Marie Andreescu, Manuel Arias-Osorio, Luis Bauluz, Nitin Kumar Bharti, Philipp Bothe, Pierre Brassac, Lucas Chancel, Mauricio De Rosa et al. · World Inequality Lab Working Paper 2025/25, October 2025

Equality and Development: A Comparative and Historical Perspective 1800-2025

18 problems found: 5 major · 5 moderate · 8 minor

What the paper claims

Strong positive long-run association between equality (post-tax B50/T10 average-income ratio) and productivity (hourly NDP): more equal countries are more productive, and today's rich countries became more equal as they became more productive.
Very large 20th-century compression of the post-tax income scale in Western and especially Nordic Europe (SE, DK, NO, NL), about two-thirds of it through pretax inequality.
The world's richest countries went through the largest inequality compression; Latin America, Sub-Saharan Africa, South/South-East Asia and MENA had little or none and have sluggish productivity.
Nordic Europe became more productive than the US around 1970 and keeps a 'significant productivity gap'; the US has been on par with Western Europe since 1980, despite much higher US total (public+private) human-capital spending; this supports a positive residual effect of equality.
The post-1980 rebound of inequality, especially in the US, was accompanied by slower, not faster, productivity growth.
Regressions show a positive, 'highly significant' residual association of equality with productivity levels (controlling for HC spending) and growth (with country FE and HC); 'we can rule out the possibility of a significant negative residual effect'.
Proposed interpretation: inclusive 'social-democratic' institutions (human capital, public services, labour rights, progressive taxation, democratic participation) raised both equality and productivity.
Wealth: limited compression; top-10% wealth share about 85% (1910) -> 55-60% in Western Europe (about 55% Nordic), middle 40% up to about 40-45%, bottom 50% nearly propertyless; the seven core countries' 'fully independent' sources show similar trends.

Every problem found

1majorconfirmed · newly identified here · methodology

The paper's statistical case that equality itself raises productivity, beyond education and health spending, comes from comparing whole world regions: rich, equal Europe against poorer, unequal regions. Comparing countries within the same region, or the same country over time, the link vanishes or reverses. The data show that equality and prosperity can coexist, but they cannot show that equality caused the prosperity, and they cannot 'rule out' that equality had some cost.

Technical detail and evidence

Why it matters, in detail. Identification comes from between-region contrasts (Europe/NAOC vs the rest) and, in growth form, from 1910-79 windows that coincide with Europe's post-war reconstruction and compression. Region x decade FE take the growth coefficient from +0.041 to -0.009 (p=0.73), and within-country level estimates are significantly negative (-60 to -89; p=0.01-0.06). Conventional SEs understate uncertainty about 5-6 fold. The evidence supports 'large compression coexisted with high productivity', not a positive residual effect, and it cannot exclude a negative one.

Evidence. Adjudicator re-estimation translating doreg.txt (scripts/robustness.py, robustness2.py, robustness3.py, exclude_copies.py, weighted_and_sweden.py; out/*.txt) on Equality2025Appendix.xlsx (A2b, B1a) and Bharti et al. (2025) appendix (C4a, C4b, A4a, C5a). Levels 1990-2025: col.1 148.9 with OLS se 3.65 vs country-clustered se 19.0; col.6 (logs, HC + education) 0.347, clustered p=0.15. Region FE: 8.0 (p=0.79); +HC -4.6 (p=0.87); +HC+edu -12.9 (p=0.66). Country FE: -88.6 (p=0.014); +HC -60.1 (p=0.057). Country+year FE: -67.3 (p=0.016); +HC+edu -23.7 (p=0.50). Growth 1830-2025 (col.3 baseline 0.041, clustered p=0.028): windows ending 1830-1909 -1.36, 1910-1979 +0.160, 1980-2025 -0.057 (p=0.41), with HC -0.065 (p=0.36); country FE + region x decade FE -0.009 (p=0.73), with HC -0.008 (p=0.77); 15 countries with non-copied histories, FE+HC 0.033 (p=0.16), +year FE 0.013 (p=0.57); population-weighted FE+HC 0.010 (p=0.83). Survives: +year FE 0.051 with HC (p<0.001) / 0.028 without (p=0.065); initial log productivity + country and year FE 0.069 (p<0.001); non-overlapping windows without HC 0.050-0.073.

