Facundo Alvaredo, Lucas Chancel, Thomas Piketty, Emmanuel Saez, Gabriel Zucman · Economie et Statistique / Economics and Statistics 517-518-519: 41-59 (special issue 'Beyond GDP')

Towards a System of Distributional National Accounts: Methods and Global Inequality Estimates from WID.world

18 problems found: 2 major · 6 moderate · 10 minor

What the paper claims

DINA distributes total national income and national wealth among all individual residents (adults 20+, couples equal-split), so that inequality statistics and growth by group are consistent with national-accounts aggregates.
Top income shares rose in nearly all countries since 1980 but by very different amounts (moderate in Europe; large in North America, India, China and Russia), suggesting country-specific institutions and policies matter; the US-Canada bottom-50% share collapsed while the top-1% share nearly doubled.
Within China, Europe, India, Russia and US-Canada, real pre-tax income growth per adult over 1980-2016 was systematically higher for upper income groups.
Private wealth-income ratios rose in nearly all countries while public wealth declined, becoming negative in several rich countries.
Top wealth shares rose sharply in the USA and China and more moderately in France and the UK.
Global income inequality has increased since 1980 despite the catching-up of China and India, driven by income growth at the global top; the global top 1% captured about twice as much of total growth as the bottom 50%; neither post-2000 emerging-country growth nor the 2008 crisis stopped the rise.
Conditional projections to 2050 under three within-country inequality scenarios with common between-country growth.

Every problem found

1majorconfirmed · newly identified here · overclaiming

The paper's headline says world inequality went up after 1980. In the authors' own numbers that is true only if you look at the share going to the very richest. By the most common overall yardstick (the Gini coefficient) world inequality was flat, and the poorer half of the world increased its share of income by about a fifth. So 'the top pulled away' is supported, but 'inequality rose' as a general statement is not.

Technical detail and evidence

Why it matters, in detail. The 1980 and 2016 global Lorenz curves cross (bottom-50% ordinate +1.75 pp; bottom-99% ordinate -4.3 pp), so no index-free ordering exists. In the WIR2018 world file the Gini change is -0.0006, T10/B50 falls 13% (30.9 -> 26.9) and the MLD falls 0.026, while the top-1% share rises 4.3 pp and the Theil index 0.15. In the authors' dynamic-calibration variant, under the corrections in F2/F3, under China's official deflator (F6) and in the current WID vintage, the Gini falls by 0.012-0.018. What the data support is 'the global top-1% and top-10% shares rose and the top 1% captured about twice the bottom half's share of per-adult growth', not 'global inequality increased'.

Evidence. Authors' file Results/PPP/Scenario 4 (Static calibration)/gpinterized.dta (iso WO; WIR2018 computer-codes.zip), recomputed in scripts/world_indices.py (work/world_indices_S4static.txt): Gini 0.6523 (1980) vs 0.6517 (2016); T10/B50 30.93 -> 26.95; MLD 0.925 -> 0.899; Theil 0.918 -> 1.066; bottom 50% 7.92% -> 9.67%; middle 40% 43.10% -> 38.21%; top 10% 48.98% -> 52.12%; top 1% 16.16% -> 20.44%. Display data (WIR2018 C2.1.xlsx, F2.1.7): bottom 50% 7.98% (1980), 9.21% (1985), 9.66% (2016). Authors' dynamic-calibration robustness file: Gini 0.6523 -> 0.6338, T10/B50 30.9 -> 23.6. Current WID (WO, downloaded 2026-10-25): Gini 0.687 -> 0.671, T10/B50 41.0 -> 32.2. Under the construction corrections of F2/F3 the Gini also falls (combined 0.667 -> 0.654; work/sensitivity_table.md). Chancel & Piketty (2021, JEEA 19(6)): global T10/B50 'reached an all-time high of 53 in 1980 and 50 in 2000, before declining to 38 in 2020' (p. 3034); Gini '0.72 in 2000 and 0.67 in 2020' (p. 3035); for 1980-2020 'the effect on synthetic inequality indicators like the global T10/B50 ratio was again ambiguous' (p. 3039).

Why this severity. The abstract-level qualitative claim is not supported under the most common synthetic index (the Gini does not change) and is reversed under T10/B50 and the mean log deviation in the paper's own data; it holds only for top shares and the Theil index. Not critical, because the top-share version that the paper actually displays rises in every variant tested.

The authors' response. No response to this paper located. Co-authors Chancel & Piketty (2021, JEEA 19(6): 3034-3039) later describe the 1980-2020 effect on synthetic indicators as 'ambiguous' and place the global T10/B50 peak in 1980.

2majorconfirmed · newly identified here · data error

Hidden in the authors' computer code is a line that, without explanation, assumes Middle Eastern incomes in the early 1980s were a third lower than in 1985, when the authors' own data say they were about a fifth higher (the oil boom was ending). Because that region had many very rich people in 1980, this one line makes the world's richest look poorer in 1980 and so makes their later gains look bigger. Fixing it removes about a third of the reported rise in the top 1%'s share of world income, though the share still rises.

