Fifty-Six Percent

In February 1988 the Novosti press agency ran a round table at the Viru Hotel in Tallinn. Seventy-eight people came, thirty-three of them foreign journalists from a dozen countries, and they had questions about what was happening in the Baltics. The historian Toomas Karjahärm, who was there, recorded what the answers sounded like: pessimistic. Most of the session circled around migration and the critical decline in the share of the indigenous population. The economist Arvo Kuddo supplied the specific number. On current projections, he said, by the year 2000 Estonians would be 56 to 57 per cent of the population of Estonia.

I want to be precise about where I am standing when I tell you this. I am a Russian-speaking Estonian. The inflow Kuddo was describing is how my family got here. In that room, in that arithmetic, I was not the audience. I was the variable.

I have a particular interest in what happened next, which is that the number was wrong by about eleven percentage points. At the 2000 census ethnic Estonians were 68.3 per cent of the population whose ethnicity was known, or 67.9 per cent of the total population depending on which denominator you use. The line went the other way, and it has stayed there for twenty-five years.

This essay is about why that projection failed, and about the much larger version of the same projection that is currently running about Muslims in Europe. The two run on the same arithmetic and fail at the same joint.

What the number actually did

The Estonian share had collapsed. Across the Soviet censuses it ran 74.6 per cent in 1959, 68.2 in 1970, 64.7 in 1979, and 61.5 in 1989. That last figure is the lowest ever recorded.

Nobody in 1988 was inventing a decline. They were extrapolating one, which is a different act, and it is the one that failed.

Here is what actually reversed it, from Estonia's own statistics office. In the years after 1991 an estimated 80,000 people left, most of them east: Soviet military families, workers from the big plants. Between 1989 and 2000 the number of ethnic Estonians in Estonia fell by about thirty thousand. It fell by another thirty thousand over the following decade. And across those same two decades the Estonian share rose from 61.5 per cent to 68.3 and then to 69.8, which is Statistics Estonia's census series, all three measured against the population whose ethnicity is known.

That is the whole mechanism. No Estonian out-reproduced anyone: they became relatively more numerous by becoming absolutely fewer, while the denominator shrank faster than they did. Ene-Margit Tiit, Estonia's most senior population statistician, puts it plainly in a 2023 paper: the share moves with the direction and volume of foreign migration, and in recent decades it has fluctuated between 68 and 70 per cent.

There is a second thing the 1988 projection could not have priced, and it is the one that matters for everything that follows. The differential ran out.

In the interwar republic the fear had a real basis. Katus, Puur and Sakkeus, working on the 1920s and 30s data, found that Estonia's Russian minority had the highest birth rate of any group, exceeding the average persistently, and was the only minority with constantly positive natural growth. That was true. It stopped being true. Today the relationship is inverted: ethnic Estonian women who have finished childbearing average about two children, while women of other ethnicities, mainly Russians, average 1.5 to 1.7. A 2025 census-linked study of women born between 1971 and 1977 puts the adjusted gap at 0.19 children below the ethnic Estonian level, against a national mean of 1.9. The group that was going to out-breed the Estonians now has fewer children than they do.

I should not oversell any of this. The Estonian share has drifted down since the last census, from 69.4 per cent in 2021 to 68.5 per cent on 1 January 2026, because Estonia has taken in a large number of Ukrainians. The shape is not a triumphant recovery. It collapsed, rebounded about eight points on emigration, and has been roughly flat for a generation while the country stopped talking about it.

Here is the objection that nearly stopped this essay.

Estonia got a deus ex machina. The population that made Kuddo's projection frightening was attached to an empire, and when the empire dissolved, eighty thousand people had somewhere to go, and the military families had orders. Europe's Muslim population has no metropole to be recalled to and no garrison to withdraw. Worse for me, the fertility inversion I have just described is the wrong exhibit: it is measured on cohorts born in the 1970s whose completed families we can only see now, so it cannot have been what falsified a number that came due in the year 2000. Emigration falsified Kuddo. Emigration is the one mechanism Europe does not have on offer.

The claim I can actually defend is narrower, and narrow is enough.

Estonia is not evidence that Europe's projections will reverse. It is evidence about where projections of this kind break, and they break at a specific joint: the migration term. Composition trends are set overwhelmingly by migration flows, migration flows are the least forecastable quantity in demography, and Pew says so itself in its own methodology. Kuddo's arithmetic was fine. His confidence was the error, and his confidence lived entirely in an assumption about people continuing to arrive. That much transfers. The happy ending does not.

Everybody guesses high

About 0.5 per cent of Estonian residents aged fifteen and over who answered the 2021 census's voluntary religion question call themselves Muslim, up from 0.1 per cent in 2011. That question went out through the census sample survey rather than the register, so it is an estimate with a confidence interval rather than a headcount. The same census found that 58 per cent of people that age belong to no religion at all. Estonia is a country where the largest religious movement of the past thirty years has been away from religion.