Why this severity. The residual association is the paper's main argument beyond the simple correlation. In the published data it disappears within regions, reverses within countries and is absent after 1980, so a core claim rests on cross-region contrasts that cannot separate equality from regional or historical confounders. The descriptive association (C1) is not affected.

The authors' response. The paper says the regressions are 'for illustrative purposes only' and 'inherently fragile' (p.37) and that the residual association 'remains fragile' (p.39), but still states the 'rule out' conclusion (pp.4, 38). No response to these tests found.

2majorconfirmed · newly identified here · replication failure

When the paper's regression code is re-run on the data the authors published, the numbers printed in the paper do not come out. Most importantly, the estimated boost to productivity growth from more equality is less than half as large as reported. The authors also ran, but did not show, a version they themselves called 'very unstable'.

Technical detail and evidence

Why it matters, in detail. The printed tables are not what the released inputs imply. The controlled growth coefficient is 0.041 (clustered se 0.019) rather than 0.090 (se 0.004), and the HC coefficient is 0.052 rather than 0.008, so HC is far from irrelevant. The log-level coefficient with HC and education is 0.347 rather than 0.197. An unfavourable specification the authors ran is omitted. The tables cannot be regenerated by outsiders.

Evidence. scripts/reproduce_tables.py -> out/reproduce_tables.txt (N = 1,728 and 9,408 match; A2b = 1/A2a = (B50/0.5)/(T10/0.1) to 1e-16). doreg.txt lines 121-135; 'sy1050' occurs only in doreg.txt. Tables 3-4 are numerically identical in the 20 Oct draft (gabriel-zucman.eu TNEF container), the 27 Oct replication-zip PDF, the 31 Oct Dropbox PDF and the WID PDF. TN 2025/10 pp.4-6 (revision rounds April and September 2025).

Why this severity. The paper's only quantitative statement of the growth 'residual association' (about +0.9 pp/yr per +10 pp of equality, with HC) becomes about +0.4 pp/yr in the released data, a change of more than half in a core number. The Table 3 level coefficients move the other way (larger), so the discrepancy is not one-directional, and the sign survives.

The authors' response. None found; the paper labels the regressions illustrative (p.37).

3majorconfirmed · newly identified here · methodology

The US spends much more than Europe on education and health, but mostly on private health care. The paper treats all of it as investment in workers' skills, so the US 'should' be more productive, and equality gets the credit when Europe does as well. Counting only public spending, which the authors themselves say matters more, removes most of the extra effect attributed to equality.

Technical detail and evidence

Why it matters, in detail. With total HC as the control, high-private-spending, unequal countries (US, Latin America) look under-productive given their HC, and the shortfall loads onto equality. Controlling for public HC, the component the authors argue matters, leaves an equality coefficient of 34-70 that is imprecise. The paper's 'little effect' statement is contradicted by its own data, and part of the remaining residual is mechanical because post-tax equality contains public in-kind transfers.

Evidence. scripts/robustness.py, hc_variants.py, pwt_and_overlap.py -> out/robustness.txt, hc_variants.txt, pwt_and_overlap.txt; Bharti et al. (2025) appendix A4a (public, age-adjusted), C5a (private), C4a = A4a + C5a exactly; main appendix B2a-B2d (2023: total HC US 22.8%, Nordic 14.8%, Western 15.4%; health 15.4% vs 9.0%). Paper pp.22-24, 36.

Why this severity. Under a defensible alternative control, one the authors say they ran and which follows from their own argument that private HC spending is less productive (pp.23-24), the residual level association shrinks by 53-71% and is not significant with clustered errors. The headline residual claim therefore rests substantially on treating US private health spending as productive human-capital input.