Technical detail and evidence

Why it matters, in detail. The rule lowers MENA's 1980 mean by 45% while MENA's 1980s distribution is a 1990 backcast with a top-1% share of 30.7%, so the 1980 global top is mechanically depleted. This raises measured top-group growth and the top-share trend: change in T1 share +4.28 -> +3.01 pp (-30%); change in T10 -28%; top-1% growth -21 pp; top-0.001% growth -51 pp; world mean growth -5.8 pp. Capture shares barely move (T1 27.7% -> 26.1%; B50 12.6% -> 13.0%). Even a neutral backcast (1985 level) removes 18% of the T1-share rise.

Evidence. Code: 'gen aa=anninc if iso=="ME" & year==1985 ... replace anninc=annincME85/1.5 if iso =="ME" & year <1985'. An exact re-implementation of 0c on the authors' inputs (Data/Imported/national-incomes-ppp.dta, populations.dta, regions-codes.dta) reproduces regions-aggregates.dta for every region and year except ME 1980-84 (scripts/mena_uncorrected_exact.py; work/mena_uncorrected_exact.txt): uncorrected 20,566 (1980), 19,906, 18,349, 17,651, 17,454 (1984), 16,998 (1985); authors' file 11,332 for 1980-84. Country ratios 1980/1985: Saudi Arabia 1.96, Kuwait 1.58, Iraq 1.68, UAE 1.38 (work/mena_coverage.txt). Data/Added/middle-east.dta: totinc 8,225 constant 1980-85; shares constant 1980-89 (top 10% 65.11%). Global effect from a pooled rebuild that reproduces the authors' merged world file (1980: T1 16.16%, T10 48.98%, B50 7.92%, Gini 0.6524 vs 0.6523; 2016 identical; work/rebuild_baseline.txt), MENA uncorrected (work/rebuild_mena.txt): 1980 T1 16.16% -> 17.43%; rise of T1 share 4.28 -> 3.01 pp; rise of T10 3.14 -> 2.26 pp; 1980-2016 growth: top 1% 101% -> 80%, top 0.1% 133% -> 94%, top 0.01% 184% -> 140%, top 0.001% 233% -> 182%, world mean 58.9% -> 53.2%, bottom 50% 94% -> 88%; Gini 0.657 -> 0.652; growth capture T1 27.7% -> 26.1%, B50 12.6% -> 13.0%. Holding 1985 level (ratio 1.0): T1 rise 3.53 pp, top-0.001% growth 202%, mean 55.3%. Printed World column 60/94/43/70/101/133/185/235 vs with-rule 59/94/41/69/101/133/184/234 and without-rule 53/88/38/60/80/94/140/182. Chancel & Gethin (2017) Table B1: 'Middle-East ... Full population 89%'.

Why this severity. Using the authors' own national-accounts inputs removes about 30% of the headline 1980-2016 rise in the global top-1% share and 20-30% of the featured growth rates of global top groups (101% -> 80%; 235% -> 182%). The direction of the top-share rise and the 2:1 capture ratio survive.

3moderateconfirmed · newly identified here · internal inconsistency

The paper says that neither the rise of emerging economies after 2000 nor the 2008 financial crisis stopped global inequality from rising. Its own numbers show the opposite: by every main measure, global inequality was lower in 2016 than in 2007, and by most measures lower than in 2000.

Technical detail and evidence

Why it matters, in detail. Over 2007-2016: top-1% share -1.68 pp, top-10% -2.50 pp, bottom-50% +0.56 pp, Gini -0.018. Over 2000-2016: top-1% -0.46 pp, top-10% -3.16 pp, bottom-50% +0.98 pp, Gini -0.025. In these data the 1980-2016 increase in top shares occurs entirely in 1980-2000/2007.

Evidence. WIR2018 C2.1.xlsx F2.1.7: top 1% 20.89% (2000), 22.12% (2007), 21.79% (2008), 20.44% (2016); bottom 50% 8.68%, 9.09%, 9.20%, 9.66%. F2.1.8: top 10% 55.26% (2000), 55.40% (2004 peak), 54.60% (2007), 52.10% (2016). Authors' world file (work/world_indices_S4static.txt): Gini 0.6767 (2000), 0.6773 (2003), 0.6694 (2007), 0.6517 (2016); Theil 1.117 (2000) -> 1.066 (2016); top 0.1% 8.59% (2000), 9.74% (2007), 8.87% (2016). Current WID: top 1% 20.27% (2000) -> 19.79% (2016); Gini 0.709 -> 0.671. Chancel & Piketty (2021, p. 3035): 'most of the global inequality decline took place after the 2008 financial crisis'.

Why this severity. A secondary but explicitly featured claim is reversed by the paper's own data. It does not affect the 1980-2016 endpoint comparison.

The authors' response. No response to this paper located; co-authors Chancel & Piketty (2021, p. 3035) later state that most of the global decline occurred after 2008.

