Emmanuel Saez, Gabriel Zucman · New York: W.W. Norton, October 2019

The Triumph of Injustice: How the Rich Dodge Taxes and How to Make Them Pay

21 problems found: 3 major · 7 moderate · 11 minor

What the paper claims

In 2018, for the first time in a century, the 400 richest Americans paid a lower total tax rate (all taxes, all levels of government) than the working class (bottom 50%).
The US tax system is a 'giant flat tax' (every group pays 25-30%) that becomes regressive at the very top.
Top-end effective tax rates have collapsed since mid-century, mainly through the erosion of the corporate tax, while working-class rates rose.
Capital is now taxed less than labor for the first time (2018), and the macroeconomic tax rate has fallen by almost four points since the late 1990s.
The corporate tax has collapsed and profits are shifted to havens: federal corporate revenue fell by almost half from 2017 to 2018; ~60% of US multinationals' foreign profits are booked (the book also says 'shifted') in havens; 40% of global multinational profits are shifted; >20% of non-US profits (~$100bn) sit in 'stateless entities'.
Policy scoring: a 2%/3% wealth tax raises ~1% of GDP; the revenue-maximizing top marginal rate is ~75% (60% average for the top 1%, ~4% of NI extra); a 25% country-by-country minimum tax would have raised close to $100bn in 2016; a 6% national income tax raises 5.6% of NI; the package raises 9.8% of NI.
Tax evasion rises with income: ~10-12% of taxes owed for most groups vs ~25% for the ultra-rich (2018).
Context (inherited from DINA / Saez-Zucman wealth series): top 1% income share ~10% (1980) -> ~20%; bottom 50% ~20% -> ~12%; bottom-50% income stagnant (~$18,500); top 1% wealth share 22% -> 37%.

Every problem found

1majorconfirmed · previously known · overclaiming

The famous 'billionaires pay less than the working class' result rests on a gap of about one percentage point. Count the tax credits that low-income workers receive through the tax system as reducing their taxes, as government agencies do and as the authors themselves did in a later version of their online calculator, and the working class pays less than billionaires. Other reasonable choices push the other way, so the honest conclusion is that the two groups pay roughly similar rates, not that billionaires clearly pay less.

Technical detail and evidence

Why it matters, in detail. Gap (top 400 minus bottom 50%) moves from -1.17 pp to +2.00 pp with refundable EITC+CTC netted (authors' simulator), or about +3.5 pp with all refundable credits netted (public-file delta). It moves to about -2.7 pp under PSZ-2018 or 25%-to-labor corporate incidence, and to -1.9 pp with the full adult population. The sign of the headline comparison is therefore inside the range spanned by defensible conventions. Only the gradient within the top (top 400 4-10 pp below P99.9-99.99/top 0.1%) is convention-robust.

Evidence. Appendix Table B1, row 2018: D=0.24214, K=0.23042; reproduced exactly from StataOutput/full2018.xlsx (scripts/recompute_2018_groups.py). taxsimulator2.xlsx (readme: revised 1/20/2020), sheet calculatorbig, col EM 'Trump (2018)' vs col FW 'Taxes-refundable credits (2018)', aggregated over p0-p49 with adults x income weights: 24.22% -> 21.04%; top-400 row 23.04% unchanged; implied refundable EITC+CTC $91.7bn (scripts/simulator_credit_profile.py). Sheet figure-taxcredits note: 'The figure also depicts the effect of factoring in existing federal refundable tax credits (the Earned Income Tax Credit and the Child Tax Credit)'. Public 2018 DINA file (scripts/toi_port.py): netting all dicred ($144bn) lowers the bottom-50% rate by 4.64 pp. Opposite-direction variants: corporate tax on all non-housing wealth (corptax0) -1.59 pp for the top 400; 25% of corporate tax on labor -1.28 pp (bottom 50% +0.26 pp); full adult population +0.57 pp for the bottom 50% (book-vintage full-population series 24.97%). Book Ch. 1 'Plutocracy': 'we've done nothing to exaggerate our numbers, quite the contrary'.

Why this severity. The literal ordering in C1 flips under the convention used by all US agencies and under the authors' own alternative profile, and the 1.2-point margin is smaller than many single convention choices. It is major rather than critical because the robust part of C1 and C2 survives every variant: the top 400 pay roughly the economy-wide average or less, and well below the top 0.1%.

The authors' response. Zucman (as reported by Gleckman, TPC, Oct 11 2019): refundable credits are spending, no different from SNAP. Saez & Zucman (2026, JPubE 259:105675, Table 1 notes) keep the convention ('Refundable tax credits are not included as negative tax (as they are treated as transfers ... in the national accounts)'). The Jan-2020 simulator added a credits-netted profile, but its summary still states that all groups pay about 28% except the top 400 (23%).

2majorconfirmed · previously known · version instability

The 2018 figures were estimates made before that year's tax data existed, and they used early numbers that later proved to understate corporate taxes. When the authors updated their own data in 2020 and 2022, billionaires' 2018 tax rate rose to 27-29%, above the working class's 25-26%. A separate 2025 study by the authors put it at about 24%. In the authors' own updated numbers, the 'first time in a hundred years' claim does not hold.

