Where The Schooling Went

Essay · economics · October 2026

Since 2013 Portugal has had the fastest improvement in the education of its workforce among the countries compared, yet output per hour has barely grown. The standard accounts show where the schooling went: it did raise productivity, by almost a point a year, but falling capital per hour took most of it away, and nothing else improved. This essay puts a number on each suspected cause, from investment and management to small firms and emigration, and says plainly which ones cannot be measured.

Reading time ≈ 12 min The split · the capital · everything else · management · the level gap · the tally

The question

What Schooling Bought showed that Portugal schooled its workforce faster than any other European country, and calculated that the extra schooling alone should have added about half a point a year to the growth of output per hour since 2013. Output per hour grew by 0.17 per cent a year. The Stalled Hour pointed to weak investment, and The Business Side to small firms and their managers. This essay tries to put numbers on all of them: how much each cause is worth, how sure we can be, and which ones the data cannot settle.

+0.88points a year added by a better-educated workforce, 2013–2019

−0.62points a year lost as capital per hour fell

−0.08points a year from everything else; Germany +0.74

5–10%of Portugal’s efficiency gap with Europe explained by management practices

The split

The schooling showed up

Economists split the growth of output per hour into three parts. The first is the workforce: if more workers have skills that firms pay more for, each hour produces more. The second is capital: more machines, buildings and software per hour worked. The third is a residual, called total factor productivity, which collects everything else: technology, organisation, management, how well resources move to the best firms, and every measurement error. The European KLEMS project does this split for every EU country, using official accounts and the pay of workers by education, age and sex.[1]

For Portugal from 2013 to 2019 the result is clear (Figure 1). The better-educated workforce added 0.88 points a year to the growth of output per hour, the largest contribution of any country compared: more than three times Spain’s or Italy’s and twelve times Germany’s. About four-fifths of it was education; the rest came from age and other changes in the workforce. But capital per hour subtracted 0.62 points a year, and everything else subtracted 0.08. Output per hour grew by 0.18 per cent a year.[1][2]

Fig. 1 — Contributions to the annual growth of value added per hour worked, total economy, 2013–2019 (mean of annual values 2014–2019), percentage points: change in the composition of the workforce (education, age, sex, weighted by relative pay), capital per hour (tangible, ICT and intangible), and the residual (total factor productivity). Data: EU KLEMS & INTANProd, 2024 release, statistical module.

So the schooling was not wasted, at least by this measure. It was cancelled out. If capital per hour had merely held steady, Portuguese output per hour would have grown about 0.8 per cent a year, close to the EU average for the period; if everything else had improved at Germany’s pace, about 0.8 points more. In accounting terms, roughly half of the gap with a German-style performance came from capital and half from the residual.[1]

The numbers depend on the years chosen, and the residual most of all (Figure 2). In 2014, the last year of the bailout, output per hour fell 1.5 per cent and the residual 2.4 per cent; starting the count in 2014 instead gives a residual of +0.39 a year. Including the pandemic years 2020 and 2021, when low-paid jobs disappeared and the residual collapsed, gives −0.69. Across all the windows, two things hold: the workforce added between 0.7 and 1.2 points a year, and capital per hour subtracted.[1]

Fig. 2 — Portugal: cumulative contributions to the growth of value added per hour since 2013, percentage points (sum of annual contributions), 2013–2019. Data: EU KLEMS & INTANProd, 2024 release, statistical module.