4moderateconfirmed · previously known · version instability

The authors' own database now reports much smaller gains for the world's richest over 1980-2016 than this paper does: for example 72% instead of 101% for the top 1%, and 120% instead of 235% for the top 0.001%. It now also shows the poorer half of the world growing faster than the top 1%. The broad picture, that the top gained a disproportionate share of growth, still holds.

Technical detail and evidence

Why it matters, in detail. Relative to the paper: top-group growth -29% (top 1%) to -49% (top 0.001%); world mean growth -13 pp; change in the T1 share -31%; the B50/T1 growth ranking reverses (93% vs 72%). The capture ratio (26.0% vs 12.6%) is stable across vintages, which identifies it as the robust statistic. The near-exact match between the corrected WIR2018 rebuild and the current WID for the 1980-2016 changes in T1, T10 and B50 shares indicates that the revision is largely driven by the MENA treatment and the proxy structure rather than by new top-income data.

Evidence. scripts/current_wid.py on /mnt/zdata/downloads/wir2022/WID_data_WO-live-2026-10-25.csv (work/current_wid_WO.txt): mean anninc992i 18,001 (1980) -> 26,381 (2016); sptinc992j p99p100 0.1689 -> 0.1979, p0p50 0.0632 -> 0.0833; gptinc992j 0.6873 -> 0.6713; growth-incidence points p25 95%, p50 97%, p75 20%, p80 20%, p99 58%, p99.999 120%. China anninc992i (adults) +532% and Russia +57% (WID_data_CN/RU_filtered, bulk 2026-10-15). Paper: Table p. 50; text p. 54; Figure IV.

Why this severity. The qualitative top-share rise and the capture ratio survive, but the global growth figures the paper features are 30-50% lower in the authors' current series. Graded moderate rather than major because the paper flags its series as provisional and because most of the global revision is already counted under F2/F3.

The authors' response. Paper p. 45 ('our methods and time-series are imperfect, fragile and subject to revision') and p. 52 ('imperfect, provisional, and subject to revision').

5moderateconfirmed · newly identified here · documentation gap

China's spectacular growth figure (831%) depends on the authors adjusting China's official inflation numbers downward to its growth. Using China's official statistics the figure would be about 1,286%, and the authors' current database says 532%. Readers aren't told this. It doesn't change who gained most within China, but it changes how much the world's poorer half is estimated to have gained.

Technical detail and evidence

Why it matters, in detail. The growth factor of every Chinese group is proportional to the aggregate real-growth factor: x9.31 (WID/PYZ) vs x13.86 (official) vs x6.32 (current WID). At the global level, China in 1980 sits in the global bottom half, so a lower 1980 Chinese income (official path) lowers the 1980 global B50 share by 0.93 pp, raises B50 growth by 28 pp and the B50 capture share from 12.6% to 14.1%, and makes the global Gini fall by 0.012. This reinforces F1.

Evidence. WDI (cached /mnt/zdata/downloads/worldbank-wdi-dina2020/): China real GDP NY.GDP.MKTP.KN x28.11 1980-2016, adults 20+ x2.03 -> +1,286% per adult; GDP deflator 15.47 -> 91.11 (x5.89). WIR2018 national-incomes-ppp.dta: China 1,501 -> 13,973 PPP EUR (x9.31). PYZ (2019, AER 109(7), p. 2474): 'Young (2003) makes systematic downward corrections to Chinese growth by re-estimating the GDP deflator ... We follow his methodology ... our real GDP growth rate lies in between the official GDP growth rate and the conservative estimate of Maddison and Wu (2008); it is closer to Maddison and Wu's.' Global sensitivity with the validated rebuild, 2016 level fixed: work/rebuild_china_official.txt. Current WID China anninc992i (adults) +532%.

Why this severity. The choice does not alter the ranking of groups within China (all growth multiples scale together) or the global top-1% trend. But featured China and global-bottom growth numbers move by 25-50% across defensible deflator choices, and the choice is not disclosed.

6moderatelikely · previously known · speculative assumption

For about half of the world's 1980 population, and for most of the people at the very top of the 1980 global picture, the authors had no direct data and borrowed other countries' patterns. For example, Japan was treated as if its income were spread like China's and India's, which were very equal in 1980. That makes the world look more equal in 1980 than it was, so the later increase looks bigger. With more realistic choices the increase shrinks by a third or more, though it does not disappear.

Technical detail and evidence

Why it matters, in detail. Ignoring within-block between-country dispersion understates inequality most where that dispersion was largest, i.e. Asia in 1980 (Japan 3.5 times the rest of its block, 2.7 times by 2016), so the bias shrinks over time and inflates the measured rise. The Japan split alone removes 1.0 pp of the 3.1 pp T10 rise and 0.3 pp of the 4.3 pp T1 rise; combined with F2 the T1 rise falls by about a third (to 2.7-2.9 pp). The opposite-signed choices (Eastern Europe, UK backcast) are several times smaller. Footnote 13's robustness claim holds for the sign of the top-1% trend, not for its magnitude or for other indices.