The authors' response. The paper argues private HC spending has a lower aggregate impact than public spending (pp.23-24, fn 47) and reports that the public/private split has 'little effect on the equality coefficient' (p.36).

4majorconfirmed · newly identified here · external inconsistency

The paper's figures make Scandinavia and the Netherlands look 20-40% more productive per hour than the US. Official OECD statistics show only Norway (oil) and Denmark ahead, with Sweden, Germany and the Netherlands slightly behind the US. Most of the difference comes from the paper counting fewer hours worked in Europe than the OECD does, plus an error in Sweden's employment figures around 1990.

Technical detail and evidence

Why it matters, in detail. European levels relative to the US are biased up about 4-9% by hours per head. LFS 15-64 employment, rather than national-accounts employment, is inconsistent with a GDP numerator that covers all workers. Single-benchmark PPP backcasting further raises 1970-90 European levels relative to current-PPP benchmarks. In OECD data US GDP per hour exceeds Western Europe's by 3-12% and roughly equals Nordic-ex-Norway, while the US spends about 8 pp of GDP more on HC (mostly private health). The 'spends more but is not more productive' argument survives only in weakened form.

Evidence. scripts/productivity_benchmarks.py, hours_decomposition.py, employment_check.py, weighted_and_sweden.py -> out/*.txt. OECD Productivity Database DF_PDB_LV (GDPHRS, USD PPP current; 2023 components incl. HRSPOP, GDPPOP; vintage May 2026); PWT 10.01; ILOSTAT DF_EMP_DWAP_SEX_AGE_RT; FRED/OECD LREM64TT*. doprod.txt lines 37-48 (2024 PPP applied to all years) and 74-89 (hours from Andreescu et al. 2025, sheet B8). Andreescu et al. (2025) appendix sheet B2 (employment rates 15-64).

Why this severity. The Nordic productivity advantage that anchors the residual-equality argument shrinks from about 20-25% (paper, 2019-23) to about 7% in OECD data, and that 7% is due to Norway (oil) and Denmark. Three of the five countries named as above the US are below it. The group-level direction survives, but the magnitude of a core supporting claim falls by well over a quarter.

The authors' response. fn 41 (p.21) asserts that recent hours match OECD/BLS; fn 55 (p.27) argues comparisons must account for hours and price levels. No response to these specific discrepancies found.

5majorconfirmed · newly identified here · selective presentation

To show that US productivity growth slowed as inequality rose, the authors' spreadsheet averaged the World War II years into the 1950 starting point, for the US only and without saying so. Without that adjustment, US productivity grew at roughly the same pace (about 1.8-2.1% a year) in every period. So the US record shows no growth boost from rising inequality, but no clear slowdown either.

Technical detail and evidence

Why it matters, in detail. A smoothing of one endpoint, applied only to the US, generates the equality-growth co-movement shown in Fig. 34. Unsmoothed, US hourly productivity growth is flat at 1.8-2.1% in all periods. The post-1990 shortfall (-0.10 pp; -0.15 pp with symmetric 5-year endpoint means) is break-year dependent (+0.03 pp with a 1980 break). The US time series is uninformative about the sign of the equality-growth relation.

Evidence. scripts/us_period_growth.py -> out/us_period_growth.txt; out/c1_growth_formulas.txt (sheet C1 rows 11-15; all other columns use plain endpoints); B1a column V. Paper p.27; Fig. 34 (p.86); captions of Figs 32-35 (pp.84-87).

Why this severity. The specific claim that the inequality rebound came with slower US growth shrinks by 62% with plain endpoints and disappears with a 1980 break. The claim that the high-equality 1950-90 period out-grew earlier periods is reversed. The weaker statement that rising inequality was not accompanied by faster growth survives.

