Atlas of the player pool

German players in Europe's 9 top-ranked leagues per million inhabitants, 2025/26

A floodlit football stadium at night seen from above, in black and white with only the green of the pitch kept

Research · Player pool atlas, German edition · Germany and 7 peers, 2025/26

3.13 players per million in Europe’s 9 top leagues, 2025/26

Per head, Germany ranks 8th of 8.

* 262 players who appeared (any minutes), per 83.6 M inhabitants; an administrative count. Methodology.

Photo: Emessem T · CC BY-SA 4.0, colour held

The question and what this atlas adds

How many of a country’s players reach Europe’s top-ranked leagues, and how that number relates to the size of the country and to the path its players take, is usually studied in pieces. Research on player development looks inside clubs and national systems: at the environment of a Danish club with a record of producing senior players (Larsen et al., 2013), at how selection and de-selection shape who progresses in German talent promotion (Güllich, 2014), at the working relationship between youth and professional departments at elite European clubs (Relvas et al., 2010), and at selection biases such as the relative age effect in European youth football (Helsen, Van Winckel and Williams, 2005). Counts of players abroad are published by the CIES Football Observatory (Poli, Ravenel and Besson, 2024), without a per-head comparison of one nation against a fixed peer set. This atlas takes Germany as a worked example and puts, on one set of definitions and from public data only, the per-head count of players in the 9 top-ranked leagues beside measures of the path to them: youth minutes at home, the age and route of the first move abroad, playing time after it and the national-team squad. It is exploratory and descriptive and estimates no causal effect.

Findings

  1. 1

    German players with 450+ Big-5 minutes: 253 at the 1995/96 peak, 161 in 2025/26; most of them play in their own league.

    The change-point model’s most probable season for the fall is 2000/01 (posterior 78 %), a level change of ×0.75 (90 % HDI 0.65–0.89). A break model dates a change in level; it identifies no cause.

    More on this

    The median age of the nation’s Big-5 players was 27 at the 1995/96 peak and 28 in 2000/01, when 24 players made a first Big-5 season of 450+ minutes.

  2. 2

    The per-head gap to France and Spain, split over three measured channels (descriptive).

    In a ridge fit over 8 countries, against France (a gap of 2.66 per million) youth minutes go with −1.55 (90 % bootstrap interval −4.75 to +2.05), league strength with −1.68 (−2.38 to −0.36) and export age with +4.26 (+0.23 to +9.68), residual +1.63; against Spain (a gap of 6.41 per million) youth minutes go with +0.78 (90 % bootstrap interval −1.02 to +2.46), league strength with −6.67 (−9.36 to −0.95) and export age with +2.13 (+0.34 to +4.80), residual +10.16.

    More on this

    The league-strength coefficient in this fit is negative (−51.1 per unit of the league multiplier): a weaker home league goes with more players in the top-ranked leagues per million, so a peer whose league is weaker than Germany’s (m_L 0.777) gets a positive league-strength contribution. The split describes an association across few, correlated national measures; it is not a causal accounting.

  3. 3

    Germany’s own under-21 nationals’ share of home-league minutes over 6 seasons: a trend of +0.50 pp a season (90 % CI +0.08 to +0.92).

    The share by season, 2020/21 to 2025/26: 3.4, 4.3, 3.2, 5.6, 4.7, 6.1 %. An OLS line through the 6 seasons rises by 0.50 percentage points a season (90 % CI +0.08 to +0.92 pp). The last step, 2024/25 to 2025/26, for the comparison countries and for other leagues whose share fell by 2 points or more: France 8.7 → 8.8 %; Spain 7.8 → 4.8 %; Czechia 11.1 → 6.4 %; Poland 12.5 → 9.3 %.

  4. 4

    Where Germany ranks among 8 countries on the pathway measures, 2025/26 (1 = highest share, youngest age, most moves).

    Under-21 share of home-league minutes 6.1 %, 4 of 8 (highest: Netherlands 13.6 %); regular under-21 starters per club 1.06, 4 of 8 (highest: Netherlands 1.89); median age at the first season with real playing time abroad 25 (n = 316), 7–8 of 8 from youngest (youngest: Belgium 22); first moves abroad in the covered seasons 316, 1 of 8; home-league UEFA-coefficient multiplier ×0.79, 4 of 8; median share of club minutes for exports 42 % (90 % bootstrap interval 34–46 %, n = 262), 1 of 8 by the median. These are descriptive ranks on medians and counts; apart from the minutes share they carry no interval, so neighbouring ranks can swap with a few players; none of them is shown to cause the per-head count.