Evidence. Code: '1b - gpinterize-subregions.do' (DE+FR+GB -> EU; IN+CN -> AS), '1c - create-other-subregions.do' (WE, EE = EU; NA = US; SA = BR), '1d - calibrate-subregions.do' (rescale to block means), '1a - prepare-data.do' lines 110-151 (constant-share extension); block membership in regions-codes.dta. Composition from the validated rebuild (work/composition.txt): proxy blocks are 31.6% of world adults and 32.0% of the 1980 global top 1%; proxy plus backcast blocks (BR, ME, GB, AF) are 47.1% of adults, 52.3% of the global top 1% and 65.9% of the top 0.1%. Sensitivities (work/sensitivity_table.md): Japan with its own distribution (current-WID JP g-percentile shape at WIR2018-vintage JP income and adults): 1980 T10 48.98% -> 50.01%, rise of T10 3.14 -> 2.13 pp (-32%), rise of T1 4.28 -> 3.95 pp (-8%), Gini 0.665 -> 0.654 (falls). Japan + MENA fix: T1 rise 2.72 pp (-36%), T10 rise 1.29 pp (-59%). Eastern Europe 1980 with Russia's Soviet-era shape (top 10% 21%): T1 rise +0.19 pp, T10 rise +0.47 pp (opposite sign). Combined (MENA + Japan + Eastern Europe + UK 1980 shape): T1 +2.93 pp, T10 +1.71 pp, B50 +2.06 pp, Gini 0.667 -> 0.654, almost identical to the current WID (+2.90, +1.73, +2.01; 0.687 -> 0.671). Chancel & Gethin (2017, section 2) call constant-share backcasting 'conservative' 'in the context of a general rise in inequality since the 1980s'; that holds for the UK but not for MENA, whose inequality fell over 1980-90 in later WID data.

Why this severity. The proxy choices move the top-10% trend by about a third and flip the sign of the Gini change, but the top-1% share (the paper's displayed headline) still rises under every tested alternative, by 2.7-4.5 pp. The method is disclosed in general terms and the critique is known.

The authors' response. Paper p. 53 and fn 13 acknowledge the simplification and assert robustness. Chancel & Piketty (2021, JEEA 19(6): 3055-3056): corrections 'often rely on a limited set of countries'; robustness checks show general conclusions unaffected, though this 'can certainly affect some of the finer decompositions for the more recent period'.

7moderatelikely · previously known · external inconsistency

The paper's US numbers come from the authors' earlier studies, whose methods other researchers dispute. Those researchers' alternatives still show rising US inequality, but by less (in one leading alternative, about half as much). The authors have accepted one correction, which trims the US top 1%'s income share by about half a percentage point.

Technical detail and evidence

Why it matters, in detail. For the US top-1% pre-tax share, the AS-PSZ gap is about 6 pp in 2014 (roughly 2 pp underreporting allocation, 1 pp retirement income); the conceded pension piece is 0.6 pp in 2016, about 6% of the printed 9.5 pp US-Canada rise. For wealth, the heterogeneous-returns literature (BHH, SZZ) and DFA reconcile partly once units and concepts are aligned; the remaining dispute concentrates in the top 0.1-0.01%.

Evidence. Corpus entries us-income-08, -10, -11, -13, -16, -17, -21; us-wealth-taxes-01, -02, -03, -06, -08, -09 (critic sources and magnitudes as recorded there). PSZ 'US Distributional National Accounts: Updates' (September 2020), sections 1.1.7 ('reduces the top 1% wealth share by 1-1.5 points in recent years') and 1.2.1 ('reduces the top 1% pre-tax income share by about 0.6 point in 2016'), text verified. WIR2018 C4.2.xlsx F4.2.1: US top-1% wealth share 23% (1980), 39% (2012), 37% (2014). Auditor B reports the revised equal-split 2014 value as 36.55% (SZ 2020 JEP data; not re-read here).

Why this severity. The direction of the US rise survives in every alternative. The magnitude of the US top-share increase may be overstated by up to about half under the critics' allocations, which are themselves disputed; the acknowledged pension correction is small (0.6 pp of income share).

The authors' response. Saez & Zucman (2020, NBER WP 27921, 'Trends in US Income and Wealth Inequality: Revising After the Revisionists') defend most allocations. PSZ Updates (September 2020) accept the pension correction and adopt heterogeneous fixed-income yields for wealth. Gale, Sabelhaus & Thorpe (2023) side partly with PSZ.

8moderateplausible · previously known · speculative assumption

For China, India, Russia, the Middle East and Africa, many of the numbers rest on assumptions that other researchers have challenged, such as using one country's tax data to stand in for a whole region or one period for several decades. Rising inequality in these countries is supported by other evidence, but the exact sizes, especially the extreme growth rates printed for the very richest, are much less certain than their precision suggests.