In 2015, under the EU relocation scheme, Estonia was allocated 329 places. It had taken 77 people by the end of 2016, and it rejected 171 of the 321 requests Greece sent it, a refusal rate of 53 per cent that put it near the top of the EU table. When the programme closed, Estonia had received 206 people in total, 141 from Greece, 59 from Turkey and six from Italy, and declared its obligation fulfilled. By March 2018, 88 of those 206 had already left the country. This was, at the time, a national political emergency.

Two years before that, in May 2013, Martin Helme had already given the argument its permanent Estonian formulation on a Tallinn television programme. Immigration policy should follow one simple rule, he said: kui on must, näita ust, if he's black, show him the door. He added that he wanted Estonia to be a white country, and that the most problematic group was Muslims. Asked about the wording by Postimees, he said he categorically refused to let political correctness dictate how he expressed himself. He was then a board member of EKRE. He is now its chairman.

Then Russia invaded Ukraine, and Estonia took in more refugees per head than almost anyone. By the end of August 2025 it was hosting 25.4 beneficiaries of temporary protection per thousand inhabitants against an EU average of 9.7, and about 30,500 people were still under temporary protection in July 2026, having peaked near 40,000. That is roughly one resident in forty-five. A UNHCR survey in 2025 found 79 per cent of working-age refugees in employment, 99 per cent of school-age children in Estonian schools, and refugees reporting hostile treatment down from 39 to 30 per cent.

I am not going to pretend this is a clean comparison. Estonians did not treat these as the same category of person, and the reason is obvious: the 2015 arrivals were Syrians and Iraqis, the 2022 arrivals were European Christians fleeing the same country Estonians fear. The obvious reading is that Estonia found Christians easy and Muslims hard, which is his point rather than mine, and I am not going to pretend the sentence does not sit there. What the outcome numbers actually establish is narrower and more useful: absorption capacity is not a fixed property of a country. Estonia was told for a decade that it could not take people, and then took two and a half per cent of its population inside three years, and the schools and the labour market held. That is a fact about institutions and money and political will. It says nothing about which arrivals are easy, and everything about the fact that "we cannot" was a choice being described as a limit.

That inverse relationship between the size of the number and the size of the fear is not an Estonian quirk. Ipsos has been measuring it for a decade. In 2016 the French, asked how many out of every hundred people in their country were Muslim, answered 31 against an estimated 7.5. Germans said 21 against 5. Poles said 7 against essentially zero, and Hungarians 6 against essentially zero. Repeated in 2024, the French and the Belgians both guessed 28 against about 8, Germans 24 against 7, Britons 21 against 6, and Poles and Hungarians 6 against under 1.

Two complications, since the pattern is not as tidy as it looks: in that 2016 round the most accurate country in Europe was the Czech Republic, which guessed 3 against under 0.1, and these are mean guesses, which a handful of people answering "ninety" can drag a long way from the typical answer.

The third complication is more serious. Those numbers look like a measure of prejudice, and substantially are not. A 2025 paper in PNAS argues that people systematically rescale estimates of proportions toward a central expectation whenever they are uncertain, so they overestimate every small group and underestimate every large one, and that they do the same thing when you ask them what share of Americans own a dishwasher. The overestimate of Muslims is not, on this account, mainly about Muslims. It is what innumerate people do with any percentage.

That finding is contested, and it does not make the wrong number harmless: people still walk around with it. It does dismantle the version of this essay I might have preferred to write, in which the public is bigoted about arithmetic.

Worse for the debunking impulse: correcting the number does not do what you would hope. Hopkins, Sides and Citrin ran the experiment seven times over eleven years and found that giving people the accurate figure reliably fixed their estimate and barely moved their attitudes. Alesina, Miano and Stantcheva, surveying 22,506 people across six countries, found that the perceived share of immigrants is not what predicts hostility at all; the belief that immigrants free-ride on the welfare state is. In the same study, merely making people think about immigration before asking an unrelated question changed their answers, with no information supplied.

I am not writing this in the belief that a corrected figure changes a mind. The perception gap is worth knowing about for a narrower reason: it establishes that the anxiety is not being produced by the number, because the number is not what people have. Which means an argument that only fixes the number is not going to be enough.

The belief also goes well past a bad estimate. Asked by Harris Interactive in 2021 whether the "great replacement" would happen in France, in the pollster's wording of European, white and Christian populations threatened with extinction by Muslim immigration from the Maghreb and sub-Saharan Africa, 61 per cent of French respondents said it probably or certainly would. This is not a fringe position being smuggled in from the internet. It is most of a country.

Three futures, and only one of them gets quoted

The number everyone is arguing about comes from one report: Pew's Europe's Growing Muslim Population, published in November 2017. Most people quoting it have not read it.

Its baseline is 25.8 million Muslims as of mid-2016, 4.9 per cent of the population, across the twenty-eight countries then in the EU plus Norway and Switzerland. It then gives three projections to 2050, not one. Under zero migration, meaning every border sealed immediately and permanently, the share reaches 7.4 per cent. Under medium migration, 11.2 per cent. Under high migration, which assumes the record 2014 to 2016 refugee flows continuing at the same volume and composition every year for thirty-four years, 14 per cent.

