# Getting It Right: anti-Muslim crime narratives, evidence and publication boundaries

Research cut-off: 6 October 2026. Scope: public factual education, not audience targeting or political persuasion. This is a scoped evidence review, not a systematic review of every European jurisdiction or every study. England and Wales, Sweden, Germany and the United States are the principal jurisdictions; Denmark supplies a causal-design example. No person's religion is inferred from a name, appearance, nationality or ancestry.

## Editorial conclusion

A defensible lead is: **Crime data should be reported honestly, with the group, offence, place and time actually measured. Nationality, birthplace and ethnicity are not interchangeable with religion. Some recorded disparities shrink after demographic and socioeconomic adjustment, and some remain. Neither a raw disparity nor an unexplained residual establishes that Islam causes violence.**

The proposed alternative, “it is disadvantaged young men rather than religion,” is too categorical. Age, sex, material conditions, peer environments, selection into migration, institutional practices and other factors merit investigation; their explanatory contribution varies by outcome and setting. Evidence can support a particular contribution without establishing a complete explanation. Religious affiliation, religious practice, doctrinal belief and ideological motivation are also different exposures.

Avoid a single “Muslims and crime” scoreboard. Publish narrow questions with visible limits. Reject collective blame without denying victims, known offending, or institutional failures. An individual's innocence does not depend on their group's average, and an offender's guilt does not transfer to people who share a religion or origin.

## 1. Narrative map and evidence of prominence

Prominence grades here mean documented public circulation, not agreement by most people. Search results, one viral post and many articles repeating one source are not prevalence estimates. No representative cross-national estimate of belief in these exact narratives was established in this review.

### A. False attribution after a real violent event

Example: Southport, July 2024. A 2025 parliamentary report documents rapid circulation of the invented name and Muslim-asylum-seeker account after the killings. This is a strong primary record of a consequential narrative, not merely a hypothetical trope. [UK parliamentary report](https://publications.parliament.uk/pa/cm5901/cmselect/cmsctech/441/report.html)

ISD's original monitoring reported millions of views for false claims. Treat views as platform exposure estimates, not unique viewers, belief, or a clean estimate of the misinformation's causal effect on rioting. Its specialist monitoring has a relevant counter-extremism mission; parliamentary corroboration adds institutional evidence but may partly draw on the same underlying monitoring. [ISD investigation](https://www.isdglobal.org/digital-dispatch/from-rumours-to-riots-how-online-misinformation-fuelled-violence-in-the-aftermath-of-the-southport-attack/)

Publication angle: verify the specific claim, timestamp corrections, and distinguish an incorrect identity assertion from later verified case developments. Do not classify an entire crime as fabricated because an online attribution was false.

### B. Selected violent videos presented as evidence of an entire faith

The US president's amplification of Britain First videos in November 2017 generated an urgent parliamentary question. This establishes unusually high-level transatlantic amplification; it does not establish how many people believed the captions. [Hansard, 30 November 2017](https://hansard.parliament.uk/commons/2017-11-30/debates/9ED5779E-DAA6-4DF6-A558-8C7E533A469E/OnlineHateSpeech)

PolitiFact documented a false migrant attribution in one video and missing context across the set. Use that contemporaneous fact-check for the caption issue rather than claiming every video was fake. [PolitiFact](https://api.politifact.com/article/2017/nov/29/donald-trump-retweets-anti-muslim-videos-far-right/)

Prominence: high historical visibility; no current prevalence estimate. A verified incident remains a selected observation with no denominator.

### C. Real group-based child sexual exploitation generalized to all Muslim men

The UK debate combines real abuse, failures of protection, uneven local patterns, incomplete national data and unjustified religious extrapolation. Treat those propositions separately. National attention is well evidenced by the government's commissioned audit; the narrower claim's population prevalence is unmeasured here. See the Casey evidence entry below.

### D. “Sweden's rape statistics prove Muslim immigration causes rape”

Two distinct issues: comparability of countries' recorded rape counts, and within-Sweden associations by migration background. Neither resolves the other. This review establishes the statistical issue and relevant studies, but does not measure this exact narrative's current reach. Do not inflate prominence merely because the claim is familiar.

