Facundo Alvaredo, Lucas Chancel, Thomas Piketty, Emmanuel Saez, Gabriel Zucman · World Inequality Lab, 2017; Cambridge, MA: Harvard University Press (Belknap), 2018

World Inequality Report 2018

19 problems found: 1 major · 7 moderate · 11 minor

What the paper claims

Global 'elephant curve': since 1980 the global top 1% captured twice as much of world income growth as the poorest 50%; the top 0.1% captured about as much as the bottom 50% ('perhaps the most striking finding'); growth was 'sluggish or even zero' between the global median and the top 1%.
The global top-1% income share rose strongly from 1980 and then declined slightly; the global bottom-50% share 'stagnated'/'oscillated around 9%'.
Income inequality increased in nearly all regions since 1980 at different speeds (so institutions and policies matter); 2016 levels differ greatly and the Middle East is the world's most unequal region.
Diverging trajectories of the US and Western Europe: similar top-1% shares in 1980, very different by 2016; the US bottom-50% share collapsed.
Countries became richer while governments became poorer: private wealth rose relative to national income, public wealth fell (negative in the US and UK); very large public-to-private transfers in China and Russia.
Wealth inequality among individuals rose: very strongly in the US (mostly the top 0.1%) and Russia/China, moderately in France and the UK.
Global wealth inequality (world proxied by China, Europe and the US) rose: the top-1% share increased while the bottom-75% share hovered around 10%; the top 1% captured over a third of wealth growth.
Conditional projections to 2050: under 'business as usual' global income and wealth inequality keep rising; if all countries followed Europe's trajectory the bottom half's income would be about twice that under a US-type trajectory.
Policy-supporting claims: wealth held in tax havens exceeds 10% of world GDP and has risen since the 1970s; tax progressivity is a 'proven' tool that also reduces pre-tax inequality; access to higher education is very unequal.

Every problem found

1majorconfirmed · newly identified here · data error

The authors' computer code quietly replaced the Middle East's average income for 1980-84 with a much lower number, and that number is contradicted by the data they were using. This makes the world's richest 1% look as if they started from a lower point in 1980, so the rise in their share of world income looks about a quarter larger than the data support. The headline 'twice as much' comparison barely changes.

Technical detail and evidence

Why it matters, in detail. Lowering MENA's 1980-84 income per adult (EUR 11,332 instead of 20,566 in 1980) pushes Gulf top earners out of the 1980 global top tail. The 1980 global top-1% share falls by about 1.1 pp, so the 1980-2016 rise is about +3.5 pp rather than +4.5 pp in pooled terms (-24%). World growth is overstated by about 6 pp (59-60% vs 53%). Top-1% capture is overstated by about 1.2 pp and bottom-50% capture understated by 0.4 pp (ratio 2.26 vs 2.10). Statistics from 1985 on are unaffected.

Evidence. Adjudicator re-implementation of 0c (scripts/adj_regions_override.py; output/01) reproduces the shipped regions-aggregates.dta to 6.5e-8. MENA income per adult (PPP 2016 EUR) used 11,332 in 1980-84 vs WID-2017 values 20,566 / 19,906 / 18,349 / 17,651 / 17,454; 1985 = 16,998. Country coverage unchanged 1980-1989 (scripts/adj_mena_coverage.py; output/02); Gulf incomes fall sharply in the data (Saudi Arabia 142,765 -> 72,876; UAE 192,641 -> 140,050 between 1980 and 1985). MENA growth 1980-2016: 88.8% as used vs 4.1% in the data (TN 2017/5 Table B1 prints 89% for the Middle East without explanation). World per-adult growth 1980-2016: 58.9% (code) vs 53.2% without the override; WID-2017's own world aggregate (WO anninc992i, wid-final.dta) gives 52.6%; Table 2.1.1 prints 60%. Pooled re-aggregation of the 15 archived sub-regions (scripts/adj_global_pool.py; output/03) and of the 7 archived regions (scripts/adj_regions_pool.py; output/05): 1980 global top-1% share 15.9% -> 17.0%; 1980-2016 rise +4.5 -> +3.5 pp; capture top 1% 28.1% -> 26.9%, bottom 50% 12.5% -> 12.8%, top 0.1% 14.0% -> 13.2%; ratio 2.26 -> 2.10.

Why this severity. Removing the undocumented override shrinks the 1980-2016 rise of the global top-1% income share, an Executive-Summary headline (C2), by about a quarter (published figures: roughly +4.2 pp becomes +3.1 pp) and cuts world growth by about 6 pp. The C1 capture ratio moves only from about 2.3 to 2.1, so the qualitative conclusions survive. This is borderline major: it reaches the rubric's quarter-of-trend threshold, and the adjustment is undocumented, contrary to the data and in the direction of the narrative.

2moderateconfirmed · newly identified here · methodology

The report's 'global' wealth estimate silently treats about 150 million people in Russia, Ukraine and neighbouring countries as if they were as wealthy as typical people in Britain, France and Spain. Fixing this leaves the main point intact: the richest 1% got a growing slice. It undoes the claim that the poorest three-quarters' share stayed steady at about 10%, because their share was nearer 6% for two decades and then rose.