The authors' response. The paper notes that 1950-1990 'was unusual for all sorts of reasons, including post-WW2 reconstruction' and that world/European comparisons are 'very difficult to interpret' (p.27); the US endpoint averaging is not disclosed.

6moderateconfirmed · newly identified here · internal inconsistency

The paper's most quoted number says the gap between a top-1% and a bottom-10% income in Nordic countries shrank from about 50-to-1 to under 5-to-1. Its own data say it started near 30-to-1, so the shrinkage was about 8-fold, not more than 10-fold. Other measures do show a very large shrinkage, so the basic story stands, but the headline is exaggerated.

Technical detail and evidence

Why it matters, in detail. Compression of the post-tax income scale to 1990 is 6-17 fold depending on the ratio (T10/B50 6.2x, P99/P10 8.4x, T1/B50 13.6x, P99.9/P10 16.5x), and 3-8 fold to 2025. The repeated '50 -> 5' is not in the data. About 40-55% of the log-compression comes from the bottom (B50, P10) rising, which is weakly observed in 1910 (see F7).

Evidence. scripts/headline_ratios.py -> out/headline_ratios.txt (sheets A3b, A3d, A3e, A4b, A2a, A1a-A1c, A6, D1a-D1b). Paper pp.2-3, 17, 20, 38; Figs 14 and 16 (pp.66, 68).

Why this severity. The enormous Nordic compression is real in the data, and 'a factor of more than 10' holds for P99.9/P10 to 1990. One of the two headline ratios is overstated by about 65% at its starting point, and its compression factor is 8.4 (to 1990) or 5 (to 2025) rather than more than 10. That is a noticeable correction, not a transformative one.

7moderateconfirmed · newly identified here · misrepresentation of source

Before 1980, the paper's charts show Denmark, Norway and the Netherlands as separate countries with their own histories, but their numbers are identical copies of a generic 'rest of Western Europe' profile; the same is true for many poorer countries. The paper presents them as independent sources that happen to agree, when they agree because they were copied. Sweden's genuine data still show the big drop in inequality.

Technical detail and evidence

Why it matters, in detail. For 1820-1970 the 'Nordic Europe' aggregate is three-quarters the OC profile, whose 1820-1910 top-10% shares are hand-set targets. The 1910 bottom shares (B50, P10), which drive 40-55% of the measured log-compression, are unobserved in tax data and partly set by unsourced targets. On the 15 countries with own histories, the HC-controlled growth coefficient is 0.033 (p=0.16; 0.013, p=0.57 with year FE). The claim that the main long-run facts 'do not depend on specific assumptions and data sources' (p.9) is not demonstrated.

Evidence. scripts/copy_footprint.py, wid_compare.py, exclude_copies.py, cross_sections.py -> out/*.txt; WIL TN 2025/10 p.3 and Annexes 1-2 (pp.19-21); WID bulk 2026-10-15 (sptincj992, sfiinct992 p90p100). Paper pp.9, 14, 31.

Why this severity. The 'independent corroboration' claim is false for three of the four Nordic countries, and most pre-1980 cross-country variation in the long-run panel is imputed. But Sweden's own series, and those of Germany, France and Britain, show the same European compression, so C2 survives qualitatively. Dropping copied countries mainly costs precision (Table 4 FE+HC 0.033 vs 0.041). C3's historical part for developing regions rests on a few regional profiles.

The authors' response. The paper calls pre-1900 series 'a strong limitation' (p.9) and its series 'imperfect and provisional' (p.8); TN 2025/10 later documents the residual-region attribution and the target values.

8moderateconfirmed · newly identified here · overclaiming

The paper says the countries that became most equal also grew fastest, and that equal countries have long been the most productive. Its own numbers show the fastest-growing economies (South Korea, Taiwan, China, the Gulf states) mostly did not become more equal. The link between equality and productivity across countries only became strong after about 1990.

Technical detail and evidence

Why it matters, in detail. In these data the cross-sectional equality-productivity gradient is a post-1990 phenomenon. Twentieth-century catch-up growth was fastest in East Asian and resource economies without compression. The long-run evidence is essentially the co-movement in a handful of European countries plus Japan and Korea.