### E. “Foreign suspect shares are Muslim offending rates”

Common error structure to teach, using Germany as a worked example: changing citizenship into religion, shares into rates, suspects into convictions, and correlation into causation. The current official figures demonstrate that overrepresentation is a real statistical question; its transformation into a religious conclusion is a separate inference. No platform-wide frequency estimate was collected.

### F. “Sex assaults are either all invented or all evidence of a religious threat”

Cologne's 2015/16 New Year sexual and property offences were serious real events investigated by the North Rhine-Westphalia parliament. Their occurrence must not be denied to correct generalizations. The official committee record is an event/institutional source, not a population-rate study. [Landtag inquiry](https://www.landtag.nrw.de/home/der-landtag/ausschusse-und-gremien-1/untersuchungsausschusse/fruhere-untersuchungsausschus-1/16wp-pua-iv-abgeschlossen.html)

A separate Frankfurt Freßgass story about the following New Year was challenged by police investigation. Keep these events separate. This review located the police release and corroborating public-broadcaster coverage, but a publication-ready incident card should inspect the release in full and record its precise finding rather than paraphrase a headline. [Police release](https://www.presseportal.de/blaulicht/pm/4970/3560139); [Deutschlandfunk](https://www.deutschlandfunk.de/silvesternacht-in-frankfurt-polizei-dementiert-bild-bericht-100.html)

### G. Prison religious affiliation used as a rape or general-offending rate

This deserves a methodological card, not automatic dismissal of the underlying count. England/Wales prison statistics really do include religion. A prison stock is not all crimes committed by that religion; it reflects entry, detention and sentence duration, and religious identity at recording need not establish identity when an offence occurred. See the prison-source entry below. Prominence was not quantified.

## 2. Evidence register

Each entry separates observed result from this review's inference. Source descriptions are intentionally concise; follow the linked originals for tables and qualifications.

### S1. Sweden: official register analysis, Brå 2021:9

[Report and English summary PDF](https://bra.se/download/18.45e4b8e192705389a34c2b/1729515966490/2021_9_Registered_offending_among_persons.pdf)

Residents aged 15+ at end-2014; suspected offences committed in 2015–18. Relative to Swedish-born people with two Swedish-born parents, all-offence suspect-registration ratios fell from 2.5 to 1.8 for foreign-born people and from just over 3 to 1.7 for Swedish-born people with two foreign-born parents after standardization for age, sex, income, education and municipality type. For rape suspicion, the respective ratios fell from 3.2 to 2.2 and from 3.2 to 1.4 (PDF pp. 5–7).

**Interpretation:** useful evidence against both “adjustment changes nothing” and “adjustment eliminates every difference.” These are migration-background groups, not religious groups; the outcome is registration as a suspect. Reported adjusted ratios are model-dependent comparisons, not percentages of individuals committing rape. The descriptive study cannot assign the remaining gap to religion, culture, discrimination or any other unmeasured cause. Historical period, not a 2026 estimate.

### S2. Sweden: rape-conviction case-control study

[Khoshnood, Sundquist & Sundquist, DOI 10.1177/08862605241311611](https://journals.sagepub.com/doi/10.1177/08862605241311611), online 2025; Journal of Interpersonal Violence 41(1–2), 2026, pp. 372–390.

4,032 individuals aged 15–60 convicted of rape/attempted or aggravated rape against adult women during 2000–20; 20,160 age/sex-matched controls. Table 2: for foreign-born arrivals aged 15+, odds ratios were 6.25 initially, 3.14 after socioeconomic controls, then 6.22 after psychiatric/substance-use and prior-crime controls. The Swedish-born/one-foreign-parent estimate was 1.12 (95% CI 0.95–1.32) in the final model: not statistically significant, despite broader wording in the abstract. Other groups remained elevated.

The authors explicitly could not measure religion or culture, establish specific causes, or observe equivalent pre-migration histories. Shorter observation for recent arrivals and residual confounding are acknowledged. **Interpretation:** do not describe every adjustment as reducing every coefficient; do not treat model D as automatically the best causal estimate. The discussion's speculation about integration is not a measured religious mechanism. Convictions and adult-female-victim cases cannot represent all offending.