Technical detail and evidence

Why it matters, in detail. Giving 141-159m ex-USSR adults the FR/UK/ES mixture adds a mass of upper-middle wealth holders. Replacing it with Russian wealth levels and shares raises the top-1% share by about 2.5-3 pp in every year and lowers the bottom-75% share by about 1.5-3 pp: it is about 6% through the late 1990s and rises to about 10% by 2016 (a 60% rise rather than stability). The 1980-2016 top-1% rise shrinks only from +4.7 to +4.2 pp. MER valuation, a disclosed and defensible convention for internationally priced assets, accounts for a further 35-40% of the top-1% rise relative to PPP. The 2050 'business as usual' projection extrapolates these group growth rates and inherits the sensitivity.

Evidence. WID-2017 (wid-final.dta, refetched by the adjudicator): QE adults 514.6m (1980), 590.5m (2016); sum of Western+Eastern Europe+BY/MD/UA+RU = 590.3m in 2016 (scripts/adj_qe_population.py; output/06). Adjudicator emulation of the R code, pooling bracket means instead of gpinter (scripts/adj_global_wealth.py; output/07), reproduces Fig. 4.1.1: top 1% 27.6% -> 32.3% (published 27.9% -> 32.6%); bottom 75% 9.0% -> 11.0% (published 8.7% -> 10.9%). Variants: (a) ex-USSR adults given Russia's own wealth, i.e. private wealth/national income from Fig. 3.5.1 x own national income per adult, with WID-2017 Russian fractile shares (scripts/adj_global_wealth_rublock.py): top 1% 30.5% -> 34.8% (+4.2 pp), bottom 75% 6.0% -> 9.6% (1980-2016 mean 7.1% vs 9.1%); (b) ex-USSR adults dropped: 29.6% -> 34.1% (+4.5 pp), bottom 75% 6.2% -> 10.0%; (c) PPP instead of MER: 26.9% -> 29.7% (+2.8 pp), bottom 75% 9.7% -> 14.3%; (d) PPP with the Russia block: 28.8% -> 31.8% (+3.1 pp), bottom 75% 8.4% -> 13.5%; (e) equal FR/UK/ES weights: +5.1 pp. Even the published bottom-75% series rises 25% (8.7% -> 10.9%), with a low of 7.5% in 1998-99. Report text: ch. 4.1 (HTML part-4 paras 56-58, 64; printed pp. 198-200); ES sec. V.

Why this severity. The top-1% component of C7 survives every variant: its 1980-2016 rise ranges from +2.3 to +4.7 pp and is only about 10% smaller once the population is fixed. The 'bottom 75% hovered around 10%' component does not survive. It is a secondary part of a finding the authors flag as provisional, so the issue is moderate rather than major.

The authors' response. TN 2017/8 notes that 'levels and growth values differ from Purchasing Power Parity estimates'; ch. 4.1 stresses that global wealth estimates are fragile and that adding Latin America, Africa and the rest of Asia would probably raise concentration. No statement on the European population definition was located.

3moderateconfirmed · previously known · selective presentation

For about a third of the world's adults, including Japan, Korea and Indonesia, the report has no income-distribution data and borrows other countries' patterns. Under a different borrowing rule that the authors themselves describe in a technical note, the richest 1% got about 1.7 times as much of global growth as the poorest half, rather than twice as much. The claim that the top 0.1% alone got as much as the poorest half no longer holds under that rule.

Technical detail and evidence

Why it matters, in detail. Static calibration transmits the very fast rise of Chinese and Indian top shares to slower-growing Other Asia (and similarly elsewhere), depressing bottom-50% growth there. Switching to the dynamic calibration raises bottom-50% capture by 2.8 pp (+22%) and lowers top-1% capture by 1.9 pp, so the capture ratio falls by about 24% (2.19 to 1.67). The top-0.1% vs bottom-50% comparison reverses (12.6% vs 15.5%), and also reverses in the current WID vintage (11.4% vs 12.6%). Combined with F1, the ratio is about 1.57.

Evidence. Authors' archived world outputs (scripts/adj_world_capture.py; output/04): - Static (PPP, scenario 4): top 1% 27.71%, bottom 50% 12.65%, top 0.1% 13.65% (ratio 2.19). - Dynamic: 25.77%, 15.47%, 12.63% (ratio 1.67). - MER: 38.39%, 9.23%. Adjudicator 7-region pooling (output/05): dynamic plus the MENA fix (F1) gives 24.8%, 15.8%, 12.0% (ratio 1.57). Current WID (output/10): 26.0%, 12.6%, 11.4%. Proxied sub-regions hold 31.2% of world adults and 32.8% of income in 2016 (scripts/adj_coverage.py; output/09). Start-year sensitivity (pooled): ratio 2.26 (from 1980), 2.56 (1985), 2.20 (1990), 1.55 (2000-2016). Claude also reports (not re-run by the adjudicator) that the capture shares stay close under alternative proxy rules: - applying proxies country by country: 29.1/12.2; - giving high-income Asia European shares: 27.5/12.3; - giving Eastern Europe Russia's shares: 30.0/12.2.

Why this severity. C1 survives qualitatively under every PPP variant: the top 1% captures 25-28% and the bottom 50% 12-16%. But 'twice as much' falls to about 1.6-1.7x under the authors' own disclosed alternative, and the 'top 0.1% equals bottom 50%' claim fails. This is at the upper end of moderate.

The authors' response. TN 2017/5 sec. 3.5 and App. C: results 'qualitatively similar'; 'overall impacts on the results are however limited'. Chancel & Piketty (2021, JEEA 19(6)) acknowledge that corrections 'often rely on a limited set of countries'.

4moderateconfirmed · previously known · version instability

The database behind the report has been revised many times since 2017. The headline comparison between the richest 1% and the poorest half barely changed. But the shape of the famous 'elephant curve' and some country rankings did: the Middle East, which the report called the world's most unequal region, is no longer the most unequal in the current data.