Technical detail and evidence

Why it matters, in detail. Imputations that apply one period's or one country's tax-revealed tail to other years or countries fix the shape of the top in unobserved cells; trends then come from national-accounts growth and from the observed years only. The very-top growth rates are ratios of extrapolated Pareto-tail means, so they are highly sensitive (Russia's top 0.001%: 25,269% vs 18,307% across vintages). Direction of bias is not uniform (formalisation can inflate trends; informality can depress levels).

Evidence. PYZ (2019, AER 109(7), p. 2477 fn 19): 'we choose to apply the average profile of upgrade factors f(p) ... estimated for the 2006-2010 period to all years of the period 1978-2005'. WIR2018 F2.1.1.a: India top-10% share 55.46% constant 2013-2016. Data/Added/middle-east.dta: shares constant 1980-89. Chancel & Gethin (2017), section 2. Corpus: international-global-01 (Ravallion: NA anchoring, no variance estimates), -02 (formalisation bias), -06 (Kharas & Seidel: Lebanon and Cote d'Ivoire donors), -11 (Kapeliushnikov 2020: Russia overstated by about one third), -12 (Hlasny & Verme: Lebanon extrapolation). Current WID Russia: mean growth 1980-2016 +57%, top 0.001% +18,307% (sptincj992 p99.999p100 0.0001 -> 0.0117).

Why this severity. Robustness concern with limited demonstrated impact: the rise in inequality in China, India and Russia is also seen in other sources, but magnitudes, especially at the very top, depend on unvalidated imputations and move by tens of percent across vintages.

The authors' response. Bharti, Chancel, Piketty & Somanchi (2024, WIL WP 2024/09, section 7.5) defend the India trend. Alvaredo, Assouad & Piketty (2019, RIW, fns 10 and 12) defend the Lebanon-based correction and report variants. Novokmet, Piketty & Zucman (2018) report deduction/declaration variants for Russia.

9minorconfirmed · newly identified here · arithmetic or transcription

A formula in the authors' code that was supposed to borrow France's income pattern for the poorer 90% of Germans and Britons was applied upside down. As a result, Britain in 1980 looks much more equal than its own data say, and Germany less equal. Fixing it changes the European figures a little (Europe's rise in inequality becomes slightly smaller) and the world figures hardly at all.

Technical detail and evidence

Why it matters, in detail. The bug inflates the UK's 1980 bottom-90% mass by about 50% (1.2216/0.8143) and deflates Germany's by about 16%. After renormalisation this transfers the distortion to the top decile. Because the UK and German errors largely offset in the DE+FR+GB merge, the Europe aggregate moves modestly (T10 rise -11%, B50 decline -18%), and the world aggregate by under 0.2 pp.

Evidence. In 1a-prepare-data.dta (Scenario 4 static), the ratio of modified DE/GB bracket averages to France's over p5-p50 equals avg90fr/avg90 (DE 1980 0.9165; GB 1980 1.2216), against mass-preserving factors of 1.092 and 0.814 (scripts/degb_scaling_check.py; work/degb_scaling_check.txt). Propagation with the rebuild, whose same-method baseline reproduces the authors' Europe aggregate (T10 33.55% -> 37.80%, T1 10.21% -> 12.52%, B50 21.28% -> 19.16%) (scripts/rebuild_degb.py; work/rebuild_degb.txt): - Europe 1980-2016: change in T10 4.26 -> 3.78 pp; change in B50 -2.11 -> -1.72 pp. - Europe growth: B50 26.3% -> 28.6%, T10 58.0% -> 55.7%, T1 71.8% -> 74.0%, top 0.001% 119.7% -> 139.7%. - World: change in T1 4.28 -> 4.31 pp, change in T10 3.14 -> 2.97 pp, Gini unchanged.

Why this severity. A confirmed coding error, but correcting it slightly strengthens the paper's 'moderate rise in Europe' claim and leaves the global results essentially unchanged. Printed Europe-column growth rates would move by 2-20 pp.

10minorconfirmed · newly identified here · documentation gap

Two of the six columns in the paper's main table (Russia and US-Canada) can't be recreated from the computer files the authors point readers to; those files describe slightly different populations (for instance, Russia plus Ukraine). This makes the numbers hard to check, though nothing suggests they are wrong.

Technical detail and evidence

Why it matters, in detail. The discrepancy reflects different aggregates and income series (Russia vs Russia+Ukraine with an imputed Soviet-average 1980 level; PSZ-type US national income vs WID-2017 PPP series), not arithmetic errors. The world column differs by 1-2 pp (e.g. middle 40% 41% vs 43%), indicating a later run than the archived one.

Evidence. regions-codes.dta: RU block = RU, UA. regions-aggregates.dta: RU 15,655 -> 18,143 PPP EUR (+15.9%); national-incomes-ppp.dta: RU 1980 missing (set to USSR average 17,209), UA -41.7%; US+Other North America +71.0%. wid-final.dta (authors' WID-2017 snapshot): RU anninc992i missing for 1980; US +74.5% (scripts/wid2017_country_growth.py; work/wid2017_country_growth.txt). Chancel & Gethin (2017) Table B1: 'Ex USSR' 16% / -36% / -9%; 'North America' 71% / 8% / 50%; World 59/94/41/69/101/133/184/234. Paper Table: World 60/94/43/70/101/133/185/235.