  5. 5

    Youth minutes in Germany and England after two academy reforms (descriptive).

    Germany after its DFB/DFL academy licensing (2001/02; Honigstein, 2015): its own under-21 nationals played 5 % of Bundesliga minutes the season before, 11 % at the highest point within twelve seasons (2009/10) and 6 % in 2025/26. England after its Elite Player Performance Plan (2012/13; Premier League, 2011): its own under-21 nationals played 3 % of Premier League minutes the season before, 6 % at the highest point within twelve seasons (2019/20) and 3 % in 2025/26. Each is one sequence in one country without a comparison group, so it describes timing and does not estimate an effect of either reform.

Each finding is generated from the pipeline outputs of this edition and of the six-edition comparison; all are descriptive, and the evidence is on the question pages.

  • published 15 September 2026 · updated 1 October 2026 · version 2 (change log)
  • status research, exploratory, not pre-registered. The one registered item is the Big-5 forecast for — (registered forecast: numbers committed at — on —; commit times are self-reported).
  • data FBref season tables via soccerdata 1.9.1, fetched 14 September 2026; Big-5 history from 1995/96, committed 22 September 2026; Wikipedia squad lists (CC BY-SA 4.0), Wikidata (CC0), Eurostat, population on 1 January 2024 (table demo_pjan), UEFA association coefficients; snapshot commit ef7dbd4. Sources, licences and hashes: data availability.
  • code github.com/sandovabarbora/czefootball-player-pool-atlas, data snapshot at ef7dbd4; rebuild: uv sync && make restore-snapshot && make figures && make pages (reproduction)
  • cite as Šandová, B. (2026). German football player pool atlas. football.bsandova.com/ger/, version 2.
  • licence code MIT; text, figures and data CC BY 4.0; player portraits are Wikimedia Commons files under their own licences (footer).

One-page brief

The five pathway measures on one screen with their samples, the largest channel in the gap decomposition, and two model checks with their intervals.

Show the one-page brief
Number GERFRAESP n (GER)
Share of league minutes that went to players aged 21 or under6.1 %8.8 %4.8 %51 players, 18 clubs
Average age of a minute played in the league26.125.626.9604 575 league minutes
Median age at the first season with real playing time in a foreign league, over the players abroad today252325316 players
Share of players abroad in a league no stronger than their own (UEFA-coefficient multiplier)92 %——288 players abroad
Players in Europe's top-ranked leagues, for every million people3.135.799.54262 players

In the gap decomposition (descriptive, see its page), the channel with the largest contribution is how old players are when they move abroad for France and how strong the domestic league is for Spain.

How strong is the home league
A season in the home league is worth about 0.78 of a season in the Premier League (90 % HDI 0.74–0.81).
Most probable season of the dated step in the Big-5 count
2000/01 (posterior 78 %); level ×0.75 (90 % HDI 0.65–0.89).

What none of this shows is listed once, under limitations on the methodology page.

How to cite

Šandová, B. (2026). German football player pool atlas. football.bsandova.com/ger/, version 2. Code and data: github.com/sandovabarbora/czefootball-player-pool-atlas, data snapshot commit ef7dbd4. Text, figures and data are CC BY 4.0; the code is MIT.