Technical detail and evidence

Why it matters, in detail. Revisions add DINA countries and new PPPs. They leave top-1% and bottom-50% capture within about 1 pp, but lower the bottom-50% share level by about 1.5 pp (about 18%) and change percentile growth rates by up to about 50 pp. The 'elephant's back' moves to P30-P50 and the trough deepens to about 20%. India and Latin America overtake the Middle East at the top of the 2016 ranking.

Evidence. Adjudicator, WID bulk 15 Oct 2026 (scripts/adj_wid2026_vintage.py; output/10): - World per-adult growth 1980-2016: 46.6% (WIR 60%). - Capture: top 1% 26.0%, bottom 50% 12.6%, top 0.1% 11.4%. - World bottom-50% share 6.3% -> 8.3% (WIR 8.0% -> 9.7%); top-1% share 16.9% / 20.3% / 21.4% / 19.8% in 1980 / 2000 / 2007 / 2016. - Growth by percentile: P20 76%, P40 102%, P50 95%, trough 19% at P79. WIR: P20 115%, P50 62%, trough about 40% at P75. - 2016 top-10% shares: Middle East (XM) 56.3%, MENA (XN) 56.1%, Latin America 58.1%, India 59.2%, Brazil 58.4%, sub-Saharan Africa 56.2% (WIR: Middle East 61%, India 55%, Brazil 55%). - US 2014 top-1% income share 19.0% (WIR 20.2%); US 2014 top-1% wealth share 35.8% (WIR 37.2%). Claude also reports Part III revisions (US 2015 net public wealth -31% vs -17% of national income; China 1978 household wealth/income 70% vs 115%); these were not re-run.

Why this severity. The headline capture shares are stable across vintages. The shape of the report's iconic curve, the level of the bottom-50% share and the 'world's most unequal region' ranking are not.

The authors' response. The report presents the global estimates as 'a first step' to be updated in future editions as new data become available.

5moderatelikely · previously known · external inconsistency

The report's estimate of how fast America's richest gained wealth used a method that assumed the rich earn the same interest rate on their savings as everyone else. Later research, including the authors' own updates, shows the rich earn more. That implies they own somewhat less than estimated: US wealth concentration still rose a lot, but roughly 15-30% less than the report says, and more so for the years after 1989.

Technical detail and evidence

Why it matters, in detail. With top fixed-income yields above average, capitalization factors at the top fall, especially in low-rate years after 2008. As a result the level and the post-2000 acceleration of top shares fall. The defensible range for the 1980-2014 top-1% rise is about +10 to +12.6 pp, against +14.7 pp in the report's data (+17 pp as stated in the text). The claim that most of the increase came from the top 0.1% still holds in the revised WID series (+10.9 of +12.6 pp). The US also enters the China+Europe+US aggregate, so C7 levels are affected; that propagation was not quantified.

Evidence. SZ (2016) appendix Table B41c (estate-based higher top fixed-income yield; AppendixTables(Distributions).xlsx) puts the 2012 top-0.1% share at 18.70% vs 22.01% in Table B1. The 2000-2012 rise falls from 6.0 to 2.8 pp (-53%) and the 1978-2012 rise from 14.9 to 11.6 pp (-22%) (Codex; verified). Equal-split-adult top-1% share: WIR data/WID-2017 22.5% (1980), 26.6% (1989), 37.2% (2014); current WID (Oct 2026) 23.2%, 28.8%, 35.8%. The 1980-2014 rise is +12.6 vs +14.7 pp (-14%) and the 1989-2014 rise +7.0 vs +10.6 pp (-34%) (output/10 and adjudicator check). Smith, Zidar & Zwick (2023, QJE 138(1), fn. 1 p. 518): 1989-2016 top 1% +6.6 pp vs +10.0 in PSZ; top 0.1% +4.6 vs +7.9. Catherine, Miller & Sarin (2025, fn. 15): SZ's printed tax-unit top-1% share 27.8% -> 41.8% (1989-2013) vs 28.6% -> 35.7% in the Feb-2022 update.

Why this severity. The direction is robust: every series shows a large rise, and SZZ conclude that inequality 'has risen substantially'. The 1980-2014 US top-1% rise is about 14% smaller in the authors' own revised series and about 25-30% smaller in heterogeneous-return estimates. The post-1989 and post-2000 rises shrink by a third to a half. This is at the upper end of moderate.

The authors' response. SZ (2016, QJE sec. IV.F) reported the estate-based and Treasury-yield robustness checks but kept the equal-yield baseline. SZ (2020, NBER w27921 and JEP) adopted a top interest premium since 2008 (about 1.4x average for the top 1%) and other corrections. They dispute SZZ's larger corrections (billionaire equity under dividend capitalization; pass-through interest biasing survey yields).

6moderatelikely · previously known · external inconsistency

The report's estimate of how much more of America's income goes to the top 1% depends on assumptions about income that never shows up on tax forms. Another respected team, using the same tax data with different assumptions, finds a rise about half as large, and the authors themselves later trimmed their numbers. Under either estimate, inequality rose more in the US than in Western Europe.

Technical detail and evidence

Why it matters, in detail. Allocation keys for income missing from tax returns move the 1980-2014 US top-1% rise between about +8.6 pp (current PSZ/WID) and about +4-5 pp (AS). The pension correction alone is -0.6 pp (2016); misreporting allocation is the largest disputed item (about 2 pp of the 6 pp 2014 AS-PSZ level gap). The US bottom-50% decline is also concept-sensitive (pre-tax vs post-tax: the report's own post-tax bottom-50% growth is +21% over 1980-2014). The global C1 numbers depend only modestly on the US input.

