OECD Regions and Cities at a Glance 2026

Annex C. Methodologies

Contribution of industries to coefficient of variation of gross value added (GVA) per worker

The decomposition identifies the contribution of each industry to inequality in GVA per worker across regions. It first expresses total GVA per worker as the sum of additive industry components, so for each region i and industry f:

$y_i=\frac{{GVA}_i}{{EMP}_i}$, $y_{fi}=\frac{{GVA}_{fi}}{{EMP}_i}$, with the region-industry components adding up to total GVA per worker: $y=\sum _f{y}_f$.

The coefficient of variation (CV) of GVA per worker is the ratio of the standard deviation of GVA per worker over the mean of GVA per worker, so the CV for total GVA per worker and per industry are:

${CV}_{total}=\frac{{\sigma }_y}{\stackrel{-}{y}}$, ${CV}_f=\frac{{\sigma }_f}{{\stackrel{-}{y}}_f}$, with ${\sigma }_y=\sum _f{\rho }_f{\sigma }_f$

The total coefficient of variation can be expressed as the sum of the contributions of each industry: ${CV}_{total}=\sum _f{C}_f$

Where: $C_f={\rho }_f\left(\frac{{\stackrel{-}{y}}_f}{\stackrel{-}{y}}\right){CV}_f$, with ${\rho }_f$ equal to the correlation between industry $f$and total GDP per capita: ${\rho }_f=Corr(y_f,y)$. With $\left(\frac{{\stackrel{-}{y}}_f}{\stackrel{-}{y}}\right)$ interpreted as the share of industry f in the sum of regional GVA contributions (so as an unweighted average-region share1), ${CV}_{total}=\sum _f{\rho }_f\left(\frac{{\stackrel{-}{y}}_f}{\stackrel{-}{y}}\right){CV}_f$.

The sectors initially considered are grouped as: construction (ISIC F), manufacturing (C), consumer and personal services (trade, transport, accommodation (G-I) and arts, recreation and other services (R-U)), and knowledge-intensive business services (J, K, M-N). To focus on the comparison between KIBS and manufacturing two sectors are excluded: construction and primary industries. The contribution of construction is small across all 32 countries. The contributions from primary industries are low in almost all countries, except Australia and Canada, where they contribute the most, and Colombia, where it fluctuates between a medium and high contribution.

To show the effect of each component more clearly, Figure C.1 focuses on KIBS and manufacturing and shows for each sector the CV (Panel A), the correlation with total GVA per worker (Panel B), the (unweighted) shares in GVA (Panel C) and the resulting contributions (Panel D). The combination of a negative correlation with other components may result in negative contributions from an industry. However, additivity ensures that these negative contributions will be compensated by other industries.

A sector's contribution to regional inequality is large when three elements reinforce each other: (1) high regional dispersion of sector productivity (Panel A), (2) a strong positive correlation between sector productivity and total regional productivity (Panel B), and (3) a large sector share in GVA (Panel C). Thus, a sector can have a very high CV but still make only a moderate contribution if its economic weight is small; similarly, a large sector contributes little if its regional productivity pattern is weakly related to the overall pattern. A negative correlation is particularly important: it can generate a negative contribution, meaning that the sector offsets rather than reinforces overall regional inequality. The level of total CV then provides the context: the same 80% contribution can represent a dominant source of a relatively small total disparity or of a much larger one.

Figure C.1. Components of CV of regional GVA per worker decomposition

Note: OECD unweighted country average.

Statlink

To identify which industries systematically underpin regional inequality, the figures by country (Figure C.2) show the contribution of five broad industries to regional inequalities over 2005-2024. To obtain a sufficiently complete and comparable regional time series, some missing regional observations are estimated by applying the corresponding national-level growth rate to TL2 regional values.