The figure that travels is always the last one. When Russian-language Telegram tells you Sweden will be nearly a third Muslim, the 30.6 per cent it is quoting is Sweden's high-migration scenario. The medium scenario for Sweden is 20.5 and the zero-migration scenario 11.1. The scenario name is the difference between a report and a prophecy.

Two more things in that report matter.

The first is the zero-migration scenario itself, which is the most interesting number Pew produced and the one nobody uses. Seal every border tomorrow and the Muslim share still rises, from 4.9 to 7.4 per cent. Two things are doing that work. The fertility gap does not disappear when the borders close. But the larger and less discussed part is age structure. Muslims in Europe had a median age of 30.4 against 43.8 for everyone else, and 27 per cent of them were under fifteen against 15 per cent of non-Muslims. A younger population grows faster at identical fertility, because more of its women are in childbearing years and fewer of its members are near the end of life. Demographers call this momentum, and it is real, and it is also bounded: the age gap is largely an artefact of migration, since migrants arrive young, and it compresses as a population ages in place.

The second is that Pew's models assume convergence. All of them. Every scenario has Muslim and non-Muslim fertility sliding toward identical rates by 2110. So anyone citing Pew as proof of runaway growth is citing a model that has already conceded the demographic point, and anyone citing Pew as proof of convergence should know that the convergence is partly an input rather than a finding. Pew's own methodology note is blunt about the weakest link: estimating migration is difficult, and projecting it is harder.

Before I use any of that, one concession the report also contains. Of the roughly 1.6 million people granted refugee status in Europe between mid-2010 and mid-2016, Pew put 78 per cent at Muslim. The 53 per cent usually reached for covers all migrants of every kind. The 2015 wave really was overwhelmingly Muslim. The argument I am making does not require it to have been anything else.

A decade of that projection has now run. The refugee peak did not repeat. The largest displacement into Europe since then has been Ukrainian, which is to say overwhelmingly Christian, and after the fall of Assad in December 2024 Syrian asylum applications collapsed. First-time asylum applications across the EU came in at 669,365 in 2025, down 27 per cent on 2024 and a little over half the 2015 peak, and for the first time since 2013 the largest single group of applicants was Venezuelan.

The British figures are the most striking, because Britain is where the ratchet was supposed to be clearest. Net migration to the UK peaked at 944,000 in the year to March 2023. In the year to December 2025 it was 171,000, a fall of more than 80 per cent in under three years, driven by visa rules a government changed because it wanted to. This is the part the model cannot contain. Migration is a policy setting, and policy moves faster than any projection to 2050.

The institutional direction has moved the same way. The EU's Pact on Migration and Asylum, a decade in the making, began applying on 12 June 2026, and it is a tightening. That matters, because of what the Eurabia thesis actually claims: European elites bound by something like contract to keep admitting this migration, unable to stop. Ten years of legal hardening ending in a restrictive Pact is a refutation at the level of the mechanism rather than the arithmetic.

There is one more trap here. Pew published a retrospective in June 2025 putting Muslims at 6.0 per cent of Europe in 2020, up 0.7 points on the decade. That looks like a data point on the same graph as the 4.9 per cent. The 2025 report's Europe is forty-two countries including Russia, Belarus, Ukraine, Bosnia, Kosovo and Albania, where Muslim populations are indigenous and centuries old. Putting the two numbers in one sentence manufactures a trend out of a change of map.

The number nobody lives in

Here my own side argues badly.

Every figure above is a national or continental average, and a national average is close to the least informative number available here. Nobody lives in a country-level percentage. The serious version of the worry is about a particular set of streets, and about what a school intake looks like there in fifteen years, and on that the numbers are much less comfortable.

Birmingham, the city Fox News called a no-go zone, was 22 per cent Muslim in 2011. At the 2021 census several of its wards ran between 70 and 80 per cent Muslim, Bordesley Green at 80.2 and Sparkbrook and Balsall Heath East at 77.4, which is a fact about how people actually live, and waving the 22 at someone who is thinking about the 70 is not an answer.

Vienna is the sharpest example, because a team at the Austrian Academy of Sciences modelled it four ways and every scenario points the same direction: Muslims at 20 to 30 per cent of Vienna by 2046. That includes the scenario in which international migration halts entirely and Austria seals its border, which still lands at 14 to 20 per cent. Under Pew's medium scenario Sweden reaches 20.5 per cent nationally, and Swedish Muslims are not evenly sprinkled across Norrland. Add the age structure from Pew's zero-migration scenario, and the school-cohort share in a given district runs well ahead of the population share for a generation.

None of that is Eurabia. A city district is not a continent, and a 25 per cent Vienna is a city with a large minority in it, which is a description that already fits Toronto, Tallinn and half of London without anyone's civilisation ending. But it is the thing people are actually looking at, and an essay that answers a question about Molenbeek with a statistic about Belgium has changed the subject. Concentration is real. It is a segregation problem, with a segregation literature and segregation policy behind it, and it is not evidence for the continental claim, because those are different claims with different mechanisms and different remedies.

