France's 2023 riots peaked in three nights. In 2005 it took eleven

On 27 June 2023 a police officer in Nanterre, west of Paris, shot dead Nahel M., a 17-year-old driver, at a traffic stop. A video of the shooting contradicted the police's first account. That night the riots began. Over the following week, by the French Senate's later count, rioters in 672 towns burned 12,031 vehicles and set fire to or damaged 2,508 buildings, among them 273 police and gendarmerie stations, 243 schools and 105 town halls. Insurers paid out €793m.

France had been here before. In October 2005 two teenagers hiding from the police in Clichy-sous-Bois, east of Paris, died in an electricity substation, and three weeks of riots followed. Set side by side, the two episodes say something about how riots now start and stop. This autumn, as protests over the budget turned violent across France, the comparison has a use.

Three nights, not eleven

The Interior Ministry publishes the number of vehicles burned each night. It is the one measure counted the same way in both episodes.

In 2005 the riots stayed in Clichy-sous-Bois and the rest of Seine-Saint-Denis for five nights and reached cities outside the Paris region only around the seventh. The peak came on the eleventh night, and the nightly count did not fall below 100 until the twenty-first. In 2023 there was no slow start. The count went from 115 vehicles to 788 on the second night and 1,919 on the third. After the peak it fell by almost half each night; in 2005 it fell by about a quarter.

The riots travelled the same route both times. Researchers at Sciences Po who mapped both episodes from the press, Lorenzo Guillaume-Le Gall and Marco Oberti, find the same order: the suburbs of Paris first, then the suburbs of other big cities, then medium and small towns. In 2005 the small provincial towns joined in the second week. In 2023 the Paris suburbs were already a minority of the riot towns from the third night, overtaken mainly by small and medium towns.

To see this for ourselves we built a list of towns from news reports: 3,654 articles from the French and international press, read one by one, each place checked against its source (see the box at the end). By the end of the second night, rioting had been reported in 122 towns in 34 of France's 96 mainland départements, more than half of them outside the Paris region.

What compressed two weeks into three nights was, most likely, the phone. In 2005 news of a riot travelled by television and word of mouth. In 2023 it travelled as video, within minutes. The Senate's inquiry found that riot content online slightly preceded the riots themselves, and a draft study of Instagram by three economists finds that a town whose residents saw more posts of riots elsewhere was more likely to riot the next day.

The state's response was different too. In 2005 the government deployed 11,500 police and gendarmes and, just after the peak, declared a state of emergency that let prefects impose curfews. In 2023 it put 45,000 on the streets from the fourth night, and some mayors set curfews of their own. Arrests were made faster: at the peak, 0.46 arrests for every vehicle burned, against 0.28 in 2005.

It is tempting to conclude that the bigger deployment ended the riots sooner. The numbers cannot show it. The police were sent because the riots grew, and the decline began the night the 45,000 arrived, so the two move together whatever the cause. A wave that rises fast may also burn out fast.

Which towns

The list of towns can be set against the census. We took every French town of 10,000 people or more, 993 in all, and split them three ways by poverty and three ways by the share of residents who are immigrants.

Both matter. Among the least poor towns with few immigrants, none on our list rioted. Among the poorest with many, more than half did. And within each band of poverty, towns with more immigrants were more likely to riot: among the poorest third, 14% of those with few immigrants against 56% of those with many. The pattern is the same outside the Paris region. Almost nine in ten of the riot towns contain one of the state's "priority neighbourhoods", the poorest districts singled out for urban policy.

Our list misses many towns, so the true shares are higher; Guillaume-Le Gall and Oberti, with a fuller list, find that 57% of towns of 10,000 people or more had some rioting. But the pattern matches theirs and that of a second Sciences Po study using police records, which finds that a town's share of immigrants and the segregation of its poor households predicted rioting, and its poverty rate alone did not.

This is about places, not people. It does not show that immigrants rioted; most of those convicted were born in France (see below). Nor can the data separate immigration from what comes with it in these towns: the large housing estates, the young population, the history of tension with the police. What it does show is that the riots were not random. They came to the same kind of place, all over the country, at the same time.