Why this severity. A reproducibility gap for two of six columns. The direction of the Russian and US results is the same in both constructions, and there is no evidence the printed values are wrong.

11minorconfirmed · newly identified here · documentation gap

The authors' forecasts to 2050 quietly drop the past income losses of some groups (for example poorer Russians and Brazilians) instead of extending past patterns as described. If anything, this makes the forecast look less unequal than the stated method would.

Technical detail and evidence

Why it matters, in detail. Zeroing negative shares removes the offsetting historical losses at the bottom of Russia and Latin America, so projected bottom incomes in those regions rise faster than the stated rule implies; after rescaling to the macro mean, top groups are scaled down slightly. Net effect: a business-as-usual global B50 share in 2050 about 1.2 pp higher than under the signed rule, per Auditor B's reconstruction.

Evidence. Code lines as cited. shares-growth-BAU.dta (Scenario 4 static): negative-share sums by block computed here (RU -1.0267, BR -0.6618, SA -0.3730, AS -0.0809). Magnitudes of the alternative rule from codex scripts/audit_construction.py (results/construction_results.json); not re-run here.

Why this severity. Affects only the illustrative 2050 scenarios, and in the direction of making the business-as-usual scenario less unequal, i.e. against the paper's message.

12minorconfirmed · newly identified here · internal inconsistency

A handful of numbers and labels are wrong or out of date. For example, the paper says Japan's government has negative net wealth when its own chart shows it positive, and it gives China's public share of wealth as 35% when the data say 31%. These are careless errors, but they do not change the paper's conclusions.

Technical detail and evidence

Why it matters, in detail. The Japan statement inverts the sign of a plotted series (Figure II-B); the China public-share and top-1% values are AER P&P 2017-vintage numbers not updated to the WIR2018 data plotted in Figures I-II; the table-period label would imply 776% rather than 831% for China (1980-2015 vs 1980-2016).

Evidence. WIR2018 ExecSumm.xlsx E7 (share of public wealth in national wealth, 2015): Japan +2.63%, US -3.54%, UK -4.01%, Germany +4.11%, France +2.79%, China 31.40% (1978: 69.17%); E6: Japan public wealth +16.2% of national income (2015). WIR2018 C2.1.xlsx F2.1.1.d: China top 1% 13.92% (2015); T2.1.1 title '1980-2016'. Paper pp. 45, 49, 50, 52, 53, 54, Figure V notes. AER P&P 2017 (pp. 405-407) contains the 13% / 35% / Japan statements (per auditor A). WIR2018 C4.2.xlsx F4.2.1 US top-1% wealth 0.39 (2012), 0.37 (2014). French web PDF (Insee), conclusion.

Why this severity. Transcription and labelling errors (1-4 pp) and one sign error (Japan) in secondary statements; no effect on the main conclusions.

13minorconfirmed · previously known · internal inconsistency

The paper says the incomes of the poorer half of Americans and Canadians fell massively; its own table shows they rose slightly (by 5% over 36 years). What fell sharply was their share of total income.

Technical detail and evidence

Why it matters, in detail. Level vs share confusion: real bottom-50% income per adult rose 5% while its share fell 7.4 pp; the policy inference about pre-distribution is not affected by the correction, but the rhetoric overstates.

Evidence. Paper p. 50 text and Table (Bottom 50%, USA-Canada: 5); WIR2018 T2.1.1 (0.05) and F2.1.1.e (19.89% -> 12.54%).

Why this severity. A wording error. The correct meaning (stagnant levels, collapsing share) is recoverable from the surrounding text and the Table.

14minorconfirmed · newly identified here · documentation gap

The paper's one-line definition of 'post-tax income' leaves out government services like defense and police, which the authors' own guidelines include. It's an incomplete description rather than an error in the numbers.

Technical detail and evidence

Why it matters, in detail. As written, the definition would leave collective consumption (and the government balance) unallocated, violating the DINA adding-up property the paper emphasises; the underlying country series (e.g. PSZ) do allocate it, which is itself contested (us-income-13).

Evidence. Paper p. 47; DINA Guidelines (WID.world WP 2016/2, cached text): 'Our post-tax national income series include all in-kind transfers and public spending (using various procedures for imputation to individuals ...), so that aggregate post-tax national income is equal to aggregate national income.' Paper p. 43 itself mentions public spending on education, police and defense.

Why this severity. An incomplete exposition in a methods summary; the paper's empirical results concern pre-tax income and wealth.

15minorconfirmed · newly identified here · overclaiming

The summary says top shares rose almost everywhere, but in the Middle East, Brazil and Africa they did not rise after 1990, and in Britain the richest people's share of wealth was about the same in 2012 as in 1980.