Evidence. Report: US top 1% 10.7% (1980) -> 20.2% (2014), bottom 50% 20.7% -> 13.1% (ExecSumm.xlsx E3, verified). Authors' revisions: PSZ Updates (Sept 2020, secs. 2.1.7, 2.2.1) cut the pension weight on non-taxable distributions, lowering the 2016 top-1% share by about 0.6 pp; the 2014 top 1% is 20.20% in the Nov-2017 tables vs 18.98% in the 2022 tables (Codex). Current WID: 10.4% (1980) -> 19.0% (2014), 18.7% (2016) (output/10). Auten & Splinter (2024, JPE 132(7)): pretax top 1% +4.4 pp 1979-2019. Iselin & Reck (2025): misreporting assumptions explain about a third of the AS-PSZ divergence, PSZ at the high end of the plausible range, AS's implied falling relative misreporting implausible. Gale, Sabelhaus & Thorpe (2023): the preponderance of evidence shows a rise; magnitude and timing debated.

Why this severity. The US-Western Europe divergence survives even Auten-Splinter's low estimate (+4.4 pp against about +2.2 pp in Western Europe). The magnitude of the US rise is uncertain by roughly 10% (the authors' own revision) to 50% (Auten-Splinter), with neutral reviews placing it in between.

The authors' response. Saez & Zucman (2020) accepted the pension-rollover point ('We are grateful to AS'; -0.6 pp in 2016) and dispute the underreporting allocation (Auten & Splinter's misreporting aggregate includes legally exempt depreciation; ranking by true income matters).

7moderatelikely · previously known · methodology

The report measures income before taxes and most government benefits. That is a legitimate choice, but it is not a measure of living standards, and it places about 30 million Americans among the poorest half of humanity. Studies based on household surveys find very different income growth for the world's middle over the same years. So the picture of who gained depends heavily on how income is defined and adjusted, and the report gives no margins of error.

Technical detail and evidence

Why it matters, in detail. Excluding most transfers and flooring negatives at zero places many rich-country adults with low market income (students, the unemployed, transfer recipients) at the very bottom of the world distribution. The global bottom 50% therefore mixes the world's poor with rich-country adults who have substantial post-transfer resources. Together with NA scaling and tax corrections, this flattens middle-of-distribution growth relative to survey-based (disposable income or consumption) curves. For the same 1988-2008 window, median growth is about 18% in the WIR distribution vs about 75% in survey-based estimates. Pre-tax and consumption curves answer different questions, so neither refutes the other.

Evidence. The report's own 2016 distribution, pooled (scripts/adj_bottom50_composition.py; output/11): 13% of US adults (30.5m) and about 9% of French, German, British and other Western European adults are below the global median, while 45% of US adults are in the global top 10%. Fig. 2.1.6 data: the poorest global percentile in 2016 is 23.9% USA-Canada and 22.6% Europe (India 0.6%). For Lakner-Milanovic's 1988-2008 window, the archived WIR world distribution gives cumulative growth of 16-18% at P50-P70, a trough of 13% at P82 and +53% for the top 1% (scripts/adj_gic_window.py; output/12). Survey-based curves show about 75% near the global median over that period (as characterized by Sandefur 2018). Yonzan, Milanovic, Morelli & Gornick (2022) find that aligned survey and tax data agree up to the top 1% in the US and Germany, and that the French survey top 1% is higher.

Why this severity. This does not alter C1 as defined (shares of pre-tax income growth). It materially qualifies the welfare reading of the 'poorest half' and the 'squeezed middle', and the curve's shape depends on concept and corrections. Its impact on the capture shares themselves is not demonstrated.

The authors' response. Box 2.4.1 explains pre-tax vs post-tax income; Piketty, Saez & Zucman (2022, JEI 20(1)) argue surveys cannot capture the top tail; Chancel & Piketty (2021, JEEA) acknowledge limited correction data in several regions.

8moderateplausible · previously known · external inconsistency

Several country estimates that feed the report are disputed by other researchers: India, Russia, the Middle East and China's wealth. The critics argue the methods could overstate inequality in some places (and understate it in others). These disputes are unresolved, and no one has measured how much they would change the report's numbers.

Technical detail and evidence

Why it matters, in detail. India's tax-based trend may embed a formalization bias, while its level may be understated because of informality; the net bias is unknown. Kapeliushnikov's Russia critique compares corrected pre-tax shares with survey Ginis, which are partly incommensurate. The Middle East level depends on transporting Lebanon's percentile corrections and on imputed Gulf national income. China's wealth tail depends on generalized-Pareto splicing to Hurun lists. The share of global growth captured depends mainly on China and India, so only the Indian concern bears materially on C1.

Evidence. Ravallion (2022, JEI 20(1); NBER w28890 pp. 14-15) on India; Kapeliushnikov (2020, Voprosy Ekonomiki 2020(4)) claims NPZ inequality is about one third too high, benchmarked on survey Ginis that NPZ argue miss the top (abstract only); Hlasny (2020, JESM 44; ECINEQ 2020-549) on the Lebanon donor correction; Alvaredo, Assouad & Piketty (2019, RIW 65(4), p. 707) note that both within-country corrections and between-country gaps drive the Middle East level; Tao, Tang & Chen (2026) and Luo & Chen (2021) on Chinese rich-list tails. Later WID vintages already lower the Middle East's 2016 top-10% share to 56% (F7).