Change log

  • 1 October 2026
    • The Showcase cards page no longer repeats the pool: the static cluster maps (the page's interactive atlas shows the same projections), the historical-analogs list (each card carries its analogs) and the player index (the pool list on the page and the Players page) were removed, about 7 000 words. Old links to them open the Players page.
    • Correction: each edition's Big-5 count applied the 450-minute floor to each club stint, not to a player's minutes summed across the five leagues as the method states. A player who moved between two Big-5 clubs during a season with fewer than 450 minutes at each was left out (England 2025/26: Ethan Nwaneri, 171 minutes at Arsenal and 324 at Marseille). The editions now count as the Nations page does and agree with it for every country and season: England 2025/26 is 128 (was 127), Spain's 1995/96 peak 324 (was 323), and Denmark, Croatia, Poland, Switzerland, France, Italy, Germany and others move by 1–3 players in some seasons (design/count-and-lono-protocol.md).
    • The series models were refitted on the corrected counts under the same convergence rule. Denmark's fit now fails it (one divergent transition after both sampler stages), so the Denmark edition and the Denmark rows of the Czech and Norwegian editions date no step; the pages say so. Every other dated season is unchanged: Czechia 2014/15 ×0.59 (90 % HDI 0.37–1.02), Germany 2000/01 ×0.75 (0.65–0.89). Backtest errors move by at most 0.35 players; the naive baseline's for Denmark is now 3.0 (was 2.7), Spain's 8.9 (was 8.6), Germany's 8.3 (was 8.6).
    • The leave-one-nation-out check of the export-age model now runs at the main fit's settings (four chains, 1000 draws) and meets the convergence rule in every edition (minimum bulk ESS was 177–324); the shift it reports moves by at most 0.0005. The page states its diagnostics.
    • The front page's header now carries the per-head figure and one sentence (the rank). The three other measures it listed are in the findings just below it; the counts at 450 and 900 minutes are on the Methodology page.
    • Each edition's own series model (the Big-5 page, its forecast and backtest, and the first finding on the Nations page) was refitted under a convergence rule written down on 1 October 2026, before the refit (design/edition-refit-protocol.md): four chains, 2000 tuning steps and 2000 draws, target acceptance 0.99 only where 0.95 failed; max R-hat at most 1.01, bulk and tail ESS at least 400, no divergent transitions. At the earlier settings (two chains, 500 draws) no edition's fit met it. Now every fit passes: the six home fits, the peers' fits, the forecast fits and every backtest origin.
    • What moved: England's fall, dated 2007/08 at ×0.95 (90 % HDI 0.77–1.17), is now ×0.96 (0.76–1.18), within 5 % of no change, so the England edition dates no fall (best candidate 2019/20, posterior 10 %). Germany's fall stays at 2000/01, ×0.75 (90 % HDI 0.64–0.90; was ×0.76, 0.65–0.98), posterior 75 % (was 69 %). Czechia's fall stays at 2014/15, ×0.59 (0.39–1.02; was ×0.60, 0.39–1.03), and its rise at 2000/01, ×1.67 (0.70–2.32; was ×1.65, 0.87–2.46). Every other dated season is unchanged. In Norway's backtest the model's mean error is now 3.4 players, the same as the naive baseline (was 3.2), so the model no longer beats it. The live forecasts move by at most two players; the registered forecasts are unchanged.
    • The Big-5 page of an edition that dates no step (England, Spain) no longer calls its best candidate season a rise or a fall in What this means.
    • The cross-edition change-point fits on the Nations page were refitted under a convergence rule written down before the refit (design/nations-refit-protocol.md): four chains, 2000 tuning steps and 2000 draws, target acceptance 0.95 and 0.99 only where needed; max R-hat at most 1.01, bulk and tail ESS at least 400, no divergent transitions. 15 of 17 fits pass and the page now lists every fit's diagnostics. Denmark and Switzerland fail (one divergent transition each) and are no longer dated.
    • Dated steps that moved in the refit: Poland's fall from 2014/15 to 2019/20 (×1.33, 90 % HDI 0.88–1.82) and its other step from 1998/99 to 2014/15; Italy's other step from 2010/11 to 2009/10. Every other country keeps its seasons with small changes in size. Czechia's fall stays at 2014/15, ×0.59, but its 90 % HDI widens from 0.38–0.97 to 0.37–1.02 and now includes no change; its rise stays at 2000/01, ×1.67 (0.74–2.40). The edition's own Big-5 fit, used in the findings, is unchanged.
  • 30 September 2026
    • The Big-5 chart caption now describes only the dated rules the chart draws in each edition; for Spain, where no step is drawn, it no longer mentions a rise and a fall.
  • 29 September 2026
    • Checked against editorial standard v1 on 29 September 2026. Pages now carry a research kicker, a metadata block (status: exploratory, not pre-registered; data snapshot; code; cite as; licence), a question-and-contribution paragraph with five external references, How to cite and this change log; every question page ends with What this means, What this does not show and a metadata line.