Which is the whole point of separating them. If the mechanism you fear is displacement by headcount, the remedy is the border. If the mechanism is concentration and the isolation that comes with it, the border does nothing at all, and thirty years of arguing about the border is thirty years not spent on Malmö.

The gap narrows, unevenly

The claim underneath all of this is that Muslims in Europe have far more children and always will. The first half is true and shrinking. The second half is not a finding at all. It is an assumption, and it is the assumption the whole argument is standing on.

Pew's fertility figures for Europe are 2.6 children per Muslim woman against 1.6 for non-Muslims, for the 2015 to 2020 period on that thirty-country definition. Pew projects the gap to narrow to 0.7 by 2045-2050. Note also that Pew's own earlier estimate, on a fifty-country Europe including Russia, put the Muslim rate at 2.1 for 2010 to 2015, half a child above Christians rather than a full child. Both numbers are Pew's. They are routinely quoted at each other.

The global Muslim fertility rate of 3.1 has nothing to do with any of this. It is driven by Niger, Somalia, Afghanistan and northern Nigeria, and it tells you nothing about a woman in Marseille.

The largest cross-national study of the question, covering twenty-five countries, found Muslim fertility averaging 47 per cent above the national level, and then split it: immigrants 62 per cent above, native-born Muslims 19 per cent above. Nineteen per cent is a real gap and a long way short of a civilisational trajectory, and the size of the drop between those two figures points at migration status as the operative variable.

The measurement itself is a problem, and INSEE has quantified it. The standard fertility rate attributes to immigrant women a lifetime pattern distorted by the catch-up births that cluster just after arrival. Correct for it and France's immigrant figure falls from 2.98 to 2.37. That correction is 0.61 of a child, on the same order as the gaps being argued about, and almost every immigrant fertility number in public circulation is uncorrected.

Look instead at completed families, which cannot be distorted this way. For French women born between 1960 and 1974, INSEE finds immigrant women averaged 2.3 children, daughters of two immigrant parents 2.0, daughters of one immigrant parent 1.8, and women with no migrant background 1.9. In Norway, Norwegian-born women with immigrant parents had a lower fertility rate in 2018, 1.48, than the national average of 1.56. Fertility among Somali-born immigrant women in Norway, the standard example of a group that supposedly never adapts, was 2.42 that year, down 0.41 in twelve months.

Now the other side, which the people making my argument usually skip. Convergence is partial and it is group-specific. Sylvie Dubuc's work on Britain finds second-generation Pakistani and Bangladeshi women running 55 and 65 per cent above the UK rate, down from 85 and 105 per cent in the immigrant generation but nowhere near level. Germany's Turkish second generation still differs markedly from non-migrant Germans. The six-country study by Kulu and colleagues finds descendants sitting between the immigrant and native levels with third births in particular staying elevated in some groups, and their decade review of the whole literature reports some groups at native levels, some below, and some significantly higher, explicitly leaving open whether that diversity declines over generations or is here to stay.

The summary the evidence supports: immigrant fertility converges a long way, at different speeds, and in some groups it has not finished. Anyone telling you the literature has settled this in either direction has not read it.

The origin countries are meanwhile doing the thing nobody scripted. Turkey's fertility rate in 2024 was about 1.48, below France's 1.62 and, more to the point, above the two countries it was supposed to be Islamising: England and Wales came in at 1.41 that year and Germany at 1.35. Turkey now out-reproduces Germany. Iran is around 1.7, Tunisia 1.8, and Morocco's 2024 census put it at 1.97, below replacement for the first time in the country's history and down from 2.2 a decade earlier. The demographic transition did not stop at the edge of the Muslim world, and it is running faster there than it did in Europe.

Before the arithmetic, the case that cuts the other way, because it is in Europe and it is not disputed. In Kosovo the Albanian share of the population went from 68.5 per cent in 1948 to 81.6 per cent in 1991, while the Serb share fell from 23.6 to 9.9. Those are the figures the demographer Helge Brunborg put before the ICTY as an expert witness. Composition can change fast and permanently, driven by exactly the combination people fear: a large fertility gap plus sustained out-migration by the other group. It happened in Europe, inside living memory.

The case does not carry as far as it looks. It took an extraordinary fertility differential, Kosovo's rate being above six children per woman when the change began. And Kosovo's fertility rate today is about 1.54, below the EU average and only a shade above Sweden's own record-low 1.43. The engine that did it ran down within two generations, in the place it was running fastest.

The next calculation is mine rather than anyone's published finding, and it puts a number on what a permanent gap would have to do. Start with a 5 per cent minority. Hold Pew's 2.6-against-1.6 gap fixed forever. Allow no migration, no mortality difference, nobody leaving the religion. The minority's share of each birth cohort then runs 5, 7.9, 12.2, 18.4, 26.8, 37.4, 49.2 per cent, reaching just under half of births in the sixth generation, about a hundred and eighty years out if you allow thirty years to a generation. Narrow the gap the way Pew projects and the horizon moves out past any date the word "soon" can survive.