The KLEMS accounts end in 2021. A rougher calculation of my own, with the same method applied to Eurostat and European Commission data to 2025, gives the same picture for 2013–2025: education +0.72 points a year, capital −0.59, everything else +0.04 (Figure 3). Since 2019 the fall in capital per hour has eased, to −0.35 a year, but the residual has turned slightly negative.[3] This also corrects a common reading. The OECD reports that Portuguese “multifactor productivity” grew by about 0.4 points a year in 2014–2019, but its measure counts hours worked without adjusting for skills, so the schooling is inside that number; taking it out leaves a residual close to zero or below.[4]

Fig. 3 — Contributions to the annual growth of GDP per hour worked, 2013–2025, percentage points; a rough extension: education composition only (three levels, weighted by Portuguese relative pay for every country), net capital stock rather than capital services, residual by difference. Data: Eurostat (nama_10_lp_ulc, lfsa_egaed); European Commission, AMECO; EU KLEMS pay weights; author’s calculation.

The capital

More hours, not more machines

Capital per hour fell because investment did not keep pace with work. Investment fell to 14.8 per cent of GDP in 2013 and averaged 16.6 per cent in 2014–2019, against 20.5 per cent in the EU, while total hours worked grew about 2 per cent a year as unemployment fell and, later, immigrants arrived.[5] Net private investment, after replacing what wears out, was recently 0.6 per cent of GDP, against 3.8 per cent in the EU, according to the European Commission.[6] The Banco de Portugal’s summary is that capital per worker has stagnated since the sovereign crisis “because investment did not keep pace with employment”, and the European Commission and the government’s own economists say the same.[7][6]

Why firms invested so little is less certain. Many were still paying down the debts of the 2000s, which fell from 120 to 65 per cent of GDP; bank credit was tight in the first years after the crisis; and low-skilled labour was easy to hire, which makes machines less pressing. That last reason is a hypothesis, consistent with the data but not tested here.

Everything else

What is inside the residual

The residual is where better technology, organisation and management should show up, and for Portugal it has been about zero since 2013, against 0.7 in Germany and 1.7 in Czechia. It cannot be measured directly; its parts can only be estimated one at a time, and the estimates overlap.

Moving between industries: about zero. Hours moved into hotels, restaurants and other low-productivity services, but the exit of workers from agriculture, the least productive sector, offset it; the net effect on productivity from 2013 to 2023 was about nil.[8]

Resources stuck in weak firms: large in 2013–2015, unknown since. A Banco de Portugal study of all firms found that, in 2013–2015, the way workers were spread among surviving firms lowered labour productivity growth by about 1.1 points a year and efficiency by about 0.55; its series ends in 2015, and no later one exists.[9] “Zombie” firms, those that could not cover their interest for three years, fell from 14,600 in 2014 to 7,800 in 2019 and held about 5 per cent of jobs; they weigh on the level of productivity, but they cannot explain a worsening after 2014.[10]

Small firms: about a sixth of the gap in level, no worsening. Giving Portugal the EU’s mix of firm sizes would close about 15 per cent of the gap in output per worker, or about a quarter counting the mix of sectors too. The share of jobs in micro firms has not risen; but the distance between the typical Portuguese firm and the best European ones widened from 2010 to 2020 in every size class.[11]

Management

Measurable in level, not in time

The best international measure of management comes from the World Management Survey, which scored manufacturing firms in 35 countries on monitoring, targets and the management of people, through structured interviews with their managers. Portuguese firms averaged 2.83 on a scale of 1 to 5, fifteenth of 35: below Germany (3.21), France and Italy, above Spain and Greece (Figure 4). Portugal scored lowest on targets and on people management, such as promoting and rewarding on merit, and better on monitoring.[12][13]

Fig. 4 — Average management score of manufacturing firms with 50 to 5,000 employees, World Management Survey waves 2004–2014, scale 1 to 5; selected countries of 35. Data: N. Bloom, R. Sadun and J. Van Reenen, “Management as a Technology?”, NBER Working Paper 22327, Fig. 2.