Technical detail and evidence

Why it matters, in detail. The 'nearly all countries' generalisation is drawn from about a dozen DINA series. Among them, regions covering 14-18% of the world's adults show no rise in top-10% shares since 1990, and the UK top wealth shares show none since 1980.

Evidence. WIR2018 C2.1.xlsx F2.1.1.b: top 10% 1990 -> 2016, Middle East 65.86% -> 60.86%, Brazil 58.09% -> 55.33%, Africa 55.27% -> 54.45%. WIR2018 C4.2.xlsx F4.2.1/F4.2.2: UK top 1% 0.19 (1980), 0.16 (1990), 0.20 (2012); top 10% 0.52, 0.46, 0.52. Paper p. 48 (exceptions noted), abstract, p. 52, p. 56.

Why this severity. The general claim holds for the large countries covered; the exceptions are partly disclosed.

16minorconfirmed · newly identified here · misrepresentation of source

The '27% of all income growth' figure is measured per adult. Because the world's adult population doubled, the top 1%'s share of the total increase in world income was about 22%. The comparison with the poorer half (about twice as much) is unaffected.

Technical detail and evidence

Why it matters, in detail. With a growing population, per-adult capture = (s16*m16 - s80*m80)/(m16 - m80) differs from total-income capture = (s16*Y16 - s80*Y80)/(Y16 - Y80): 27.7% vs 22.4% for the top 1%.

Evidence. regions-aggregates.dta: adults 2,394m -> 4,858m (x2.03), total income x3.22, mean x1.589. With the authors' world shares (T1 16.16% -> 20.44%, B50 7.92% -> 9.67%): per-adult capture 27.7% / 12.6%; total-income capture 22.4% / 10.5% (computed here; matches auditor A).

Why this severity. A labelling imprecision. The per-adult concept is the documented convention, and the roughly 2:1 ratio between top 1% and bottom 50% is unchanged.

17minorconfirmed · documentation gap

Some shortcuts are not mentioned in the article: Australia and New Zealand are left out of the 'world', some '2016' numbers are really older numbers carried forward, and every Soviet republic is given the same average income in the 1980s. None is large on its own, but together they make the starting point of the comparison more guesswork than the paper suggests.

Technical detail and evidence

Why it matters, in detail. Excluding Oceania removes a small, rich population from the global top deciles. The Soviet-average imputation compresses 1980 between-republic dispersion (Central Asia vs the Baltics/RSFSR) and biases the measured 1980-2016 rise upward, like the proxies in F3; its size was not quantified here.

Evidence. runwir-program.do: '$countries AF AS BR CN DE EE FR GB IN ME NA RU SA US WE' (no OC); populations.dta for OC countries. WIR2018 F2.1.1.a: India 55.46% (2013-16), US-Canada 46.96% (2014-16), China 41.42% (2015-16), Russia 45.51% (2015-16). 1a-prepare-data.dta: BR top 10% 54.34% for every year 1980-2001; GB 36.90% for 1980-1990; F2.1.1.b Brazil 58.09% (1990). national-incomes-ppp.dta: 1980 national income missing for AM, AZ, BY, EE, KG, KZ, LT, LV, MD, RU, TJ, TM, UZ, filled with the USSR average by '0c - aggregate-regions.do' lines 27-37.

Why this severity. Each step is small or partly documented elsewhere. The Soviet-average imputation and frozen backcasts add to the constructed character of the 1980 baseline (F3) but were not separately quantified.

18minorplausible · previously known · overclaiming

The paper says its results show that government policies, not just technology or globalisation, drive inequality. That may well be true, but the paper doesn't test it; it only notes that countries differ.

Technical detail and evidence

Why it matters, in detail. Cross-country variation in trend magnitude is consistent with institutional explanations but does not identify them; country series differ in data sources, imputations and vintages, which can generate trend differences on their own.

Evidence. Paper p. 48, abstract (p. 41), conclusion (p. 56).

Why this severity. An interpretive claim, hedged in the abstract; no numerical result depends on it.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