Why this severity. These are robustness concerns of limited demonstrated impact on inputs to the regional comparisons. They could change specific levels (Russia, Middle East) but not the qualitative pattern of rising inequality.

The authors' response. Alvaredo, Assouad & Piketty (2019, fn. 10, fn. 12) answer Hlasny & Verme and report alternative tail variants; Bharti, Chancel, Piketty & Somanchi (2024, sec. 7.5) answer the Indian compliance argument; no reply to Kapeliushnikov located.

9minorconfirmed · newly identified here · internal inconsistency

Several sentences in the summary and conclusion do not match the report's own charts. For example, they put the peak of the global top 1%'s income share in 2000 instead of 2007. They describe income growth for the global middle as possibly 'zero', when the report's own curve shows at least about 40% over the period. And they say the US top 1% held 39% of wealth in 2014, when the report's own data say 37%. The charts are right; the text is careless, and mostly in the direction of making inequality sound starker.

Technical detail and evidence

Why it matters, in detail. The misdating in (a) conflates the top-10% share's post-2000 break (between-country convergence) with the top-1% share, which kept rising until the 2008 crisis. The 'zero growth' wording in (c) is not supported by the report's anonymous growth-incidence curve (at least +40% at every percentile between P50 and P99). Items (b), (d), (e), (f) and (g) overstate concentration or stagnation by 1-2 pp or by mislabeling. Item (a) understates the pre-crisis rise.

Evidence. Data checked by the adjudicator: (a)-(b) ExecSumm.xlsx E5 (16.28 / 20.89 / 22.14 / 22.12 / 20.44% in 1980 / 2000 / 2006 / 2007 / 2016); Summary sheet notes; HTML executive-summary para 97 and part-2 para 82 vs part-2 para 287. (c) E4 data: the minimum cumulative 1980-2016 growth in P50-P99 is 39.7% at P75, about 0.9%/yr. (d) E5: 7.98% -> 9.66%. (e) E8/F4.2.1: 37.24% in 2014; 38.85% is the 2012 value. (f) F4.1.3, 2050: top 0.1% 24.89% vs middle 40% 25.74%; the ES note gives '26% vs 27%' and the ch. 4.1 body 'both 25%'. (g) E6: Japan net public wealth +16.2% of national income in 2015; E7: +2.6% of national wealth. (h) The US bottom-90% wealth share is positive at 22.8-28.0% (F4.3.2, complement of F4.2.2); the negative item is its net fixed-income component (-1.8% to -2.9% in 2008-2012).

Why this severity. These are text-level misstatements. The plotted data are correct and no core conclusion changes, although several sit in the most-read parts of the report and most lean toward overstatement.

10minorconfirmed · previously known · overclaiming

The report says it has solid data for about two-thirds of the world's adults and three-quarters of world income. The actual figures are closer to 58% and 64%, and the rest is filled in by assumption. Australia, New Zealand and the Pacific were left out of the 'world' altogether, but this hardly affects the results.

Technical detail and evidence

Why it matters, in detail. The overstatement understates reliance on imputed distributions: about 42% of adults and 36% of income lack own distributional data, counting Germany and the UK as covered. Excluding Oceania and re-mapping Iran and the Maghreb shift little income across the global distribution.

Evidence. Adjudicator computation for 2016 with WID-2017 inputs (scripts/adj_coverage.py; output/09): - full-DINA countries (US, FR, CN, IN, BR, RU) plus the AAP Middle East: 55.4% of adults, 57.9% of income; - adding Germany and the UK: 57.8% and 63.5%; - the claim is approached (62.3% and 70.8%) only by counting all of Western Europe as covered. Oceania holds 0.55% of world adults and 1.09% of income; the global country list in runwir-program.do is 'AF AS BR CN DE EE FR GB IN ME NA RU SA US WE'. Region assignments are in regions-codes.dta (IR in AS; AM, AZ, GE, IL, DZ, MA, TN, LY in ME).

Why this severity. The box overstates direct data coverage by about 9 pp of adults and 12 pp of income. The consequences of proxying are graded in F4, and the omissions and mismatch have negligible effect on the headline (Claude: adding Oceania gives 28.1/12.4).

The authors' response. Box 2.1.1 and TN 2017/5 describe the proxying as 'an over simplification' with limited impact on general conclusions.

11minorconfirmed · newly identified here · data error

A programming mistake scrambled the estimated income distributions of Germany and the United Kingdom inside the global calculation. The two errors largely cancel out, so the world-level results are essentially unchanged.

Technical detail and evidence

Why it matters, in detail. The UK's lower distribution is scaled down and Germany's up. The two errors largely offset in the merged France-Germany-UK distribution used as the proxy for Western and Eastern Europe, so Europe's aggregate and global statistics barely move. The country-level Germany/UK distributions inside the global code are wrong, but they are not the ones shown in the country chapters.

Evidence. Code read by the adjudicator. Saved 1a-prepare-data.dta (Codex and Claude): UK top-10% share 36.9% (input) vs about 28% implied for 1980; Germany 31.9% vs 35.8%. WID-2017 holds only an individual-adult series (sptinc992i) for the UK, while the other countries use equal-split adults (adjudicator check of wid-final.dta). Effect of correcting it: European 1980 top-10% share +0.4-0.5 pp; C1 top-1% capture 27.67% -> 27.70% (Codex) or 28.10% -> 28.15% (Claude); bottom 50% +0.01 pp.