That calculation is aggressive about fertility and it is silent about migration, which I should say plainly rather than let it pass as a bound on the future. It is not one. Zeroing out migration deletes the variable that actually drives Pew's whole 7.4-to-14 spread, and half of births is not half of a population, which lags by decades. What the number bounds is what fertility alone can do, on its own, at the most extreme gap anyone has measured. Fertility alone is not what anyone is arguing about, and that is rather the point: strip the migration term out of the scary arithmetic and the arithmetic stops being scary, which tells you where the fear was actually living.

Where the story came from

The demographic panic about Muslims in Europe has a literature, and its own numbers keep failing in public.

Bat Ye'or's Eurabia (2005) is where the name comes from, and it is barely a demographic forecast. Its claim is about intent: that the Euro-Arab Dialogue, a real diplomatic process begun after the 1973 oil crisis, was in truth a bargain aligning Europe with Arab interests at the cost of its own culture. One line of scholarly attack targets the mechanism. Writing in Race & Class in 2006, Matt Carr argued that accepting the thesis means crediting a secret and unelected body with the ability to transform every major European institution without any parliament or newspaper noticing. Renaud Camus gave the other root phrase its currency with Le Grand Remplacement in 2011, a claim about racial substitution rather than diplomatic conspiracy; he calls it genocide by substitution and was fined €4,000 in 2014 for incitement to hatred against Muslims.

The single most-quoted line in the genre is Bernard Lewis telling Die Welt in July 2004 that Europe would be Islamic. The headline said that flatly. Lewis's own sentence about majorities was conditional: on current trends, Europe would have Muslim majorities in the population by the end of the twenty-first century at the latest. He did also say, without any qualification, that Europe would become part of the Arab West, the Maghreb. It was an interview remark, resting on the assumption whose failure this essay is about, and it has been circulating for twenty-two years in a form its author did not use.

Mark Steyn's America Alone (2006) made the claims checkable enough to check. Its most cited is that the evidence suggests a third of all births in France are already Muslim. France does not collect birth statistics by religion and has not asked religion in a census since 1872, so there is no evidence that could suggest it.

Then there is the viral video. In 2009 a seven-minute clip called Muslim Demographics passed ten million views, and BBC Radio 4's More or Less took it apart claim by claim. The video put Belgium at 25 per cent Muslim; the Belgian statistics office said 6. Its Dutch numbers, the BBC calculated, would require Muslim women in the Netherlands to be giving birth at at least fourteen times the rate of their neighbours. And it carried a quotation attributed to the German government: that the country's population decline could not be stopped, and that it would be a Muslim state by 2050. The first sentence was genuinely said, by Walter Radermacher of the Federal Statistics Office. The second was appended by someone else. Radermacher's response, on the record: the quotation reading as if the German government believed Germany would become a Muslim state is simply not true, and there is no source that can be quoted for it.

In January 2015 Fox News broadcast the claim that Birmingham was a totally Muslim city where non-Muslims do not go. Birmingham was 22 per cent Muslim at the 2011 census. PolitiFact rated it Pants on Fire, the guest who said it apologised for an inexcusable error, and Ofcom later found the broadcast in breach of the UK broadcasting code.

A fair objection to everything in this section: these are the easy opponents. A conspiracy theorist, a pamphleteer, a YouTube video, a cable-news error and a baby-name meme are not the strongest form of the argument, and picking them is a way of winning cheaply. The serious version has serious authors. David Coleman, a demographer at Oxford, published peer-reviewed projections of majority-minority transition in several European countries under the heading of a third demographic transition. Eric Kaufmann concedes the arithmetic I have spent this essay on and argues about something else entirely, which is what happens to a majority's sense of itself while the composition changes underneath it. Neither is touched by anything in this section, because neither made these mistakes. The case for worrying is theirs, not the video's.

The most durable item is the smallest. For years the claim that Muhammad was the most popular boy's name in Britain was true only if you combined its spellings while not combining Oliver's, and the ONS said so explicitly, pointing out that merging Sophie, Sophia and Sofia would rearrange the girls' list too. Since 2023 it is genuinely first on the single spelling, with 4,661 boys given the name that year, in a country where parents used 29,560 distinct names and spellings for boys and where the 2021 census of England and Wales counts Muslims at 6.5 per cent of the population. A name ranking measures naming convention. Muslim families concentrate on one name. British non-Muslim families have spent thirty years spreading out across a longer list.

None of which is harmless. In July 2022 Viktor Orbán told an audience in Romania that Hungarians are "willing to mix with one another, but we do not want to become peoples of mixed race", which is about as close as a sitting head of government has come to endorsing the frame outright. He lost power in April 2026, after sixteen years, to a party that beat him by fifteen points. And the man who murdered fifty-one people in two Christchurch mosques in March 2019 titled his manifesto The Great Replacement.

The country that tells this story about everyone else

The version of this meme that circulates in Russian has a problem at home.

