiska BNP-effekter ej kvantifierade

Möjliga positiva effekter på företagande, innovation och konsumtion utöver direkt skatt fångas inte. Forskningen är blandad — vissa OECD-studier ser positiva spridningseffekter, andra inga.

Independent verification

None of the reviews below is an exact replication, but they confirm the order of magnitude and the mechanism in the underlying studies.

Version history of the method

  • 2026-09-14 — methodology portal and model register: all active models and in-house calculations were given model cards with sources, assumptions, uncertainty and reproducibility status.
  • 2026-09-14 — the what-if simulator: the crime-cost item and the combined index for the second generation were removed. No volume parameter is any longer scaled to crime or behavioural outcomes.
  • 2026-09-14 — canonical definition engine: Sweden, Denmark and Germany verified against the primary source; conversion factors between definitions removed; the map always shows Statistics Sweden's definition.
  • 2026-09-13 — NIER 117: the published NIER calculation is the default; the alternative calculation is shown only on request and is labelled as not an official NIER figure.

Source code, data and datasets

Every model card states the code and data reference in the product. The entire indicator dataset can be downloaded as JSON or CSV. More downloads are available in the dataset catalogue.