The survey’s authors take as their benchmark, drawing on their own and others’ studies, that a firm one standard deviation better managed is about 10 per cent more productive. On that basis, management explains 27 per cent of Portugal’s efficiency gap with the United States.[12] Applying the same rule to Europe, Portuguese firms managed like the average of nine European countries in the survey would be about 2 per cent more efficient, and managed like German firms about 6 per cent: roughly 5 to 10 per cent of Portugal’s efficiency gap with its European peers. Using the raw association between management and sales per worker instead gives up to about a quarter, but that overstates the effect, because better firms also tend to be better managed for other reasons.[13]

Ownership is part of the story. In Portugal, as in Italy and Greece, about 60 per cent of the firms surveyed were run by their founders, by families with a family chief executive, or by the state, against 20 to 30 per cent in Germany, Sweden or the United States, and such firms are on average worse managed.[13] INE’s own survey of management practices in 2016 found that 56.5 per cent of Portuguese firms with five or more people are majority-owned by founders or their families, and that firms with better practices produce much more per worker, even comparing firms of the same size, capital and staff education.[14] The schooling of those who run firms matters too: a study of every Portuguese firm found productivity about 5 per cent higher for each extra year of the founder’s schooling. Portuguese employers have about 1.8 years less schooling than the European average, which on that estimate would be worth about 9 per cent in the firms they run; a rough figure, since the estimate comes from comparisons within Portugal.[15][16]

What cannot be measured is whether management explains the slowdown. There is no Portuguese management series over time in published form, the international survey covers only manufacturing firms with 50 or more employees, which leaves out the small firms where most Portuguese work, and the one indicator that is tracked, the schooling of employers, has improved by about 1.4 years since 2011. On the evidence, poor management is part of why Portugal is less productive than its neighbours; there is no evidence that it got worse after 2013. INE repeated its survey in 2022, and its microdata could, in principle, answer the question.[14]

The workforce

Was the schooling overcounted?

The workforce term weights each worker by pay, so it assumes graduates earn more because they produce more. If part of the graduate premium was scarcity or credentials, the contribution of schooling is smaller than 0.88 and the residual correspondingly less negative; the two trade off one for one, and their sum, about 0.8 points a year, is robust. The graduate premium fell as graduates became common, from 1.8 times the average pay in 2008 to 1.6 in 2020, which the accounts already allow for.[2]

Two other effects are small. At the same level of education, Portuguese adults score about 10 points below the OECD average in literacy, worth perhaps 3 to 4 per cent of the quality of the workforce.[17] And if the Portuguese-born graduates living in other OECD countries in 2015/16, about 224,000, or just the 146,000 added since 2000/01, had stayed and worked like those who did, output per hour would be about 2 to 3 per cent higher, roughly 0.1 to 0.2 points a year over fifteen years; this is my own calculation, not a published estimate. Emigrants as a whole are less educated than those who stayed, so emigration has not lowered the average education of those working in Portugal.[18]

The level gap

Plenty of capital, too little efficiency

The slowdown and the gap are different questions. The same accounting can be done in levels, comparing Portugal with other countries in a single year (Figure 5). In 2019 an hour of Portuguese work produced about 56 per cent of the average of the five largest EU economies. Capital is not the reason for that gap: Portugal has more capital per unit of output than its peers, much of it buildings and infrastructure from the boom before 2008. Measured on a logarithmic scale, lower efficiency accounts for nearly nine-tenths of the gap and fewer years of schooling for about half, while the extra capital per unit of output offsets about a third.[19]

Fig. 5 — Portugal’s output per hour relative to each comparator in 2019, and its three factors: capital per unit of output (raised to the power 1/2), human capital from years of schooling, and efficiency (total factor productivity); the three multiply to output per hour. Five largest EU: Germany, France, Italy, Spain and the Netherlands. Capital share of one-third. Data: Penn World Table 10.0; author’s calculation following Hall and Jones (1999).

Put together, the two accounts tell one story. Before 2008 Portugal invested heavily, but much of it went into buildings and sectors with low returns; after 2013 it invested too little to equip a growing workforce. Throughout, the efficiency with which capital and labour are combined, where management, firm size and the movement of resources to better firms sit, stayed well below its neighbours’.