Technical detail and evidence

Why it matters, in detail. In like-for-like full-population terms the gap moves from -1.9 pp (book vintage) to +1.8 pp (Sept 2020) and +3.7 pp (Feb 2022); restricted-sample equivalents would be about 0.7 pp larger. Most of the revision comes from corporate and business-property taxes allocated to the top 400 (+2.7 and +2.5 pp), after NIPA revisions and the Forbes calibration of top wealth. Holding income fixed, the corporate revision alone puts the 2018 top-400 rate at 24.5-24.8%, vs 24.3% for the bottom half.

Evidence. ALFRED FCTAX vintages (2019-04-01 to 2019-07-25: 284.6/157.7; 2019-07-27: 251.5/147.4; 2021-10-01: 245.4/208.8; 2023-10-01 and 2026-10-01: 230.3/225.0). BEA NIPA bulk file (Sept 2026): A054RC 297.2/297.4; W025RC 284.8/285.1; B075RC/A032RC 2018 1.28%, 2009 1.27%, 1982 1.19%, 2019 1.15%; (taxes + social contributions)/NI 31.5-32.6% (1995-2000) vs 29.9% (2019) (scripts/nipa_macro.py). parameters.csv ttfkcot: 2016 392.9, 2017 350.7, 2018 220.4. PSZ2020 and PSZ2022 AppendixTablesII(Distrib).xlsx, sheet 'taxrates' (taxtop400 / taxbot50 / taxall): 2018 27.23 / 25.40 / 28.34 and 29.31 / 25.65 / 28.74; 2019 26.44 / 25.54 and 30.82 / 26.01. Book vintage (PSZUpdates/PSZ2019Distributions.xlsx 'taxrates'): 2018 23.04 / 24.97 / 27.82 (scripts/later_vintage_top400.py; output/vintage_top400.txt). Top-400 components, Feb 2022 vs book: corporate 7.81 vs 5.08; estate 1.86 vs 1.11; income tax 9.53 vs 9.23; implied business property 7.4 vs 4.95. BSYZ (2025, NBER w34170) Table 2. Appendix Tables A3 (2019 copies 2018) and A5.

Why this severity. In the authors' own updated series the headline 2018 top-400 level rises by 4-6 points (18-27% of its level), enough to reverse the crossover. Their administrative estimate supports a low level, so the ordering is undetermined rather than refuted: major, not critical.

The authors' response. PSZ update notes (Sept 2020): 'Distributional series for 2017, 2018 and 2019 are provisional and subject to revision'. BSYZ (2025, p. 2) describe the book's 23% as delivering 'similar bottom-line results' to their 23.8%. No acknowledgment was found of the higher top-400 rates in the 2020 and 2022 releases.

3majorconfirmed · previously known · methodology

The book says taxes on the rich have collapsed since the 1950s. They did fall, but by how much depends heavily on who is assumed to have borne the old, high corporate tax. Under the authors' own earlier method, or the government's method, the fall is about half as large as the book shows, or less.

Technical detail and evidence

Why it matters, in detail. Assigning the then-large corporate tax (about 4-6% of NI) only to concentrated C-corporation equity raises mid-century top rates (top 0.1% in 1962: +7.5 pp vs PSZ). Recent top rates are lowered by the pure-gains denominator and the ITEP sales-tax allocation. In SZ (2026) Fig. 3, the CBO allocation halves the 1950-2021 top-1% decline (-7 vs -13 pp).

Evidence. PSZ Nov-2017 archive Appendix Tables II TG1 vs TOI Table B1 vs PSZ Feb-2022 TG1 (scripts/vintage_trend_compare.py). Top 0.1%, 1962 -> 2014: 43.6 -> 39.8 (-3.8) vs 51.1 -> 34.1 (-17.0) vs 52.0 -> 36.4 (-15.6). Top 0.1%, 1950 -> 2014: -9.3 vs -22.8 vs -21.0. Top 1%, 1950 -> 2014: -3.2 vs -13.4 vs -11.4. Top 0.01%, 1950 -> 2014: -21.6 vs -35.1 vs -37.6. Online Appendix pp. 4-5 ('This explains why effective tax rates at the very top are higher in the book than in PSZ (2018) in the 1950s-1960s'). SZ (2026) sec. 3.2 and Fig. 3. Public 2018 file: corptax0 (PSZ key) vs corptax (TOI key).

Why this severity. The direction of the long-run decline is robust: even under PSZ (2018), the top 0.01% falls about 21 points over 1950-2014. But its magnitude, the book's 'collapse' and 'rolled back to the 1910s', roughly halves or more under the authors' previous or agency-style conventions. That change is well beyond a quarter of the trend.

The authors' response. Saez & Zucman (Oct 2019, 'Clarifying Distributional Tax Incidence') and SZ (2026, JPubE): for current taxes, the corporate tax should be assigned to shareholders; conventional approaches shift taxes while holding pre-tax incomes fixed and are internally inconsistent.