Russia has more Muslims than any other country in Europe, and, on the standard ethnicity-derived estimate though not on self-identification, a higher share than any country in Western Europe. Marlène Laruelle's figure, and it is the standard one, is about 15 million nominally Muslim citizens out of 146 million, roughly a tenth, and her qualifier has to travel with the number: nominally means belonging to an ethnic group whose cultural background refers to Islam, and most are not practising. Ask people directly and you get much less. The largest religion survey ever run in Russia, Sreda's Arena atlas in 2012, found 6.5 per cent identifying as Muslim. Ask the muftiate and you get much more, twenty to twenty-five million. Russia's census asks no religion question at all, so every Russian Muslim figure is inferred from ethnicity, and the 2021 census left 11.3 per cent of those counted with no stated nationality, nearly triple the 2010 share. The uncertainty sits on the variable the whole argument runs on.

In 2019 the chairman of Russia's Council of Muftis, Ravil Gainutdin, told a conference at the State Duma that according to specialists whose competence was beyond doubt, in about fifteen years up to 30 per cent of Russia's population would be Muslim. He named no specialists and produced no model. It is the Eurabia genre in Russian, delivered by the other side, and it is quoted approvingly by the people you would expect.

Meanwhile the actual Russian numbers went the way the Estonian ones did. Russia's total fertility rate fell to about 1.37 in 2025, the lowest in two decades. Only three regions are above replacement: Chechnya at 2.60, Buddhist Tuva at 2.21, and the oil-and-gas district of Yamalo-Nenets at 2.01. Dagestan and Ingushetia, the two republics that anchor the Russian version of the fear, are both now below replacement. Bashkortostan is below the national average. The convergence that European commentators are still arguing about already happened inside Russia, and hardly anyone there has noticed.

What Russia does have is a structural labour shortage with unemployment near 2 per cent, and about four million citizens of Central Asian states present in the country as of September 2024, and remittance flows that hold up whole neighbouring economies. After the Crocus City Hall attack in March 2024, the response was a crackdown: Human Rights Watch counted 85,800 deportations in the first half of 2024, double the 2023 rate, alongside new powers to expel people without a court order and regional bans on migrants working in whole sectors. By April 2025, 70 per cent of Russians told Levada the authorities should restrict the inflow. Against which: foreign citizens accounted for 1.72 per cent of registered crimes in Russia in 2019, at a per-capita rate well below that of Russian citizens. That figure carries the same defect I complain about in the German statistics further down, and I should flag it here rather than only where it suits me: it counts by citizenship, so anyone who naturalised has moved into the Russian column, and naturalised former migrants are exactly the category Russian officials now point to.

And through all of it, Russian state television has kept running the story about Europe. In July 2026 Dmitry Kiselyov opened Vesti Nedeli with the British baby-name statistic as evidence that England and Wales are being reborn ethnically and culturally. InoSMI, part of the same state media group, published a Russian translation of a 2026 interview with Bat Ye'or under the headline "Secret documents confirm: Europe long ago capitulated to Islam". Eight years earlier the same outlet had translated a Die Welt article whose actual argument was that the popularity of one name proves nothing, and gave it a teaser asking whether Muhammad is Europe's most popular name.

The tell is what happened around 2020. The EU's disinformation monitor documents the Kremlin dropping the Islamisation line and repositioning itself as the protector of Muslims against a bigoted West. The same apparatus that had spent years warning that Europe was being overrun began accusing Europe of Islamophobia. The narrative changed when its usefulness did.

The problems that are real

If all I had to offer were the paragraphs above, I would be doing a version of the same trick, which is deciding in advance which facts are allowed. So here is the other column, with the numbers as their authors reported them.

Ruud Koopmans's 2013 study for the WZB found that among Muslims surveyed, 65 per cent said the rules of the Koran mattered more to them than the laws of the country they lived in, and 44 per cent agreed with all three of his fundamentalism items against under 4 per cent of native Christians. On out-group hostility, 45 per cent agreed that Jews cannot be trusted, against 9 per cent of Christians. The scope has to travel with it: the Muslim sample was people of Turkish and Moroccan origin only, in six countries, surveyed in 2008, and the internal range is enormous, from 15 per cent among Alevis to 50 among Moroccan-origin respondents. Koopmans also reported two things his admirers tend to drop, that socio-economic controls did not shrink the gap at all, and that in absolute numbers there are at least as many Christian fundamentalists in Western Europe as Muslim ones, most homophobes and antisemites being natives.

The 2016 ICM survey for Channel 4 found 52 per cent of British Muslims disagreeing that homosexuality should be legal in Britain, and 23 per cent supporting sharia in some areas. Its sample was drawn only from areas at least 20 per cent Muslim, which covers about half of Britain's Muslim population and skews toward the poorest neighbourhoods. That caveat belongs on the number every time, and it does not make the number go away.