Evidence. scripts/communist_and_misc.py, cross_sections.py -> out/communist_and_misc.txt, cross_sections.txt; doreg.txt lines 32-39; paper pp.3, 21, 35, 84-87.

Why this severity. These statements go beyond the authors' own data. They do not change the 2025 cross-section or the European compression, but they overstate the generality of the 'long-run' equality-prosperity link the abstract advertises.

The authors' response. The paper repeatedly hedges that mechanisms are hard to disentangle (pp.3-4, 18, 37-39) and reports the null 1910 association (p.35).

9moderatelikely · newly identified here · misrepresentation of source

For wealth, the four 'Nordic' countries in the paper share one identical history, taken from an older file rather than the database the paper cites. The current version of that database shows a smaller drop in Nordic wealth concentration than the paper reports.

Technical detail and evidence

Why it matters, in detail. There is one Nordic wealth trajectory, not four independent ones. Against WID's current series the 1910-1990 Nordic top-10% fall is overstated by about 50% (34 vs 22 pp), and the Nordic-vs-Western ranking of compression reverses. Imputing wealth from income shares for missing country-years (TN sec. 2.2) also builds part of the income-wealth correlation into the data.

Evidence. scripts/copy_footprint.py, wid_compare.py -> out/copy_footprint.txt, wid_compare.txt; doequality.txt line 15 and the commented wealth-export block; WID bulk 2026-10-15 shwealj992; TN 2025/10 sec. 2.2 (wealth imputed from top income shares for missing country-years).

Why this severity. The qualitative wealth findings survive in both sources: falling top share, rising middle 40%, tiny bottom 50%. But the Nordic magnitude, and the claim that Nordic compression exceeded Western Europe's, rest on an undocumented single series that diverges from WID's own.

The authors' response. The code comment flags the wealth series as provisional ('for now'); this is not disclosed in the paper.

10moderatelikely · previously known · external inconsistency

The finding that more equal countries are more productive relies on the World Inequality Database's estimates for poor countries, which add large amounts of missing top income to household surveys. With the World Bank's survey figures instead, countries such as India or Ethiopia look about as equal as parts of Europe, and the overall link becomes much weaker. It still holds among countries whose surveys measure income in the same way.

Technical detail and evidence

Why it matters, in detail. WID equality for developing countries is compressed near very high inequality (B50/T10 about 0.03-0.10), so the 2025 fit is a rich-vs-rest contrast. Survey decile shares give B50/T10 about 0.12-0.34 for Asian and African consumption surveys. The gradient is robust among income-survey countries (slope 156-183) but absent among consumption-survey countries. The headline R² is measurement-dependent.

Evidence. scripts/survey_crosscheck.py -> out/survey_crosscheck.txt; World Bank PIP survey-level file (api.worldbank.org/pip/v1/pip?country=all&year=all&fill_gaps=false&format=csv, downloaded 2026-10-25 to /mnt/zdata/downloads/wdi-adjudicator-andreescu/pip_surveys.csv). Paper p.15, Fig. 44.

Why this severity. The association survives qualitatively among countries with comparable income-based surveys, so C1 is not overturned. But its strength (R² about 0.6-0.7), and the paper's picture of uniformly very unequal developing regions, depend on WID corrections the paper treats as settled. Survey measures have their own biases, notably consumption vs income, so the true gradient lies within a wide range.

The authors' response. Paper p.15: 'Available data sources for these regions are more fragile ..., but this general finding and the corresponding orders of magnitude appear to be robust.' WID authors' general position (DINA Guidelines) is that surveys miss top incomes.

11minorconfirmed · newly identified here · arithmetic or transcription

Some numbers in one table were copied from the wrong columns. It is a typo that does not change the conclusions.

Technical detail and evidence

Why it matters, in detail. Transcription error in the HC and education rows of the log columns; no bearing on the equality estimates.

