Skip to content

Institute of Public Policies

Home » Publications » Policy Brief » The Left Behind: The Local Effects of Deindustrialisation in France

The Left Behind: The Local Effects of Deindustrialisation in France

Brice Berland, Marco Tabellini, Clémence Tricaud
February 2026
This paper analyses the long-term local consequences of deindustrialisation in France, from an economic, social and democratic perspective.
Reference IPP Policy Brief No. 123

Quote this publication

Berland, Brice, Tabellini, Marco, Tricaud, Clémence. ‘The Left Behind: The Local Effects of Deindustrialisation in France’. IPP Policy Brief, No. 123. Institut des politiques publiques (IPP).
Download

Presentation

The manufacturing sector has long been regarded as one of the pillars of modern economic and social development. Beyond its role as a driving force behind economic growth, industry has also been a major vehicle for social inclusion. The factory town was not merely a place of production: it was a social institution where work, identity and local solidarity intersected.

Since the 1970s, this world has changed profoundly. France entered the post-war period with one of the largest industrial sectors in Western Europe: in 1968, more than one in four workers was employed in industry, a proportion comparable to that of Germany and significantly higher than that of the United States at the time. France subsequently experienced one of the most marked industrial declines, which now places it amongst the most deindustrialised economies in the G7. Since 1968, the share of industrial employment has fallen from nearly 27 per cent to around 12 per cent in 2016, representing a net loss of more than 2.5 million jobs.

This paper analyses the long-term local consequences of deindustrialisation in France.

Key Results

  • Deindustrialisation has led to long-term economic stagnation: a 10-percentage-point decline in the share of industrial employment between 1968 and 2016 is associated with an unemployment rate that is 1 percentage point higher and an average per capita income that is 3.8 per cent lower.
  • Deindustrialisation has contributed to social isolation: a 10-percentage-point decline in industrial employment is associated with a 6.5 per cent increase in the proportion of people living alone and a fall in marriage rates.
  • Deindustrialisation has reshaped the demographic composition of French municipalities: a 10-point fall in industrial employment is linked to slower population growth, mainly due to out-migration, as well as to a higher demographic dependency ratio, reflecting the ageing of the population.
  • Industrial decline is associated with a marked weakening of contemporary community life: municipalities that have experienced a further 10-percentage-point decline in industrial employment since 1968 now have 9.5 fewer local amenities per 10,000 inhabitants and are seeing a sharp decline in the formation of new associations.
  • Deindustrialisation has lasting political consequences: the weakening of economic and social structures is accompanied by a fall in voter turnout, a decline in support for European integration and a significant increase in votes for the far right.

Method and Data

Data

This study draws on four main sources of data at local authority level:

  • the supplementary census databases made available via Saphir (INSEE), covering eight census rounds (1968, 1975, 1982, 1990, 1999, 2006, 2011 and 2016). This enables us to measure the proportion of industrial employment amongst residents, as well as local socio-economic characteristics, notably household composition, age and educational attainment.
  • Additional economic variables drawn from Piketty and Cagé (2023) to perform the evaluation of the economic impact of deindustrialisation beyond employment: per capita income, local GDP and property prices. The political variables are also taken from Piketty and Cagé (2023).
  • Municipal results from the presidential and parliamentary elections from 1967 to 2017.
  • The 2018 Permanent Database of Local Facilities (BPE) to measure the number of local amenities available in each municipality.
  • The National Register of Associations (RNA) to calculate the number of associations established in each municipality between 2018 and 2024

Analysis sample

For each supplementary census, a random sample of 20% to 25% of individuals is surveyed in each municipality. To ensure the statistical reliability and representativeness of the data, the authors exclude municipalities with fewer than 500 inhabitants. They also exclude municipalities that underwent boundary changes during the period under study. The final sample comprises a panel of 10,258 municipalities that we track across the eight census waves, from 1968 to 2016.

Methodology

The authors use two complementary empirical strategies

(1) First-difference regressions (panel). This approach analyses changes in the proportion of residents employed in industry between each of the eight census waves and links these to changes in our economic, social and demographic indicators over the same period. It controls for municipal fixed effects and fixed effects specific to each census period, thereby capturing intra-municipal variations over time. The estimated coefficient measures the average association, within a municipality, between changes in industrial employment and concurrent changes in economic or social indicators, whilst taking into account municipal characteristics that remain constant over time and national trends affecting all municipalities simultaneously.

(2) Long-run difference regressions. This approach relates the overall changes in the indicators between 1968 and 2016 to the corresponding long-term change in industrial employment. Municipal fixed effects would account for all the variation in this specification; the authors therefore include departmental fixed effects to account for different regional trajectories. This approach has two objectives. Firstly, by focusing on cumulative long-term effects rather than short-term adjustments, it captures the full extent of the structural transformation observed over five decades. Secondly, by condensing the panel into a single period, it limits potential biases associated with heterogeneous treatment effects in fixed-effects models.

All regressions are population-weighted.

Partners

Agence nationale de la recherche (ANR)

This paper was supported by the ‘Investissements d’avenir’ programme (ANR-11-IDEX-0003/Labex Ecodec/ANR-11-LABX-0047).

Last modified: July 21, 2026