Figure C.2. Industry contributions to regional labour productivity disparities by country

Statlink

Labour market areas definitions

The labour market area (LMA) analysis builds on official definitions for 15 OECD countries, drawing on national statistical offices and other government agencies. Although their names differ across countries, these geographies follow the same broad principle: they group small administrative units or other subnational areas so that most working residents work within the area and most jobs are filled by people living there. Academic researchers and other institutions have defined similar labour market geographies for additional countries, but this analysis uses only definitions published or provided by official national sources. Countries use four broad approaches to delineate LMAs, most of which rely primarily on commuting flows between small administrative units or other territorial building blocks:

• LMA/Travel to Work Area (TTWA) algorithm: Canada, Denmark, France, Hungary, Italy, Switzerland, United Kingdom;

• Commuting zones based on hierarchical clustering: Japan, Mexico, United States;

• Tailored algorithms based on commuting flows and other variables: Australia, Germany, Norway, Sweden;

• Bottom-up governance approach: Netherlands.

The LMA/TTWA algorithm iteratively combines small administrative units or other building blocks into areas that meet criteria for labour force size and self-containment. Self-containment has two dimensions: demand-side self-containment measures the share of people working in an area who also live there, while supply-side self-containment measures the share of working residents who also work there. The algorithm evaluates the lower of these two measures and applies four parameters: minimum and target levels of self-containment, and minimum and target numbers of workers. A grouping can compensate to some extent for a smaller workforce through higher self-containment, or for lower self-containment through a larger workforce, but it must meet the minimum threshold for both dimensions. The methodology builds on the TTWA approach developed in the United Kingdom (Coombes, Green and Openshaw, 1986[1]) and is further described in Eurostat methodological guidance (Eurostat, 2020[2]) and an R package developed by Istat (Istat, 2014[3]).

The resulting LMAs share a broad methodological basis, but differences in parameter choices and additional country-specific processing steps limit full comparability across countries. Table C.1 shows that countries apply substantially different parameter values, particularly for the minimum and target numbers of workers. National authorities also undertake post-processing with input from national and local experts to reflect territorial circumstances, including assigning building blocks that the algorithm places on a reserve list.

Table C.1. Parameters used by countries that implemented the LMA/TTWA algorithm

Country	Min. self-containment	Target self-containment	Min. number of workers	Target number of workers	Note
Canada	75%	90%	0	25 000	Canada delineates self-contained labour areas for areas outside of census metropolitan areas and census agglomerations
Denmark	65%	80%	2 500	35 000
France	60%	70%	15 000	25 000	Special method for Corsica, Île-de-France and the oversea departments
Hungary	76%	78%	5 000	20 000
Italy	60%	75%	1 000	10 000
Switzerland	70%	90%	50 000	80 000	Two-step procedure in which small administrative units are combined to 101 Arbeitsmarktregionen, which are then combined to the 16 Arbeitsmarktgrossregionen used here
United Kingdom	66.7%	75%	3 500	25 000	Building blocks vary by country

The other approaches pursue the same broad objective through different procedures. Hierarchical clustering groups areas according to the strength of their commuting links. Tailored national algorithms combine commuting flows with additional criteria: Australia considers the number of employed people and distance; Norway uses travel time; Germany uses travel time and administrative boundaries; and Sweden relies on commuting flows alone. Netherlands uses a bottom-up governance approach that allows municipalities and regional labour-market partners to propose and agree changes to labour market region boundaries.

The median population of Labour Market Areas varies substantially across countries. Table C.2 shows that the median LMA has considerably more residents in densely populated countries such as the Netherlands than in more sparsely populated countries such as Australia or Norway, reflecting differences in national settlement patterns. Differences in delineation methods, parameter choices and territorial building blocks can also contribute to this variation. A substantial share of LMAs consists entirely of towns, semi-dense areas or rural areas, whereas most TL3 regions contain at least one city.