    • Advice drawn from descriptive data was removed: the “A federation tracking this would watch” line on every question, the “For a federation” and “What this is for” blocks, the targets and the “numbers to watch” in the findings, and the heuristic sentence under the two reforms, now retitled “Youth minutes in Germany and England after two academy reforms (descriptive)” with Honigstein (2015) and Premier League (2011) cited inline.
    • Correction: the per-head count was described as players with at least 450 minutes; it counts every player who appeared (any minutes). The number is unchanged and the page now gives the count at 450 and 900 minutes beside it (new table per_capita_floors.json, src/international_benchmark.py).
    • Correction: the findings quoted the Big-5 steps from the cross-edition fit (outputs/nations/nations.json, long_run); they now quote the edition’s own change-point fit, the one its Big-5 page shows, with the posterior and the 90 % HDI of each step. The claims that a peer’s presence “came back” (resting on posteriors of 23–31 %) were removed.
    • Correction: each Big-5 step is now dated to the season where its own posterior peaks; before, it was the season the step takes in the most probable pair of seasons, which for Spain was 2001/02 (posterior 2 %) where 2021/22 has 7 %. The findings follow the Big-5 page in not dating a step within 5 % of no change, and report an earlier fall as the other step. The method text now says the series starts in 1995/96, not 1990/91, and that the model has two steps.
    • Cross-edition page: every dated step in the long-run chart and table now carries its 90 % HDI (the long run was refitted with the same seed; only the dated seasons of Austria, Belgium, Italy, the Netherlands, Portugal and Spain moved, to their own most probable season). The lines the page generated in the browser, which grouped countries into those that “fell and came back” and asked whether the youth share is “a cause”, were removed; the decomposition bars show their 90 % bootstrap intervals.
    • Correction: the gap decomposition’s league-strength coefficient is negative in the fit; the text now says so (a weaker home league goes with more players per million) instead of reading league strength as an advantage. The export-age channel is reported as zero by construction where the countries compared share the same median export age; the two export-age measures are named apart (age at the first season abroad; age at the first top-9 season).
    • Correction: the recent youth-minutes finding compared two endpoints; it now reports an OLS slope over the six seasons with its 90 % CI, and the slope at the upper bound of the last season where the share is a floor.
    • Stepping-stone leagues are now disjoint from the top nine (config/leagues.yaml: only the German second tier). Every tier rule already checked the top nine first, so no count changed; pathways.json, pool_table.json and season_changes.json were recomputed and are identical.
    • New intervals: 90 % percentile-bootstrap intervals on each country’s median minutes share abroad (2 000 resamples, seed 42; pathways.json); interval types are labelled throughout (90 % HDI, 90 % bootstrap interval, 90 % CI).
    • Exhibit C on the pathway page now shows the minutes-share bars and the club goals-scored percentile bars under their own labels; before, only the goals percentile was drawn under a paragraph that led with the minutes share.
    • League and club headshots, which carry no licence the atlas can cite, are no longer shown; players without a Wikimedia Commons portrait show initials.
    • Methodology: data availability with sources, snapshot dates and licence notes; reproduction commands, snapshot commit and seeds; all limitations in one list; references with access dates, DOIs and the data sources (FBref, soccerdata, Wikipedia, Wikidata, Eurostat demo_pjan, UEFA). The review and ruling counts and the build-flow section were replaced by a line on the automated tests; the tracking-data demonstration was removed.
    • Reproduction: the rebuild command now includes make figures, which refits the seeded models from the snapshot and redraws the figures; without it a clean clone stopped at the missing figures.
    • Cross-edition page: the decomposition section is retitled from “why a nation lags” to a description of the split, “mechanism” wording is replaced by “channel”, and each contrast is labelled as a gap in players per million.
  • 29 September 2026
    • Per-head count, same value, corrected label: 262 players, 3.13 per million, rank 8 of 8 with any minutes; added: 191 (2.28, rank 8) at 450 minutes and 155 (1.85, rank 8) at 900 minutes.
    • Big-5 steps in the findings: 2000/01 ×0.75 and 2010/11 ×1.03 (cross-edition fit) became 2000/01 ×0.76 (90 % HDI 0.65–0.98, posterior 69 %) from the edition’s own fit; the 2010/11 step (×1.04) is within 5 % of no change and is not dated.
    • Youth-minutes finding: “The trend runs the right way: Germany's own under-21s went from 3 % to 6 % of home-league minutes in 6 seasons.” became “Germany’s own under-21 nationals’ share of home-league minutes over 6 seasons: a trend of +0.50 pp a season (90 % CI +0.08 to +0.92).”