### S3. Sweden: international rape-statistics comparability

[Brå, Reported and cleared rapes in Europe, 2020:13](https://bra.se/english/publications/archive/2020-09-30-reported-and-cleared-rapes-in-europe)

The official methodological study finds differences in legal definitions, recording and clearing rules make raw country comparisons misleading; harmonizing selected conventions substantially changes the comparison. **Interpretation:** high recorded figures do not directly establish equally large differences in incidence. This does not prove no underlying difference, no real increase, or that sexual violence is unimportant. An improvement in the comparability argument is not a rebuttal to S1/S2, which ask different questions.

### S4. England/Wales: Casey audit, June 2025

[National audit, accessible report](https://www.gov.uk/government/publications/national-audit-on-group-based-child-sexual-exploitation-and-abuse/national-audit-on-group-based-child-sexual-exploitation-and-abuse-accessible), especially sections 4.3–4.5.

National ethnicity information was too incomplete for reliable national conclusions. Better local material from Greater Manchester, West Yorkshire and South Yorkshire showed Asian-ethnicity overrepresentation in selected group-based exploitation contexts. The audit challenged the 2020 Home Office report's national-majority inference. **Interpretation:** neither “all grooming gangs are Muslim” nor an unqualified “the national data prove most are White” is warranted. Asian/Pakistani ethnicity is not measured Muslim faith. Local operation-based observations are not nationally representative religion-adjusted rates. Acknowledging serious abuse and institutional avoidance is compatible with rejecting collective blame.

### S5. England/Wales: ONS religion measurement

[ONS, Rape by religion, 2025](https://www.ons.gov.uk/aboutus/transparencyandgovernance/freedomofinformationfoi/rapebyreligion2025)

ONS says it does not hold perpetrators' religion data and directs enquiries toward Ministry of Justice offender statistics. **Interpretation:** say precisely what ONS lacks; do not say no UK criminal-justice dataset ever records religion. Victim religion, hate-crime motivation and offender religion answer different questions.

### S6. England/Wales: actual prison religion data

[HMPPS Offender Equalities Report 2023/24](https://www.gov.uk/government/statistics/hmpps-offender-equalities-annual-report-2023-to-2024/hmpps-offender-equalities-report-202324)

The published March 2024 snapshot includes prisoners' self-identified religion; 18.2% identified as Muslim. This is a historical prison-population statistic, not an estimate of rape incidence or all criminal offending. **Interpretation:** avoid converting it to an unadjusted “Muslim criminality rate.” A valid analysis needs age/sex and geographical alignment, custody status, offence and sentence mix, time at risk, recording completeness and identity timing. Do not invoke conversion as a complete numerical explanation without longitudinal data.

[MoJ religion-code change notice, 30 October 2025](https://assets.publishing.service.gov.uk/media/6903487ecc6b3272b3913123/Changes_to_Religion_Codes.pdf): classifications changed at end-June 2025, including Muslim category content; the notice warns pre/post distributions are not comparable. This is an immediate correction trigger for any prison-religion trend chart.

### S7. Germany: official 2025 statistics, released April 2026

[Federal government summary of PKS 2025](https://www.bundesregierung.de/breg-de/aktuelles/polizeiliche-kriminalstatistik-2025-2422020)

Reports non-German citizens were 42.9% of violent-crime suspects. For rape, it reports 98.6% male suspects and a 38.5% non-German share. These figures must be labelled suspects and citizenship, not convicted offenders and religion. **Interpretation:** a share of suspects is not a population-adjusted risk. German citizens may have migration backgrounds; non-German citizens belong to diverse religions or none. Do not infer faith from citizenship categories. Use the detailed BKA offence codes before charting “rape,” as press-summary labels may compress categories.

[BKA 2024 nationality/residence-purpose tables](https://www.bka.de/DE/AktuelleInformationen/StatistikenLagebilder/PolizeilicheKriminalstatistik/PKS2024/PKSTabellen/BundTVNationalitaet/bundTVNationalitaet.html?nn=240862) and [resident suspect-rate table description](https://www.bka.de/SharedDocs/Downloads/DE/Publikationen/PolizeilicheKriminalstatistik/2024/Interpretation/Belastungszahlen/ZR-BZ-01-T50-TVBZgesamtnichtdeutsch2009_xls.html) distinguish the relevant variables. Avoid the outdated blanket claim that Germany never provides residence-adjusted rates. Match the numerator to the actual resident denominator; exclude residence-law offences when comparing general offending opportunity.