Evidence. Paper Table 3 (p.102); Equality2025MainFigures.xlsx sheet T3 (typed values, per Auditor A); scripts/reproduce_tables.py.

Why this severity. Presentation error; the equality coefficients and conclusions are unaffected.

12minorconfirmed · newly identified here · internal inconsistency

Labels and points on some charts do not match the underlying numbers, and one chart pairs the wrong countries' data, but fixing them does not change what the charts are meant to show.

Technical detail and evidence

Why it matters, in detail. With correct data: 2025 OLS R² 0.63 (48 countries), 1910 R² 0.007. The mis-pairing affects only the plotted 1910 points.

Evidence. scripts/fig44_46_check.py -> out/fig44_46_check.txt (Equality2025MainFigures.xlsx chart46-48.xml literal arrays; sheet C2 formulas); scripts/communist_and_misc.py (1910 ranks).

Why this severity. Display and annotation errors. The contrasts the figures are meant to show (strong 2025 association, none in 1910) survive correct alignment.

13minorconfirmed · newly identified here · arithmetic or transcription

In one chart the three wealth groups add up to more than 100% because Sweden was left out of one group. Correcting it makes the poorest half look even poorer, which supports the paper's point.

Technical detail and evidence

Why it matters, in detail. Inconsistent country sets across groups; the Nordic bottom-50% wealth share is overstated by 4.2 pp (3.6x) in 2025.

Evidence. out/e1b_group_formulas.txt; Equality2025MainFigures.xlsx chart39.xml literals; paper Fig. 37 (p.92) and p.32 ('around 5-10% or less').

Why this severity. Internal inconsistency in one figure; correcting it strengthens the paper's 'propertyless bottom half' conclusion.

The authors' response. Fn 64 (p.31) discusses Sweden's negative bottom-50% wealth share but not its omission from one group.

14minorconfirmed · newly identified here · data error

In a few historical years the data say impossible things, such as the poorest tenth earning more than the median. These are isolated glitches that do not drive the main results.

Technical detail and evidence

Why it matters, in detail. Shares and thresholds imputed separately are not jointly consistent in 23 country-years. Percentile-ratio statistics for DE 1946-51 and FR 1890-1905 should not be used.

Evidence. scripts/feasibility.py -> out/feasibility.txt (A3a, A3b, A1e, A1b, D3b).

Why this severity. Isolated country-years; the threshold ratios for these cells are unreliable, but the share-based equality index used in the core results is essentially unaffected.

The authors' response. The code comments that additive interpolation is used for the lower tail 'otherwise this leads to inconsistencies' (doequality.txt line 229).

15minorconfirmed · newly identified here · internal inconsistency

The paper says Nordic countries were more equal than the Soviet Union even at the very top; its own data show the opposite for 1950-1970, although the authors note that Soviet elite perks are poorly measured.

Technical detail and evidence

Why it matters, in detail. The top-end comparison favouring social democracy over Soviet communism is not in the data for matched years; it holds only for T10/B50 and for China after 1970.

Evidence. scripts/communist_and_misc.py -> out/communist_and_misc.txt (A1b, A1c, A2a).

Why this severity. Secondary argument about the communist counter-example; the authors caveat that Soviet elite in-kind incomes are under-measured.

The authors' response. Sec. 4.3 (p.25): the USSR/Maoist estimates 'probably underestimate the extent of in-kind incomes available for the elite'; fn 49 adds that in-kind incomes are taken into account 'only in a very imperfect and limited manner'.

16minorconfirmed · newly identified here · arithmetic or transcription

Several small factual mistakes. The most consequential calls the US 'oil-rich' to explain why it is very productive despite being unequal, but oil, gas and mining are only about 1-2% of the US economy.

Technical detail and evidence

Why it matters, in detail. The US lies about 35 €/h above the Fig. 44 fit. With mining at 1-2.5% of GDP rather than 6-8%, it is a genuine high-inequality, high-productivity counter-example to a tight equality-productivity link among rich countries, not an oil artefact.