The tally

What each cause is worth

Growth effects are in points a year of output per hour since 2013; level effects are shares of the gap with Europe.

1. Falling capital per hour: about −0.6 points a year, easing to −0.35 after 2019. Measured directly; high confidence.
2. Better-educated workforce: +0.7 to +0.9 points a year, the largest in the countries compared. Measured, assuming pay reflects productivity; medium-high confidence.
3. Everything else (the residual): about zero, against +0.7 in Germany. Measured only as what is left over; medium confidence, sensitive to the years chosen.
4. Hours moving between industries: about zero. High confidence.
5. Resources stuck in weak firms: −1.1 points a year for labour productivity in 2013–2015; unknown since. Low confidence after 2015.
6. Management practices: about 5 to 10 per cent of the efficiency gap with Europe in level (27 per cent of the gap with the United States); its role in the slowdown cannot be measured.
7. Schooling of employers: perhaps 9 per cent in the firms they run, in level; improving, so not a cause of the slowdown. Low to medium confidence.
8. Small firms: about 15 per cent of the gap in level, a quarter with the sector mix; not worsening. Medium confidence.
9. Graduate emigration: about 2 to 3 per cent of the level, 0.1 to 0.2 points a year. A rough counterfactual; low confidence.
10. Weaker skills at the same schooling: about 3 to 4 per cent of workforce quality. Low to medium confidence.

These do not add up to a whole, and should not be read as if they did. They are measured in different units, over different periods, and they overlap: badly managed firms also invest less and stay small, so management shows up in the capital and size numbers as well as its own.

The balance

The schooling arrived; the investment and the efficiency did not

The answer to the question is more encouraging than it first looks. The schooling did raise Portuguese productivity, by more than in any of the countries compared. It was cancelled by a decade in which firms added workers faster than equipment, and by a residual that did not grow at all. The first problem can be measured precisely and has begun to ease; the second is where management, small firms and the slow movement of resources to better firms sit, and it can be described in level but not yet broken down over time.

For policy, that points to two things the schooling cannot do alone: getting investment back above what wears out, and raising the efficiency of firms, which includes, but is not only, how they are managed. Both depend more on firms, owners and credit than on schools. And the one gap that would most help to answer the question is a measurement gap: a regular, published series on how Portuguese firms are managed, which INE’s survey could provide.

On method and tools

This piece was written collaboratively with Claude Opus 5.5 (Anthropic): human specification, editorial direction and critical review; machine data research, analysis and drafting. The figures are computed by scripts/schooling_went.py. Growth accounting uses EU KLEMS & INTANProd (2024 release, statistical module, total economy), which weights types of workers by their relative pay and assets by their user cost; Portuguese labour accounts start in 2008, so no comparable split exists for 2000–2008. The extension to 2025 is the author’s, with net capital stocks, three education levels and one set of pay weights. The level comparison follows Hall and Jones (1999) with Penn World Table 10.0 (version 11.0 could not be downloaded). Management conversions apply Bloom, Sadun and Van Reenen’s rule of 10 per cent of productivity per standard deviation of management to their size-weighted country scores. The emigration counterfactual adds graduate emigrants to 2016 employment at Portuguese relative pay. Results are in docs/schooling-went-results.json; the downloaded sources, with page or table for each number, are kept with the script’s data.

The cover photograph is Contentores, Terminal XXI by Nuno Morão; CC BY-SA 2.0, via Wikimedia Commons, cropped.

Authored by: Luis Matos Ferreira — Physicist, Developer, Writer

Related essays on this blog

1. What Schooling Bought — the education boom and what it paid.
2. The Stalled Hour — why output per hour stopped growing.
3. The Business Side — what firms face, and how much it explains.
4. The Leavers — who emigrates, and why.
5. Work, Time and Money — the reading guide to the whole series.