The audit found 'no established critical or major findings' (overall assessment, 5 moderate and 4 minor issues). — Auditor B did not detect the undisclosed MENA 1980-84 rule (F2), which on the authors' own inputs removes about 30% of the headline rise in the global top-1% share and 20-30% of top-group growth. B also rated the index-dependence of the abstract's headline claim (F1) as moderate, though the claim fails under the Gini and T10/B50 in the authors' own world file. Both are graded major here.
Every one of the proxy/backcast construction choices understates 1980 inequality and so overstates the 1980-2016 increase (report I3). — Partly incorrect. Giving 1980 communist Eastern Europe (incl. the Baltics, Belarus, Moldova) the Western European DE+FR+GB shape, and holding the UK's 1990 shape back to 1980, bias 1980 inequality upward. Replacing the Eastern European shape with Russia's Soviet-era 1980 shape raises the measured rise of the global top-1% share by 0.19 pp (work/rebuild_counterdirection.txt). The dominant choices (MENA rule, Japan in 'Other Asia') do push in the direction A describes, and the net effect is as A states.
All ex-USSR states are given Russia's distribution (report I3). — Factually inaccurate in detail: in regions-codes.dta only Ukraine shares the Russia block. The Baltics, Belarus and Moldova are in 'Other Eastern Europe' (DE+FR+GB shape), Armenia, Azerbaijan and Georgia in the Middle East block, and the Central Asian republics in 'Other Asia' (China+India shape). Missing 1980-90 incomes of 13 republics are set to the USSR average (retained in F18).
The discrepancies between the printed Table and the archived computation (e.g. Russia 34% vs 15.9%, middle 40% +5% vs -9.4%) reflect a version gap between the display data and the computational archive (report I3). — Partly misdiagnosed. The archived 'Russia' block is Russia plus Ukraine with an imputed USSR-average 1980 income, and 'North America' includes other North American countries, so the printed Russia and US-Canada columns come from different aggregates or country series rather than a later vintage of the same computation. The reproducibility gap itself is retained (F10), graded minor rather than moderate.
India's top-10% share is given as 56% while the plotted series is 55.5% (report I11g). — Immaterial: the text says 'about ... 56%'; 55.46% rounds to 55%, and India's country paper (Chancel & Piketty 2019) reports 56.1%. Not counted as an issue.
The statement that DINA inequality 'is always measured using homogenous observation units' is contradicted by Figure III, which mixes equal-split and individual (UK) series (report I11h). — Not an inconsistency of substance: both are adult-individual series, the DINA framework explicitly provides both equal-split and individualistic adult series (pp. 45-46), and Figure III's notes disclose the UK exception.
Version instability of the global series is a major issue (report I4). — Not rejected but regraded to moderate (F5): revisions in a living database are not errors in the paper, the paper flags its series as provisional, and most of the revision of the global top-share trend is already counted under F2/F3, whose corrections reproduce the current vintage.
The inverted Germany/UK scaling (report I2) and the projection zeroing of negative growth shares (report I4) are moderate issues. — Not rejected but regraded to minor (F9, F11): the DE/GB correction strengthens the paper's 'moderate rise in Europe' and leaves global results unchanged (world top-1% trend +0.03 pp); the projection constraint affects only the illustrative 2050 scenarios and works against the paper's message.

Audit notes

What reproduced. Reproduced: - All 48 Table cells equal WIR2018 Table 2.1.1. - The authors' merged world file gives top-1% shares of 16.16% (1980) and 20.44% (2016) (printed: about 16% / 20.4%), bottom-50% shares of 7.92% / 9.67%, growth capture of 27.7% / 12.6% (printed 27% / 12%), and World-column growth of 59/94/41/69/101/133/184/234% (printed 60/94/43/70/101/133/185/235). - An independent pooled rebuild from the 15 calibrated blocks reproduces that file to within 0.01 pp (Gini 0.6524 vs 0.6523). - A Python re-implementation of the regional aggregation reproduces the authors' aggregates exactly except MENA 1980-84. - The China, Europe and India columns reproduce from the archive. - The 2050 projections match WIR2018 E10 (both auditors). - Wealth ratios and public-wealth shares match WIR2018 E6/E7. - The current-WID figures quoted by Auditor A (world growth 47%, top 1% 72%, top 0.001% 120%, top-1% share 16.9% -> 19.8%, Gini 0.687 -> 0.671, China 532%) and official China growth (+1,286%) were re-computed and confirmed. Not reproduced: - The Russia (34%) and US-Canada (63%) columns cannot be obtained from the cited global programs or the shipped WID-2017 snapshot, whose aggregates are Russia+Ukraine (+15.9%) and North America (+71%). - Small differences between the printed and archived World values indicate a later run than the archived one. - The authors' Stata/R/gpinter pipeline was not executed end to end.

Limits of this audit. No Stata or R was available, so the authors' gpinter merges were approximated by a pooled rebuild with within-bracket interpolation. The rebuild was validated against the authors' merged world file (within 0.01 pp in 1980 and 2016), and sensitivities are reported against this baseline. The Japan split combines current-WID Japanese shares with WIR2018-vintage Japanese income and population. The Eastern-Europe Soviet-shape, UK-1980 and China-official-growth variants are illustrative alternatives, not 'true' values. The Germany/UK correction rebuilds the DE+FR+GB merge by pooling. Auditor B's magnitudes for the projection-rule alternative (F11) were not re-run. Country DINA series (US, France, China, India, Russia, Middle East, Brazil, Africa) were not rebuilt from tax or survey microdata; critics' magnitudes for the US and developing-country series are taken from the corpus entries, with author responses as recorded there. The rationale for the MENA factor of 1.5 could not be established; none was found in the paper, the WIR2018 technical notes or Chancel & Gethin (2017). The Soviet-average imputation for ex-USSR republics was identified but not quantified.