In the EU agency FRA's 2018 survey of European Jews, 30 per cent of those harassed described the perpetrator of the most serious incident as someone with a Muslim extremist view, second only to "someone I cannot describe". Comparing the same seven countries across both waves, the figure was 32 per cent in 2012 and 29 in 2018, which FRA reads as no significant change: stable rather than surging, and that cuts against alarm and against dismissal at the same time. In the same survey 72 per cent of Jewish respondents said they were concerned about increasing intolerance toward Muslims.

Employment gaps are real and so is the discrimination that produces part of them. FRA's 2024 report puts Muslims aged 20 to 64 at 63 per cent in paid work against 75 per cent for the general population, with the gap concentrated among women. Field experiments show why some of that happens: in a French test holding national origin constant, an identically qualified applicant with a Muslim first name and Muslim voluntary work got 8 per cent positive replies against 21 per cent for her Christian counterpart. A five-country experiment with 7,284 applications found large origin penalties in Norway, the Netherlands and Britain and a smaller additional penalty for disclosing Islam, with none at all in Spain.

Crime data is where the arguments are worst, in both directions. Sweden's crime council found people born abroad 2.5 times as likely, and Swedish-born children of two foreign-born parents 3.2 times as likely, to be registered as crime suspects, falling to 1.8 and 1.7 after adjusting for age, sex, income, education and municipality. A residual that size is not explained away. In the same report, the share of offences linked to foreign-born suspects rose from 23.6 to 29.8 per cent between 2007 and 2018 while the proportion of every single group registered as suspects fell, which is the cleanest demonstration available that a rising share and a falling rate are not in conflict. Germany's police statistics for 2025 record 40.1 per cent of suspects as non-German, a figure that drops to 35.5 per cent once immigration offences that only foreigners can commit are excluded, and which classifies by citizenship, so German-born Muslims are counted as German and are invisible in it.

The grooming-gang scandals are the case where the accusation of institutional cowardice is correct. Alexis Jay's 2014 inquiry put a conservative estimate of 1,400 children sexually exploited in Rotherham between 1997 and 2013, found the majority of known perpetrators there were of Pakistani heritage, and recorded staff being told by managers not to identify perpetrators' ethnicity for fear of appearing racist. Baroness Casey's national audit in 2025 found ethnicity still unrecorded for two-thirds of group-based CSE suspects nationally, so no national conclusion can be drawn, while in Greater Manchester's multi-victim, multi-offender investigations 52 per cent of suspects were of Asian ethnicity against 20.9 per cent of the local population. The same audit supplies its own corrective: across all open child sexual abuse cases in Greater Manchester between January 2022 and February 2025, covering 908 suspects, 44 per cent were White and 16 per cent Asian, with a third of unknown ethnicity. The group-based cases are a small and unrepresentative slice of the whole. Casey's verdict on the data vacuum is that the collective failure to answer the ethnicity question served nobody, with energy spent proving the point on one side and playing it down on the other.

One finding complicates the religious reading of all this. Pew's 2018 survey of nearly 25,000 Western Europeans found Christians of both kinds, churchgoing and non-practising, more negative about Muslims, Jews and immigrants than the religiously unaffiliated, and the two Christian groups sat close together. In Britain, 45 per cent of churchgoing Christians and 47 per cent of non-practising ones said Islam is fundamentally incompatible with their country's values, against 30 per cent of the unaffiliated. Christian identity predicts the attitude better than churchgoing does, which is an argument against treating devout religiosity of any kind as the mechanism, and against treating Islam as uniquely productive of hostility. It is not a full answer, and strictly it answers a different question: Pew is measuring what majorities think of minorities, Koopmans what a minority thinks. Koopmans's most fundamentalist Christian subgroups also still sit far below his Sunni ones.

And here is the concession I owe, because I have spent this essay saying that migration status rather than religion is the operative variable. For fertility, the evidence says that clearly. For attitudes it does not. Koopmans found that controlling for education, employment, age and marital status did not shrink the Muslim-Christian gap at all, and the internal range in his data tracks ethno-religious subgroup, from 15 per cent among Alevis to 50 among Moroccan-origin respondents. Something is surviving the socio-economic controls there. I do not think it licenses any of the conclusions this essay is arguing against, but "religion isn't the variable" is a claim I can defend about how many children people have and not about what they told a survey in 2008.

What is actually shrinking

Discrimination in hiring, the segregation of poor neighbourhoods, attitudes to gay people, a police force that would not write down an ethnicity, a survey of what people think about sharia: these are questions about how a society works, and they are answerable, and some of the answers are ugly. They are not questions about how many people there will be in 2050.

That separation is the load-bearing move in this essay, so it should be argued rather than announced, and there is a version of the other side that it does not automatically survive.

The version goes like this. Nobody serious claims a majority. The claim is that a distribution which is stable inside a group has different consequences at 12 per cent than at 5, because the absolute numbers scale with the share. Koopmans's 44 per cent, ICM's 52, the FRA's 30: hold those proportions and double the denominator, and you have doubled the count. That is a demographic argument about social outcomes, and it does not need a single one of the projections I have spent this essay taking apart.