Sources

1. EU KLEMS & INTANProd, 2024 release (Luiss Lab of European Economics), growth accounts, statistical module, total economy; F. Bontadini, C. Corrado, J. Haskel, M. Iommi and C. Jona-Lasinio, “EUKLEMS & INTANProd: industry productivity accounts with intangibles”, 2023; Eurostat (nama_10_lp_ulc).
2. EU KLEMS & INTANProd, 2024 release, Portuguese labour accounts by education, age and sex; author’s split of the composition term.
3. Eurostat, real labour productivity per hour (nama_10_lp_ulc) and employment by education (lfsa_egaed); European Commission, AMECO (net capital stock, total hours); author’s calculation.
4. OECD, Productivity Statistics: Methodological note, May 2021, p. 4; OECD Productivity Database, multifactor productivity, Portugal.
5. European Commission, AMECO, gross fixed capital formation and total hours worked; Eurostat, investment (nama_10_gdp).
6. European Commission, 2024 Country Report: Portugal, SWD(2024) 622, p. 58; 2026 Country Report: Portugal, SWD(2026) 222, p. 67.
7. Banco de Portugal, Boletim Económico, March 2025, Caixas 3–4, pp. 22–27; GPEARI, note on labour productivity, c. 2023, pp. 1–3.
8. Eurostat, national accounts by industry (nama_10_a64, nama_10_a64_e); author’s shift-share decomposition, as in The Stalled Hour.
9. S. Dias and C. R. Marques, “Every cloud has a silver lining: micro-level evidence on the cleansing effects of the Portuguese financial crisis”, Banco de Portugal Working Paper 18/2018, Table 2, p. 17; Oxford Bulletin of Economics and Statistics 83, 2021.
10. P. Alves, J. Tavares and A. Osório de Barros, “Revisitar as Empresas Zombie em Portugal (2008–2021)”, GEE Paper 178, October 2023, pp. 14–16.
11. Eurostat, structural business statistics by size class (sbs_sc_ovw), 2023; IMF, Portugal: 2024 Article IV Consultation, Annex VI, pp. 66–68; J. Amador and G. Nogueira, Banco de Portugal, Revista de Estudos Económicos, October 2025.
12. N. Bloom, R. Sadun and J. Van Reenen, “Management as a Technology?”, NBER Working Paper 22327, 2017, pp. 24–25, Fig. 2 (p. 32) and Table 7 (p. 47).
13. N. Bloom and J. Van Reenen, “Why Do Management Practices Differ across Firms and Countries?”, Journal of Economic Perspectives 24(1), 2010, pp. 8–16; author’s conversions from Bloom, Sadun and Van Reenen (2017), Table 7.
14. INE, Inquérito às Práticas de Gestão 2016, Destaque, 22 November 2017, pp. 1–3 and Tabela 1, p. 15; DREM, note on the 2022 survey, p. 4.
15. F. Queiró, “Entrepreneurial Human Capital and Firm Dynamics”, GEE Paper 116, 2018, p. 7; Review of Economic Studies 89(4), 2022.
16. Eurostat, self-employed with employees by education (lfsa_esgaed), 2011–2024; author’s years of schooling.
17. OECD, Survey of Adult Skills 2023, Table A.2.5, as presented by Pessoas 2030 and ANQEP, November 2025, p. 20; E. Hanushek, G. Schwerdt, S. Wiederhold and L. Woessmann, “Returns to skills around the world”, European Economic Review 73, 2015.
18. OECD, Connecting with Emigrants, 2015, p. 239; OECD Social, Employment and Migration Working Paper 239, 2020, Table A.2; author’s counterfactual with KLEMS pay weights.
19. R. Feenstra, R. Inklaar and M. Timmer, “The Next Generation of the Penn World Table”, American Economic Review 105(10), 2015 (PWT 10.0); R. Hall and C. Jones, “Why Do Some Countries Produce So Much More Output per Worker than Others?”, Quarterly Journal of Economics 114(1), 1999; author’s calculation.