### S8. Germany: locality-level analysis

[Adema & Alipour, ifo Schnelldienst digital 3/2025](https://www.ifo.de/DocDL/sd-2025-digital-03-adema-alipour-migration-kriminalitaet.pdf)

Analysis of districts in 2018–23 finds no positive local-crime relationship with changes in foreigner/refugee shares under its specifications. Its geography is the place of the offence; district data cannot fully align suspects' residences, and individual microdata are unavailable. **Interpretation:** informative about local aggregate changes, not a direct individual religion effect. It can coexist with suspect overrepresentation. Do not rewrite a district-level null as proof every immigrant subgroup has identical individual risk, or that all settings and offences show no effect. ifo's press release is the same research source, not independent replication.

### S9. Germany: direct religiosity measurement, with limited scope

[Dirk Baier, The Influence of Religiosity on Violent Behavior of Adolescents, 2014](https://journals.sagepub.com/doi/10.1177/0886260513504646)

Publisher abstract describes a German-wide school survey of 16,545 ninth-grade boys identifying with Christian or Islamic denominations. Among Muslim respondents, bivariate religiosity/violence association was nonsignificant; controlling alcohol consumption produced a positive association. The abstract also describes masculinity norms and violent-media use.

**Interpretation:** this directly warns against claiming religion has never been measured. But a selected-age observational survey and a conditional association cannot establish the causal effect of Islam on adult men's rape offending. Alcohol may sit on a causal pathway; adjusting it changes the question. Full article was restricted; this review assessed the abstract only. Do not use it for a detailed effect-size card without full-methods review and replication evidence.

### S10. United States: what FBI data encode

[FBI 2025.0 NIBRS technical specification](https://le.fbi.gov/file-repository/2025-0-nibrs-technical-specification-062625.pdf), offender segment, data elements 36–39A; [current documentation hub](https://le.fbi.gov/informational-tools/ucr/ucr-technical-specifications-user-manuals-and-data-tools).

The offender segment records age, sex, race and ethnicity, not religious affiliation. Religious-bias coding describes offence motivation/targeting; it is not the offender's faith. **Interpretation:** a purported national FBI “Muslim rape rate” needs another documented data source and valid denominator. Do not substitute Arab ancestry, a Middle Eastern country, or a Muslim-sounding name.

[BJS national-estimates methodology page](https://bjs.ojp.gov/national-incident-based-reporting-system-nibrs-national-estimates-api) explains imputation and weighting for nonresponse/missingness. Distinguish raw participating-agency counts from national estimates, especially across reporting transitions.

### S11. United States: migration evidence, not Muslim evidence

[Light, He & Robey, PNAS 2020, Texas 2012–18](https://pubmed.ncbi.nlm.nih.gov/33288713/), DOI 10.1073/pnas.2014704117.

Texas administrative data showed lower felony arrest rates among undocumented immigrants than US-born citizens, including violent offences; results were robust to alternatives including conviction measures. **Interpretation:** relevant to a broad “immigrants inherently criminal” assertion, but cannot estimate US Muslim offending or explain European findings. Immigration status, legal status and religious identity are different. Texas and the historical period constrain generalization; undocumented-population estimates add denominator uncertainty. Abstract/figure descriptions checked here, not a new replication.

[Ousey & Kubrin, Annual Review of Criminology 2018](https://www.annualreviews.org/content/journals/10.1146/annurev-criminol-032317-092026)

Review/meta-analysis of macrosocial research from 1994–2014 reports a very weak negative overall immigration–crime association, with meaningful variation by study design and context. **Interpretation:** an aggregate literature summary is not a subgroup religious comparison or proof immigration reduces every offence everywhere. Abstract and source bibliography reviewed; full dynamic article body not retrieved.

### S12. Denmark: stronger design for a contextual mechanism

[Damm & Dustmann, American Economic Review 2014](https://pubs.aeaweb.org/doi/abs/10.1257/aer.104.6.1806), DOI 10.1257/aer.104.6.1806.