Who was caught, and who rioted

Officials have published a portrait of the rioters, drawn from those who were arrested and tried. The inspectorates of the interior and justice ministries studied 395 adults convicted by the end of July 2023. Their average age was 23; 91% were men; 75% were born in France; 57% had no criminal record. A third of the 3,500 people arrested by 4 July were minors. Among the minors brought before a judge, six in ten came from single-parent families, against a quarter nationally.

That is a portrait of who was caught. About 4,300 people were held in police custody. The interior minister put the number of rioters at 8,000 to 12,000; a poll by Ifop suggested 50,000. On those figures, between a third and a twelfth of rioters were arrested, and arrests were not random either. In the first nights the police were told to hold the streets, not to make arrests. The Senate's inquiry warned that the arrest figures describe those caught in the act, many of them amateurs, and are too thin to say even how old the rioters were.

The one riot where both sides have been measured is England's, in 2011. The Guardian and the London School of Economics interviewed 270 rioters, most of whom were never arrested, and compared them with those who went to court. Their ages matched. But women were 21% of the rioters interviewed and 10% of those in court, and those with a past caution or conviction were about two-thirds of the interviewees (68%) against three-quarters of the suspects (76%). The courts saw more of the usual suspects and fewer women. If France's arrests worked the same way, the true rioters of 2023 included more women and fewer people known to the police than the official figures show. We cannot say by how much.

2026: a different kind of unrest

France has been in the streets again this autumn. The anger began on social media against a rise in fuel prices, then widened to the cost of living and to the €54bn of savings the government wants in next year's budget. Secondary-school pupils blockaded their schools first, from September. On Saturday 17 October they were joined by unions, teachers, farmers and former Yellow Vests: 550,000 people across the country by the Interior Ministry's count, 800,000 by the organisers'.

That afternoon and night the marches turned violent in several cities. By the ministry's count 1,522 shops and businesses were damaged or looted and firefighters were called out 2,025 times. In Limoges the old wing of the town hall, built in the 1880s, burned until its roof collapsed. In Paris a 59-year-old man died of heart and lung failure during the march; a judicial inquiry is under way. The ministry's final count for the day was 2,628 arrests and 1,144 police and gendarmes injured.

On the first day alone there were twice as many arrests as on 2023's worst night, and more injured officers than in the whole week of 2023 (782). But arrests measure policing as well as unrest. These were protest days announced in advance, and the police were ready: the government put 70,000 police and gendarmes on the streets on the following Saturdays, against 45,000 at the height of 2023, banned marches near the Élysée and the National Assembly, and checked some 40,000 people before the march of 7 November. In 2023 the police were caught by surprise and told at first to hold the streets rather than arrest. The two sets of numbers are not counts of the same thing.

The shape is what can be compared, and it is not 2023's. The 2023 riots were one wave of nights that rose and fell within a week. This movement came back every week, like the Yellow Vests of 2018-19, which also began with fuel prices. But each Saturday was smaller and calmer than the last. The ministry counted 410,000 marchers on 24 October, 320,000 on 31 October and 146,000 on 7 November, a fall of nearly three-quarters in three weeks, and arrests fell from 2,628 to about 360. The unions' strike day on 5 November drew 575,000 people by the ministry's count, and 1.6m by the CGT union's, with fewer arrests than any Saturday before it.

The government answered with force and with money. On 19 October the prime minister, Sébastien Lecornu, asked prefects for local curfews, and President Emmanuel Macron gave prefects powers to ban demonstrations. On 9 November it raised the minimum wage by 1.6% from December. More days of protest are called, among them a public-sector strike on 17 November, so the movement may yet revive. And the question that matters most for comparison with 2005 and 2023 is still open: where it struck. In 2023 the riots returned to the same places: of the 347 towns that rioted in 2005, 92% rioted again in 2023, by the count of the second Sciences Po study. Reports from this autumn describe the violence mostly in city centres. When the movement is over we will build the same list of towns, and see whether it reached the suburbs of 2005 and 2023 or new places altogether.