Why this severity. This is a genuine contradiction between code and documentation, but its effect on every core finding is negligible (under 0.1 pp on C1).

12minorconfirmed · newly identified here · internal inconsistency

The report's tables of how fast different wealth groups grew do not add up with its own charts, partly because a coding slip left Spain out of the totals. The top 1% captured about 35% of wealth growth rather than 37%, a small difference that does not change the message.

Technical detail and evidence

Why it matters, in detail. Because Spain is omitted from the totals and group means divide by the full population, top-group growth is overstated relative to the average by about 0.05 pp/yr. Separately, the capture table's 37.1% cannot be derived from the published shares (35.1%). Under any consistent computation the top 1% captured about 35% of 1980-2017 per-adult wealth growth.

Evidence. Adjudicator (scripts/adj_wealth_tables.py; output/08): Table 4.1.1 average growth of 2.876%/yr and Fig. 4.1.3 shares of 27.87% -> 32.59% imply top-1% growth of 3.31%/yr, not the printed 3.39%. Top 0.1%: 4.29% vs 4.36%; top 0.01%: 5.41% vs 5.48%. Implied capture: 35.1% / 19.0% / 10.4% vs the printed 37.1% / 21.6% / 12.4% (Codex; reproduced exactly). Emulating the code's mixed denominators (scripts/adj_wealth_spain_bug.py) gives 2.877% average and 3.36% top-1% growth, against 3.30% when computed consistently. This explains most of the growth-rate gap but not the 37.1% capture, which remains unreconciled. The ch. 4.1 text quotes '1.9% for average income' where Table 4.1.1 gives 1.39% for 1987-2017; 1.9% is world average wealth growth (Claude).

Why this severity. This is an internal inconsistency of about 2 pp in capture shares and about 0.07 pp/yr in growth rates. The qualitative finding that top wealth grew much faster than average is unaffected.

13minorconfirmed · newly identified here · replication failure

One of the report's headline projections cannot be rebuilt from the files the authors released. It estimates how much the poorest half of the world would earn in 2050 if every country followed Europe's path, and the released files give a figure about 9% lower. The broad point still holds: European-style policies would leave the poorest half with roughly twice the income they would have under US-style policies.

Technical detail and evidence

Why it matters, in detail. The December 2017 code archive predates the April 2018 figure data, so a later run is a plausible but undocumented explanation. The EU-US gap in 2050 bottom-50% income is 17% smaller in the archive (EUR 3,871 vs EUR 4,659), and the bottom-50% share under the EU path is 1.2 pp lower.

Evidence. Adjudicator, gpinterized-projections.dta (WO, 2050). Bottom-50% mean: EU8016 EUR 8,350 (share 11.71%), US8015 EUR 4,479 (6.28%), BAU8016 EUR 6,219 (8.72%). Published (C5.1.xlsx F5.1.3 and F5.1.1; ES Fig. E11): EUR 9,138 (12.88%), EUR 4,479 (6.31%), EUR 6,299 (8.88%). Top 1%: archive 19.38% / 28.03% / 24.53% vs published 19.05% / 28.02% / 24.16%. In F5.1.3 the US-path value sits under the 'Europe's trend' heading and vice versa. The EU/US ratio is 1.86 in the archive vs 2.04 as published.

Why this severity. One conditional projection level cannot be reproduced and differs by 8.6%. The qualitative contrast between scenarios, a roughly two-fold difference in the bottom half's income, survives.

14minorconfirmed · newly identified here · documentation gap

The report says its 2050 wealth projection uses trends since 1987, but the code uses trends since 1980, and a data file needed to rerun it is missing. Using the period the report describes would, if anything, make the projected concentration slightly higher.

Technical detail and evidence

Why it matters, in detail. The non-Forbes groups are calibrated on 1980-2017 rather than 1987-2017 average log growth. With the 1987 window the top groups' relative growth is faster, so the published 2050 shares are not inflated by the discrepancy. The Forbes tail cannot be rebuilt without the missing file.

Evidence. Adjudicator diagnostic (scripts/adj_projection_window.py; output/08): the published 2050 ratio of P99-99.9 to middle-40% average wealth is 24.064. Extrapolating the 1980-2017 trend gives 24.065; the stated 1987-2017 window gives 27.04. The two compounding rules differ negligibly (24.063 vs 24.065). A five-group approximation (scripts/adj_projection_shares.py) with the stated 1987 window gives higher 2050 top shares: top 1% about 42.5%, top 0.1% about 26.9% vs middle 40% about 25.6%.

Why this severity. This is a documentation and reproducibility gap in an explicitly conditional scenario. Using the documented window would, if anything, strengthen rather than weaken the projected concentration.

The authors' response. Box 4.1.1 stresses that the scenarios are conditional extrapolations of past group growth rates, not forecasts.

15minorconfirmed · newly identified here · internal inconsistency

Many statements in the country chapters quote numbers that differ from the chapters' own charts, sometimes by a lot, such as dating Russia's surge in top incomes about eleven years too early. The charts are generally correct, so the overall conclusions stand, but readers relying on the text would be misled on details.

Technical detail and evidence

Why it matters, in detail. The errors mix vintages, fractiles (top 10% vs top 1%), years (1996 vs 2007) and concepts (private vs national wealth). The France chapter overstates concentration in five statements by 3-5 pp and the long-run simulation by 8-11 pp. None alters a core estimate.