Uses quasi-random neighbourhood placement of refugee families in Denmark (1986–98). Exposure to neighbourhood young people's convictions predicted later male assignees' convictions; not every neighbourhood-crime measure did. **Interpretation:** more credible evidence for a particular peer/environment mechanism than a simple origin correlation. It does not prove socioeconomic conditions explain all disparities, nor isolate Islam, nor directly estimate rape-specific effects. Publisher abstract reviewed.

## 3. Interpretation rules for every public card

1. **Define the estimand.** Is the question the share of offenders, the risk per person, the effect of migration on local crime, or a causal religious effect? Those are different quantities.
2. **Preserve the unit.** One incident can have several offences, victims and suspects; one person can appear in several incidents. Counts of crimes are not counts of people.
3. **Use the correct justice stage.** Reports, suspicions, arrests, charges, convictions and imprisonment are not interchangeable. A conviction is stronger evidence of individual legal guilt, but a conviction dataset remains a selected view of all offending.
4. **Keep denominators compatible.** Same geography, age/sex eligibility, observation window, residence status and population definition. Visitors in the numerator with residents alone in the denominator distort rates. A small denominator makes rates unstable.
5. **Show absolute alongside relative numbers when available.** A ratio does not say the majority of a group offend. Do not invent an absolute rate from a case-control sample with an artificially selected control count.
6. **State adjustment choices.** “Controlled for socioeconomic status” may mean a few measured variables. It never automatically means every disadvantage has been removed. Income before conviction can follow the offence or earlier justice-system involvement.
7. **Avoid causal leftovers.** Unexplained residuals are not labels for religion. Missing family histories, trauma, schooling, local opportunities, policing, reporting and other factors remain possible; possibilities are not established explanations.
8. **Distinguish confounders and mediators.** A variable affected by the exposure can mediate rather than confound its effect. Controlling downstream variables can remove part of an effect or create bias. Define the causal question before selecting a model.
9. **Treat “no significance” correctly.** It does not prove equality; inspect precision, sample size and relevant effect bounds. Conversely statistical significance does not establish causation or public importance.
10. **Account for the observation process.** Reporting, recognition, identification, police attention, charging, legal rules, evidence and adjudication influence observed records. Victimization surveys avoid some police selection but introduce recall, nonresponse, interpretation and coverage limits. A respondent may not know an offender's religion.
11. **Avoid ecological reversals.** A district-level association does not identify individual behaviour; individual overrepresentation does not imply immigration raised district crime. Both can be true in the same setting.
12. **Do not replace religion with a proxy without saying so.** “Muslim-majority country,” “Asian,” “Pakistani,” “foreign,” “refugee,” “non-Western” and “Muslim” are not synonyms. Assigning religion from surnames is not a valid public evidence method.

These are analytical safeguards, not a claim that every bias always works in one direction. The correct response to a limitation is calibrated uncertainty, not automatic dismissal of inconvenient evidence.

## 4. Proposed public-facing statements

Use adjacent source links from the register. These are drafts, not permission to publish without normal review.

**Core card:** “Nationality, birthplace and religion are different things. A statistic about foreign-born suspects cannot, by itself, tell us the crime rate of Muslims or show that religion caused an offence.”

**Adjustment card:** “Fair comparisons matter. Age and social conditions can change the size of recorded differences. They do not always remove them, and a remaining difference does not reveal its cause.”

**Evidence-stage card:** “Police suspects, convicted offenders and people in prison are different populations. Check which one a statistic actually counts.”

**UK exploitation card:** “The abuse was real and victims were failed. The evidence also has limits: local findings cannot establish a national rate for an entire religion.”

**Sweden comparison card:** “Countries record rape differently. Raw police totals are not a reliable league table of how often rape happens.”

**Individual responsibility card:** “Investigate offences and protect victims. Guilt belongs to the people responsible; it does not transfer to everyone who shares their religion or background.”

Do not publish: “Muslims are no more likely to offend everywhere”; “poverty completely explains it”; “the figures are all fake”; “no one records religion”; “most national grooming offenders are definitely White”; “nationality proves Islam is the cause”; or “a study found a coefficient, so religion is causal.”

## 5. Evidence strength and publication priorities

**Ready for a carefully scoped explainer:** measurement distinctions; S1 official adjustment example; S3 comparability; S4 updated national/local distinction; S5/S6 religion-data distinction; S10 FBI schema.