Table C.2. Median population of LMAs and TL3 regions, thousands of people

Country	LMA	TL3		Country	LMA	TL3
Australia	8	191		Mexico	30	298
Canada	10	41		Netherlands	411	341
Denmark	79	582		Norway	7	314
France	126	532		Sweden	38	285
Germany	182	155		Switzerland	442	246
Hungary	43	356		United Kingdom	151	335
Italy	42	377		United States	233	848
Japan	118	1 541

Note: Values refer to the most recent year available and are rounded to the nearest thousand.

To ensure consistency between the LMA and functional urban area geographies, the analysis presented in Chapter 2, Figures 2.14 and 2.15 merges LMAs when more than 50% of the population of each LMA lies within the same FUA. Figures 2.12 and 2.13 of Chapter 2 do not apply this adjustment, as they rely partly on unemployment rates countries publish directly at the LMA level. Figure C.3 illustrates all LMA merges in France. For example, two LMAs are merged around Lyon, while three are merged around Bordeaux.

Figure C.3. Merging French LMAs that each overlap with the same FUA for more than 50% of their population

As described above, LMAs group adjacent small administrative units into functionally connected areas. Figure C.4 illustrates this for LMAs around Bordeaux, showing the LMA groupings together with the degree of urbanisation of each small administrative unit. Cities, towns and semi-dense areas, and rural areas can all occur within the same LMA. The white boundaries identify LMAs that the analysis presented in Figures 2.14 and 2.15 of Chapter 2 merges because more than 50% of the population of each LMA lies within the Bordeaux FUA, to ensure consistency between the two geographies.

Figure C.4. Degree of urbanisation of small administrative units and LMA boundaries

Note: The map shows French LMAs overlapping with the TL3 regions Dordogne, Gironde, Lot-et-Garonne, Lot, and Corrèze. Thick black lines represent LMA borders, thin black lines represent borders of small administrative units and white lines represent LMAs the analysis merges because they overlap with the Bordeaux FUA by more than 50% of their population.

Figure C.5 classifies the same LMAs according to the most urbanised small administrative unit they contain. An LMA is classified as a city LMA (shown in blue) if it contains at least one small administrative unit classified as a city. It is classified as a town/semi-dense LMA (shown in yellow) if it contains no city but at least one town or semi-dense area. An LMA composed exclusively of small administrative units classified as rural areas is classified as a rural LMA (shown in green).

Figure C.5. Classification of LMAs by degree of urbanisation

Note: The map shows French LMAs overlapping with the TL3 regions Dordogne, Gironde, Lot-et-Garonne, Lot, and Corrèze. Thick black lines represent LMA borders, thin black lines represent borders of small administrative units and white lines represent LMAs the analysis merges because they overlap with the Bordeaux FUA by more than 50% of their population.

Robustness checks

Several checks assess the sensitivity of the results to LMA delineation and differences in the number of LMAs across countries. Merging LMAs with more than 50% of their population within the same functional urban area (FUA) has little effect on the results. Similarly, applying standardised delineation parameters instead of official national parameters for countries using the LMA/TTWA algorithm leaves the broad cross-country ranking unchanged. To assess whether the number of LMAs mechanically affects the coefficient of variation, a Monte Carlo exercise drew 10 000 random samples of 29 LMAs per country, corresponding to the number of LMAs in Denmark. The average simulated coefficients of variation remain close to the values observed from the full sample, suggesting that differences in the number of LMAs do not, in themselves, drive the results.

Population weighting has a greater influence on the results. The unweighted coefficient of variation used in the analysis gives equal weight to each LMA, whereas the population-weighted measure assigns greater weight to more populous LMAs. Population weighting generally reduces measured dispersion, particularly in Australia and Canada, where LMA population sizes vary substantially. By contrast, the coefficients of variation change relatively little in Italy and Hungary. Neither approach is inherently preferable: the choice depends on whether the analysis aims to give equal importance to differences between LMAs or to reflect the distribution of LMA unemployment rates across the population.