Evidence. Russia: F2.3.5 gives 15.8% in 1996, with 26.9% first reached in 2007. The 'top 1%' figures in ch. 2.8 are top-10% values (F2.8.5: 23.7% -> 45.5%). US: the text's own 27x -> 81x ratio with flat bottom incomes implies about +200% (EUR 442,800 -> 1,340,000). Fig. 3.1.2: China 487% in 2015 equals private wealth (F3.4.2); national wealth is 710%. France: F4.4.1 gives 50.3% / 51.1% / 57.1% / 55.9%; F4.4.8 gives 73.6% (2100) and 77.5% (2150). Claude's sub-audit lists about 50 such discrepancies in chapters 2.2-2.12 and Parts III-IV; the remainder were not re-verified.

Why this severity. These are narrative errors. Figures and tables are largely correct, and the chapters' qualitative conclusions are unchanged. Their number points to weak quality control of the text.

16minorconfirmed · newly identified here · arithmetic or transcription

There are a number of small arithmetic and labelling mistakes in the text and technical notes. None affects the report's conclusions.

Technical detail and evidence

Why it matters, in detail. Several ch. 2.1 growth figures cannot be reproduced from the report's own aggregates, and they understate rather than overstate the Euro-American middle-class squeeze. The worked-example error does not propagate, because the code uses the correct formula.

Evidence. Code aggregates (regions-aggregates.dta): Euro-America 57.3%, world 58.9% (Table 2.1.1: 60%), China+India+Euro-America 85.9%. Table 2.1.1 gives US-Canada 63%; 74.5% is the US alone. The China+Europe+US universe has about 1.89bn adults, so 1/20m and 1/100m are about 95 and 19 adults (R code: round(pop/20e6), round(pop/100e6)). The worked example equals 0.4 (Codex). The R code and TN 2017/7 use FR/UK/ES. Brazil's Lei Áurea (Law 3,353) dates from 13 May 1888.

Why this severity. These are small arithmetic and labelling errors that do not affect any conclusion.

17minorconfirmed · previously known · documentation gap

Many of the '2016' figures are really 2013-2015 figures carried forward. The technical notes mention this but the main text does not. It mainly matters for countries where inequality kept changing, such as India.

Technical detail and evidence

Why it matters, in detail. Holding shares constant after the last observation assumes distribution-neutral growth in 2013/14-2016. For India, later data imply a higher top-10% share (59.2% vs 55.5%).

Evidence. 1a-prepare-data.dta top-10% shares (adjudicator): India 55.456% in each year 2013-2016; US 47.017% for 2014-2016; Germany 38.128% for 2013-2016; China and Brazil constant for 2015-2016. Current WID puts India's 2016 top-10% share at 59.2%. The Brazil item is from Claude and was not re-verified.

Why this severity. This is labelling and documentation that the technical notes partly disclose. Any error depends on unobserved changes after the last data year.

The authors' response. TN 2017/8 states that values after the last available year are extrapolated with constant shares.

18minorlikely · previously known · overclaiming

The report says research has 'demonstrated' that higher top tax rates reduce inequality even before taxes, by discouraging aggressive pay bargaining. The evidence is a strong statistical association, and the authors' own source paper says it does not prove cause and effect. Other economists offer different explanations.

Technical detail and evidence

Why it matters, in detail. The pre-tax effect of top tax rates is identified in PSS (2014) from long-run cross-country co-movements, with the bargaining share obtained as a residual after a weakly estimated real-supply elasticity. Reverse causality and alternative channels (avoidance, superstar effects, real responses) are not ruled out.

Evidence. Piketty, Saez & Stantcheva (NBER WP 17616, sec. on international evidence): 'Naturally, the strong correlation between top tax rates and top income shares does not prove a causal link. Reverse causality scenarios ... remain a possibility.' (adjudicator, local copy). Competing interpretations: Kaplan & Rauh (2013, JEP 27(3)) attribute top-pay growth to superstar and market forces; Mertens & Montiel Olea (2018, QJE 133(4)) find real effects of top-rate cuts that rent-seeking alone cannot explain; Rothschild & Scheuer (2016, REStud) show the bargaining-based optimal-tax correction is sensitive to who bears the rents.

Why this severity. This is policy rhetoric rather than a descriptive core finding. It overstates the strength of the causal evidence but does not affect any estimate.

The authors' response. PSS (2014) acknowledge reverse causality and present CEO-pay evidence (pay for luck is larger when top rates are low) as support for the bargaining channel.

19minorplausible · previously known · methodology

The report counts only wealth that people own and can sell, not the value of future pensions and Social Security. Counting those would make wealth look more evenly spread and its concentration rise less, but that is a different definition rather than a mistake.

Technical detail and evidence

Why it matters, in detail. Including non-marketable social-insurance wealth, which is large and widely held, lowers measured top shares and their trend, sharply so with low discount rates. Rent-based housing valuation would dampen the rise in wealth/income ratios. Neither shows an error in the report's marketable-wealth series.

Evidence. Catherine, Miller & Sarin (2025, J. Finance 80(3), Table III): SCF top-1% share 1989-2019 rises +6.4 pp on marketable wealth vs +1.9 pp including Social Security wealth. Weil (2015, AER P&P) on human capital and transfer wealth. Bonnet, Bono, Chapelle & Wasmer (2014) on housing-driven wealth/income ratios.

Why this severity. This is a definitional issue: the report defines its concept, and its trend conclusions hold for marketable wealth.