**Use with full-method caveats:** S2 observational conviction study and its model-specific findings; S8 local aggregate analysis. These should appear together with clear differences in outcomes and units, not as contradictory scorecards.

**Supporting context, not the lead proof:** S9 older religiosity survey (restricted full text); S11 US immigration studies; S12 Danish peer-exposure evidence. They constrain broad claims but do not settle the Muslim-men/rape question.

**Hold for further incident-level verification:** a Frankfurt fact-check card, additional US incidents involving minors, national “no-go zone” narratives, and any numerical ranking of narrative prevalence. A search hit is not enough. Do not publish identifying details of child victims or nonpublic minors.

Before expanding to France, Belgium, the Netherlands, Norway or all of Europe, repeat the country-specific measurement review. No Europe-wide Muslim crime rate was established here. A universal causal answer is outside what these sources can provide.

## 6. Corrections and refresh triggers

- New nationally complete and methodologically comparable UK CSE data; subsequent inquiry findings that alter Casey-era conclusions.
- New Swedish register work using later periods, different offence definitions, improved covariates or replication of the rape-conviction models.
- Changes to German citizenship law, residence classifications, statistical denominators or offence codes; next PKS release. Preserve historical labels rather than silently replacing figures.
- NIBRS specification changes, coverage revisions, national estimation methods or corrected source series.
- Prison religion coding breaks; changing custody populations; new longitudinal identity data. The June 2025 break already matters.
- A study retraction, correction or credible methodological response; confidence intervals showing claims were overstated.
- Verification that an illustrative post was misquoted, altered, synthetic or misattributed; new evidence about the event. Keep a visible dated correction log.
- Evidence of actual narrative reach should update the prominence label independently of the underlying truth assessment.

## 7. Methods, access and independence notes

Search method: targeted web searches with opposing-result checks, then primary-source opening, report-section searches and publisher methods/table inspection. Search families included Brå foreign/native background and adjustment; Swedish rape convictions and immigrant background; Casey 2025 ethnicity audit; ONS rape by religion; BKA nationality/residence and latest PKS; FBI NIBRS religion/schema; Texas immigration arrest studies; religiosity/violence Germany; Denmark neighbourhood natural experiments; Southport misinformation; Britain First retweets; Cologne inquiry and Frankfurt police correction.

Access date for all links: 6 October 2026. Evidence date and publication date are distinguished above. Search engine relative-date metadata was not relied on when the source supplied its own date. The Swedish conviction study's 2025 online/2026 issue distinction is intentional.

Direct review: Brå English PDF text, relevant Casey accessible-report sections, ONS response, HMPPS report, MoJ code notice, German current government release, ifo PDF methods passages, FBI 2025 technical specification offender segment, parliamentary Southport and Hansard records, publisher full text/tables of the Swedish conviction study. Brå page screenshot was requested; the browser returned a reference without a usable image display, so the numeric summary rests on clearly parsed report prose, not a claimed visual table inspection.

Limited review: Baier's publisher abstract only (full text restricted); Texas PubMed abstract and figures; Annual Reviews abstract and bibliography (dynamic full text unavailable); Denmark AER abstract. PMC returned a CAPTCHA screen for Swedish papers; no CAPTCHA was solved or bypassed. The same Swedish conviction paper was legitimately available in full at its publisher. A BKA 2024 PDF open failed; no claim here depends on pretending it was read. Statistics Denmark's 2025 immigrants-report page also failed to open; it was not used for numerical claims.

Independence: several webpages are duplicate presentations of one study. Do not count a publisher paper, PubMed record, university page and news summary as four replications. Brå's administrative work and the Swedish academic study have different designs but partly draw on the same country's justice-system machinery and periods. Casey synthesizes local records and earlier inquiries; downstream news summaries are not new datasets. ifo's press release and article are one analysis. Official crime records are primary for recorded counts, not automatically definitive for unobserved offending or causes. Advocacy/specialist monitoring can establish observed posts and reach, with its sampling and mission disclosed.

No raw person-level data were collected; no religious profiles were inferred. No external messages, uploads, or publication actions were taken. This review does not offer individual criminal-risk assessment.