Social infrastructure and Points of Interest (POI) data from OpenStreetMap (OSM)

The analysis on social infrastructure includes analysis on eating and drinking facilities, cultural facilities and public community facilities. These categories do not constitute an exhaustive measure of social infrastructure. The analysis draws on OpenStreetMap (OSM), an open, collaboratively maintained geographic database where contributors describe geographic features using tags consisting of keys and corresponding values. Table C.3 lists the tags used to identify facilities in each category.

Table C.3. OSM tags used to identify social infrastructure facilities

Social infrastructure dimension	Values for the OSM key amenity	Other keys and values
Eating and drinking facilities	bar, biergarten, cafe, fast food, food court, ice cream, pub, restaurant	shop: ice cream
Cultural facilities	arts centre, cinema, community centre, events venue, library, music venue, nightclub, theatre	leisure: dance, tourism: gallery, museum
Public community facilities	community centre, social centre, social facility	–

Note: For features tagged as community centres, an additional community centre key with the value cultural centre identifies cultural facilities. All other community centres are classified as public community facilities.

The data-cleaning procedure harmonises the geographic representation of facilities and aims to count each facility only once. It involves three main steps:

1. Resolve duplicate records. Occasionally, the same OSM identifier appears in multiple records, often with conflicting country assignments. Such duplicates can lead to double counting or the incorrect attribution of facilities to a country. The procedure uses geographic boundaries to identify the appropriate record and excludes cases that cannot be resolved.

2. Remove overlapping representations. Different OSM records may represent the same physical facility as points, lines or polygons, creating a risk of double counting. For each specific facility type identified through the OSM tags in Table C.3, the procedure removes points located inside corresponding polygons, lines covered by such polygons along more than half their length, and polygons overlapping other polygons by more than half their area. These rules apply to facility-sized polygons with a bounding-box diagonal of up to 500 metres, avoiding the removal of distinct facilities within extensive mapped areas.

3. Standardise facility locations. The procedure converts retained polygons to centroids and lines to midpoints, combining them with existing points to obtain a consistent representation of facilities.

Country selection draws on estimates of OSM building completeness in urban centres (Herfort et al., 2023[4]). Building completeness serves as an indirect indicator of the likely coverage of mapped facilities. Comparisons of cultural facility counts across settlements of similar population show that countries with particularly low building completeness also tend to record substantially fewer facilities. Geographic coverage excludes Nunavut (Canada); Ceuta, Melilla and the Canary Islands (Spain); Guadeloupe, Martinique, French Guiana, Réunion and Mayotte (France); and Svalbard and Jan Mayen (Norway).

The indicators capture the presence and number of mapped facilities, but not differences in their characteristics. Facilities can vary in size, capacity, quality, frequency of use, activities, affordability and funding. Some activities may also take place in multifunctional spaces without being reflected in their OSM classification. Furthermore, the indicators do not establish whether facilities are accessible to everyone or measure the social interactions that take place there. They are therefore primarily suited to identifying geographic regularities in facility provision rather than assessing individual facilities.

References

[1] Coombes, M., A. Green and S. Openshaw (1986), “An Efficient Algorithm to Generate Official Statistical Reporting Areas: The Case of the 1984 Travel-to-Work Areas Revision in Britain”, Journal of the Operational Research Society, Vol. 37/10, pp. 943-953, https://doi.org/10.1057/jors.1986.163.

[2] Eurostat (2020), European harmonised Labour Market Areas - Methodology on functional geographies with potential, https://doi.org/10.2785/328723.

[4] Herfort, B. et al. (2023), “A spatio-temporal analysis investigating completeness and inequalities of global urban building data in OpenStreetMap”, Nature Communications, Vol. 14/1, https://doi.org/10.1038/s41467-023-39698-6.

[3] Istat (2014), Istat implementation of the algorithm to develop Labour Market Areas, https://www.istat.it/en/files/2016/03/Description-of-the-LabourMarketAreas-algorithm.pdf.

Note

← 1.  Note that this is different from the industry’s share in total GVA in all cases except where populations shares are identical.