The authors' response. Saez & Zucman (2020, JEP) argue that government promises and unfunded pensions are not wealth in the SNA sense and are excluded by the Fed and other statistical agencies.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

Excluding Russia/Ukraine from the European proxy cuts the 1980-2016 rise of the global top-1% wealth share to +3.8 pp (+1.6 pp together with PPP) and roughly doubles the bottom-75% share; the global-wealth issue is therefore 'major'. — The magnitudes were not reproduced. Removing RU, UA, BY and MD adults from the WID-2017 QE population leaves the rise at +4.5 pp (vs +4.7 pp), and giving those adults Russia's own wealth gives +4.2 pp. PPP plus exclusion gives +2.3 pp. In these variants the bottom-75% share rises by about 60% (6.2% -> 10.0%; 6.0% -> 9.6%). The construction problem itself is upheld (F2), graded moderate.
The Fed DFA's ~31% US top-1% wealth share (2014) shows that the report's 37-39% is too high. — This is not a like-for-like comparison. The DFA uses households and includes consumer durables and unfunded defined-benefit pensions. Saez & Zucman (2020, JEP) show that harmonized DFA and SZ series both give about 38% (2018) for the top 1% of tax units. The valid evidence of overstatement comes from heterogeneous-return estimates and the authors' own revisions, which F3 uses.
Box 2.1.1 misdescribes the proxy method (Malaysia given the China-India distribution individually) because the code applies proxies to whole sub-regions, ignoring between-country inequality within them. — TN 2017/5 secs. 3.1-3.3 explicitly describe sub-regional pooling and report finer tests. Claude's own country-by-country variant leaves the headline essentially unchanged (29.1% vs 12.2%). This is a simplification in an explanatory box, not a substantive error, and is noted only as context under F10.
The lower 2000-2016 capture ratio (19.2% vs 12.4%) is evidence that the headline is fragile. — The report's claim is explicitly about 1980-2016, and a later window answers a different question. Other windows give higher ratios (1985-2016: 2.56). This is reported only as context in F4, not counted against the finding.
The wealth projection code's compounding of 1 + mean(log growth) instead of exp(mean log growth) is part of the reproducibility problem. — It is numerically immaterial: the two rules give 2050 ratios of 24.063 and 24.065 (published 24.064). Only the calibration window and the missing Forbes input are retained (F14).
The difference between the archived world capture shares (27.71% / 12.65%) and the printed 27% / 12% is a version mismatch. — The archived growth workbook rounds to 27.71% and 13%, and TN 2017/5 prints 28% / 13%. The printed 27% / 12% is a truncation-level difference with no analytical consequence; it is listed among the trivial slips in F16 rather than treated as a version problem.

Audit notes

What reproduced. Reproduced: (1) The country-to-region aggregation step (0c), exactly (max relative error 6.5e-8). (2) C1, by pooling the authors' 15 archived sub-regional distributions: top 1% 28.1%, bottom 50% 12.5%, top 0.1% 14.0% of 1980-2016 per-adult growth, against archived output of 27.71% / 12.65% / 13.65% and printed figures of 27% / 12% / 13%. (3) The authors' alternative outputs: dynamic calibration 25.77% / 15.47%; MER 38.39% / 9.23%. (4) The global top-1% income share path (15.9% in 1980, peak 22.1% in 2006-07, 20.4% in 2016). (5) The China+Europe+US wealth series (Fig. 4.1.1), by emulating the authors' R code with the shipped inputs and WID-2017 populations: 27.6% -> 32.3% and bottom 75% 9.0% -> 11.0%, against published 27.9% -> 32.6% and 8.7% -> 10.9%. (6) Saez-Zucman's heterogeneous-yield US top-0.1% wealth share (18.70% in 2012). (7) The current WID-vintage statistics (capture 26.0% / 12.6%). Not reproduced: - Table 2.1.1's 60% world growth: 58.9% with the hard-coded Middle-East override, 53.2% without it, 52.6% in WID-2017's world aggregate. - Table 4.1.2's wealth-capture shares: 37.1% / 21.6% / 12.4% printed vs 35.1% / 19.0% / 10.4% implied by the published shares. - The Europe-path 2050 bottom-50% income: EUR 8,350 in the archive vs EUR 9,138 published. - The stated 1987 calibration window of the wealth projection: the published output matches 1980-2017. Not attempted: - the authors' gpinter merges (approximated by mass-point pooling, within about 0.4 pp of published shares); - country DINA series from primary tax and survey data; - the Forbes-based wealth projection (input file missing); - the Harvard University Press and translated editions.

Limits of this audit. The adjudicator did not run the authors' gpinter (generalized Pareto) merges. Global distributions were rebuilt by pooling bracket means, which matches published shares within about 0.4 pp and capture shares within about 0.5 pp; sensitivities are stated relative to this emulation. The global-wealth emulation uses frozen WID-2017 populations and 2016 conversion factors rather than the live WID downloads in the authors' R code. Its Russia block is a crude seven-point distribution that also assigns Russia's per-adult wealth ratio to Ukraine, Belarus and Moldova, so it is an approximation, not a full alternative series. Country DINA inputs (US, China, India, Russia, Middle East, Brazil, Africa) were not re-estimated from primary data. Known critiques were assessed from the critique corpus and partial texts (Kapeliushnikov from the abstract only; Galbraith from the author's summary). Claude's country-level proxy tests, its geographic-composition result and most of its roughly 50 chapter-level text-versus-data items were not all re-run, and are marked where used. The Harvard University Press, French and German editions were not compared; the online HTML and June-2018 PDF were used. The full 709 MB code archive was not downloaded; selected members, including the frozen WID database (wid-final.dta), were fetched by HTTP range request.