Evidence. FRED VAPGDPM / VAPGDPU (BEA value added by industry, % of GDP; /mnt/zdata/downloads/fred-claude-andreescu/); scripts/communist_and_misc.py, headline_ratios.py; paper pp.7-8, 11, 14, 17, 34-35 and references.

Why this severity. No effect on estimates; the fossil-fuel claim matters only for how the US outlier in Fig. 44 is interpreted.

17minorconfirmed · newly identified here · documentation gap

The data and code released with the paper are incomplete, and some series the paper says are public are not in the database it points to, so outsiders cannot fully check the results.

Technical detail and evidence

Why it matters, in detail. Not reproducible end-to-end. The US pre-1950 post-tax override lowers measured US equality by about 5-15% in 1913-1940. The '2025' observations are projections from 2023 data.

Evidence. replication/codes/*.txt (grep sy1050; dodirectory.txt; doprod.txt; doequality.txt rows 225-226); WID bulk US sdiincj992 (2026); TN 2025/10; /mnt/zdata/downloads/andreescu2025/paper-zucman.pdf (file type TNEF).

Why this severity. These gaps block exact replication (their main consequence is F2) and verification of the pre-1980 series; by themselves they do not change conclusions.

18minorconfirmed · newly identified here · other

The paper says equality matters four times less than education spending, but that comparison mixes up units. Measured fairly, the regression gives equality the bigger role, which makes its shakiness more important.

Technical detail and evidence

Why it matters, in detail. Standardized effects reverse the paper's ranking: the specification attributes more productivity variation to the equality index than to education spending.

Evidence. scripts/standardized.py -> out/standardized.txt.

Why this severity. Interpretive slip that understates rather than overstates the weight the regression gives equality; it makes the fragility documented in F1 and F3 more consequential.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

The pre-1980 post-tax series, constructed by linear interpolation of redistribution 1910-1980, 'drive the growth-regression result' (Table 4 association concentrated in the 1910-1980 window where post-tax equality is assumed). — Not supported. With pretax equality, which does not use the post-tax assumption, Table 4 col.3 is 0.11 (clustered p=0.002), larger than with post-tax equality. Auditor B's alternative redistribution timings (sqrt(s), s^2) keep the sign (0.046 / 0.032). The concentration in 1910-79 windows reflects where the variation is (European compression and post-war growth), which F1 covers. The US PSZ override is retained as a documentation point in F17.
PWT 10.01 contradicts the paper's statement that Nordic Europe overtook the US 'around 1970' (with PWT, only around 2000). — Overstated. The 2000 date comes from PWT current-PPP levels (cgdpo). Applying the paper's own single-benchmark method to PWT data (national-accounts growth, rgdpna, anchored to 2019 levels) gives a crossing in 1976, so historical level rankings are method-dependent rather than contradicted. The discrepancies in recent levels, hours and the Swedish employment input are upheld in F4.
The very strong income-wealth correlation invoked in fn 62 ('holds separately in every subperiod') is partly mechanical because missing wealth country-years are imputed from income. — Not demonstrated. TN 2025/10 (sec. 2.2, Figs 1-4) estimates the income-wealth relationship on observed country-years before imputing, and fn 62 may refer to those observations. The imputation is noted as a caveat in F8.
The US-Nordic human-capital spending gap 'almost 10 points of GDP' (p.23) is overstated (7.4-8.0, peak 8.9). — Immaterial. In the authors' data the US-Nordic total HC gap peaks at 9.2 pp (2020) and the US-Western gap at 9.2 pp (2011); 'almost 10 points in recent years' is a fair rounding. The relevant problem, the composition of the gap, is in F3.
Data-vintage drift (2026 WID revisions to post-1980 Nordic pretax series, e.g. NO B50 1990 0.309 -> 0.284) is a flaw of the paper. — Post-publication revisions of a living database are not errors in the audited version. Noted under limitations; the substantive vintage problem (tables not matching the paper's own released data) is F2.
The PDF listed at gabriel-zucman.eu/files/AndreescuEtal2025.pdf 'returned HTML, not PDF'. — Factually wrong (materials list, not a claim about the paper). The file is a TNEF (winmail.dat) e-mail container (file signature 0x223E9F78) embedding a 20 October 2025 draft, as Auditor A found; it is recorded in F17.
The Nordic P99/P10 misstatement is a major issue; the Fig. 37 adding-up error and the distribution-ordering violations are moderate issues. — Severity not upheld (the issues themselves are confirmed). P99/P10 is graded moderate because the 'factor of more than 10' holds for P99.9/P10 and the compression remains 6-17 fold depending on the ratio (F6). Fig. 37 is minor because the correction strengthens the paper's claim (F13). The ordering violations are minor because they are isolated and do not enter the share-based equality index materially (F14).
Copied regional profiles (I1) are a major issue for the core findings. — Severity not upheld (the issue itself is confirmed, F7). Sweden's own series and those of DE, FR and GB show the same European compression, and dropping copied countries changes the regression coefficients little (precision falls), so the issue is graded moderate as an important undisclosed caveat.