It is the best argument on that side. It assumes the distributions are constant, and the evidence says they are not, in the same way and for the same reason the fertility rates are not. Koopmans surveyed a first generation in 2008. The FRA's own data show reported discrimination rising between generations while its 2018 antisemitism figures stayed flat. The second generation is not the first generation in fertility, in language, in labour-force participation, or in what it tells a pollster, and an argument that lets the share move while holding the attitudes frozen has assumed the one thing most in doubt. Where the distributions do prove stable, the scaling argument goes through and I lose that round. It is a claim about integration outcomes, which are measurable, and it will be settled by measurement rather than by either of us.

What I do reject is the fusion of the two, because it converts a set of solvable problems into an arithmetic fate. Hopelessness is politically useful to people who would rather not fix anything.

Estonia is a decent place to see what the fused version costs. The 1988 projection did not produce a policy that stopped anything, because there was nothing to stop; the inflow had already ended and the arithmetic was already wrong when it was spoken. What it produced was a frame in which a third of the country's residents were a quantity rather than a set of neighbours, and that frame outlived the numbers by decades.

I said at the start that I was the variable in that arithmetic, and I should say what I actually mean by it, since it is the only thing in this essay I know from the inside. It is not a complaint about how I have been treated; my own experience here has been unremarkable, and I have written before that I have not had the hard version of any of this. It is narrower than that. There is a particular sensation in reading a projection that has your household in it as a coefficient, and then reading the follow-up thirty-eight years later that says the coefficient behaved differently than expected. Both documents are about you and neither is addressed to you. That is the whole of what being a demographic argument feels like, and it is why I am unmoved by the version of this debate that treats the counting as harmless because the counters are only doing sums. The counting is not the problem. It is the second sentence, the one about what should therefore be done, that has never once in this history been improved by the arithmetic being wrong. \n\nThe work that changed anything was slower and less dramatic. The proportion of non-Estonians who speak no Estonian at all has fallen from 20 per cent in 2008 to 4 per cent in 2023. The stateless share of the population has gone from 32 per cent in 1992 to under 5 per cent. Neither of those was achieved by counting anybody.

And the actual demographic threat to Estonia turned out to be a different one entirely. In 2024 the country recorded 9,690 births against 15,756 deaths, a natural decrease of just over six thousand, against net migration of 1,374. Migration is now offsetting a fraction of the decline rather than reversing it, and the population fell 0.7 per cent in the year to January 2026. The fertility rate has gone from 1.61 in 2021 to 1.16 in 2025, the lowest ever recorded here. And it is not one group doing it: births to ethnic Estonian mothers fell 33 per cent over those four years, against 30 per cent for births overall. Statistics Estonia's baseline scenario has the country at about 1.2 million people by 2085.

Meanwhile the national strategy Estonia 2035 sets its identity indicator on language rather than ethnicity: the share of residents with Estonian as a mother tongue, target at least 67 per cent, against 66.6 measured. The state's stated floor is a number the 1988 forecasters would have read as the beginning of the end.

Tiit, who has spent a professional lifetime on these numbers, is impatient with the whole register. Estonia having 1.3 million people and slightly over 900,000 Estonians is not a population crisis, she writes; Estonia is simply a small nation and will remain one. Peoples with their own states do not disappear for demographic reasons. If they disappear, it is through loss of identity.

That carries across to the larger question, and it is the part the arithmetic cannot touch. Whatever Europe is at risk of losing, no plausible projection has anyone taking it. What the projections describe is a continent that will be somewhat more Muslim, considerably less Christian, older, and smaller, in which the largest single religious movement of the past fifty years has been people leaving religion altogether. There are views to be had about that. Extrapolation is not how anyone will arrive at them: it is the one method that has been tried repeatedly here, in public, with dates attached, and has not yet been right.

One objection survives all of this. Someone can say that the alarm worked: that the 2015 panic produced the deal with Turkey, the hardened borders and the Pact, and that grading alarmists on forecasts that did not happen is grading a fire alarm on the fire it prevented. That is a serious argument and it is partly right about the last decade. It cannot explain Estonia, where the reversal had no policy behind it whatsoever, and it does not rescue the specific claims, because a warning is not entitled to credit for a fall in Turkish fertility or for the Assad regime collapsing. Some of the flattening was chosen, and the concession is owed.

Kuddo's number came due in the year 2000. Lewis's comes due at the end of this century, which is a safe distance, and that distance is the whole problem with the genre.

I am not going to give myself that distance. Pew's medium scenario is 11.2 per cent by 2050, and it is Pew's central case rather than its ceiling, so that is the number I will take: if the EU-plus-Norway-and-Switzerland share comes in above it, I was wrong. Pew is not extrapolation, which is why I am willing to be graded against it: it is a cohort-component projection with its assumptions written down where you can check them, and that is the difference between it and everything else in this essay.

Twenty-four years is still too long to wait, so here is a nearer one. By 2035 I expect that share to be under 8 per cent, and second-generation Turkish-origin fertility in Germany to be below 1.8. Nine years is short enough to be embarrassing.