Audit notes

What reproduced. Reproduced (adjudicator's own Python translation of doreg.txt and of the sheet formulas): the equality index and sample sizes; all descriptive shares and ratios in Secs 3-5, with the discrepancies noted in F6, F13 and F14; the 2025 cross-section (R² 0.63); the C1 period-growth figures, including the US-only endpoint substitution; the chart data arrays of Figs 37 and 44-46; and the copy pattern across territories, also present in WID's own 2026 bulk files. Not reproduced: all nine equality coefficients of Tables 3-4. Three independent reconstructions agree with one another: Table 3 148.9 / 117.0 / 94.5, logs 0.993 / 0.660 / 0.347; Table 4 0.066 / 0.066 / 0.041 with HC coefficient 0.052. The printed values are 144.2 / 104.9 / 84.2; 0.926 / 0.481 / 0.197; 0.086 / 0.094 / 0.090; HC 0.008. Also not reproduced: the Fig. 44-46 annotations; the 'P99/P10 about 50 -> under 5, factor > 10' headline; the stated 'little effect' of splitting public and private HC; and fn 41's claim that hours match OECD. The original Stata pipeline is not runnable as released (sy1050 is never created, private Dropbox paths, no estimation .dta or logs). External benchmarks: OECD 2023 GDP per hour reverses the paper's ranking of DE, NL and SE vs the US. World Bank PIP survey deciles give a much weaker cross-country association (R² 0.13 vs 0.71 with WID equality, same 45 countries).

Limits of this audit. No Stata was available and the authors' estimation datasets (equality.dta, reghumank.dta, humankprodreg.dta) and logs are not released. Reproductions use the released Oct-2025 workbook and the July-2025 Bharti et al. HC workbook, which may differ from the inputs actually used for Tables 3-4. The WID vintage of October 2025 was not archived, so comparisons with WID series use the 2026-10-15 bulk files. The pre-1980 income copies are identical in both, but the wealth comparison (F8) may partly reflect later revisions. The OECD productivity data are a May-2026 vintage, and historical cross-country productivity levels are inherently method-dependent (PPP benchmark, hours concept). The PIP survey cross-check (F9) mixes consumption and income surveys and is a sensitivity analysis, not a preferred alternative series. The sources of the TN target values could not be traced, the construction of Bharti et al.'s HC series and Andreescu et al.'s hours outside a few OECD countries was not audited, and Nievas & Piketty's historical national accounts were not checked. The region x decade fixed-effects test is demanding and removes much of the historical variation; it is reported as one of several diagnostics. No author response to these findings was sought. Page numbers refer to printed pages of the WID PDF (sha256 c2604953...; PDF page = printed + 1).