4moderateconfirmed · previously known · internal inconsistency

For working people, the study first subtracts Social Security payroll taxes from their income and then counts those same payroll taxes as taxes paid, even though the book tells readers their income figure includes payroll taxes. Fixing this lowers the working class's measured tax rate by almost 2 points, enough on its own to put billionaires above them.

Technical detail and evidence

Why it matters, in detail. Implied published bottom-50% rate is about 22.6% (gap +0.5 pp) and P50-90 about 25.7%; top-end rates are unchanged. The inconsistency also inflates the 1950-2018 rise in bottom-50% rates by about 1.5 pp, since payroll taxes were about 3% of income in 1950. Splinter's fuller CBO reconciliation, which adds contributions and Medicare benefits, lowers bottom-half rates by about 5 pp (2010).

Evidence. USdinaonline codebook (July 16, 2019) sec. A.3.2: definitions of peinc, ptinc, plinc, plcon and ssuicontrib. Identity checks on usdina2018.dta: plinc = flinc+plcon+plbel (exact); peinc = ptinc+govin+npinc+prisupen+invpen (exact); plcon total -$2,135bn (includes private pensions); ssuicontrib $958.3bn. tax-by-gperc.do line 7 (compo includes govcontrib). Book Ch. 1 'The average income of working-class Americans: $18,500' (quoted text). Saez & Zucman (Oct 2019 incidence note): 'employer and employee payroll taxes are a tax on labor'; SZ (2026) fn 18: pre-tax national income 'is net of Social Security taxes'. Port: adding ssuicontrib back to the denominator lowers the bottom-50% rate by 1.66 pp (23.39% -> 21.73% on the public file), the middle 40% by 1.89 pp and the top 400 by 0.01 pp.

Why this severity. Lowers bottom and middle rates by about 1.7-1.9 points (~7%). By itself that reverses the 1.2-point C1 margin and makes the profile more progressive below the top. Top-end rates and the long-run decline at the top are unaffected. Graded moderate because its effect on C1 is part of the knife-edge rated major in F1.

The authors' response. SZ (2026, fn 18) restate the DINA convention: pre-tax income 'is net of Social Security taxes ... and symmetrically includes Social Security benefits'. They do not address the inclusion of the same contributions in the tax numerator.

5moderateconfirmed · newly identified here · documentation gap

For the richest, the authors add part of their stock-market gains to their income before computing tax rates, and the book does not tell readers this. Without that adjustment, billionaires' tax rate in 2018 is about 25%, slightly above the working class's.

Technical detail and evidence

Why it matters, in detail. Without pure gains the gap is +1.07 pp (exact on the published bins). The threshold binds only in high-realization years (1996-2001, 2004-08, 2012-18), so it lowers recent top rates relative to 1950s benchmarks. It is also asymmetric: unrealized housing and pension appreciation of lower groups is not added.

Evidence. tax-by-gperc.do lines 32-45 (pure_kg = max(0, frac-0.03)/frac x kg_agi; denom = peinc + pure_kg). full2018.xlsx top-400 row: tax $105.20m, peinc $416.0m, denom $456.6m. Recomputation (scripts/recompute_2018_groups.py): top 400 23.04 -> 25.29; top 0.1% 30.9 -> 34.2; top 0.01% 28.5 -> 32.2; bottom 50% 24.21 -> 24.22. Online Appendix pp. 9-11; Table B1 header ('% of pre-tax income including pure taxable capital gains'). Book Ch. 1 ('divide that figure by each group's pre-tax income').

Why this severity. A defensible but discretionary choice (the 3% threshold) that moves the 2018 top-400 rate by 2.25 points (~10% of its level). It decides the C1 ordering on its own and contributes about 3-4 points (~15%) of the 1950-2018 decline at the top 0.1-0.01%. Material to the knife-edge, not to the existence of a decline.

The authors' response. SZ (2026, fn 19): excluding such gains would make top rates 'artificially and misleadingly' pro-cyclical. The online appendix (sec. 1.1) gives the retained-earnings rationale for the 3% threshold.

6moderateconfirmed · previously known · documentation gap

The 2018 billionaire figure comes from a single statistical 'stand-in' record, which the authors adjusted by hand in ways the appendix does not describe; for example, they doubled it to match Forbes. That does not change the billionaire tax rate itself, but it makes the number harder to check and affects related claims, such as how much a wealth tax would raise.

Technical detail and evidence

Why it matters, in detail. The cap is the dominant lever (the rate goes from 23.0% to about 29.5% without it). The BPEA triangulation (51/50/41%) and BSYZ's 42% support about 45%. The cap is applied to DINA income (about 8-10% of top-400 wealth) rather than to wealth x macro return (about 6.5%), which if anything overstates top-400 income tax. The agreement with BSYZ (23.8%) reflects offsetting differences in components and in income/wealth ratios (TOI income about 9.7% of wealth vs about 5.6% in BSYZ).

Evidence. tax-by-gperc.do lines 54-69; tax-by-gperc-details.do lines 66-81. full2018.xlsx top-400 row: nb 514.67, thres = avg = $416.0m, hweal $4.27bn. Recomputation: top 0.001% rate 25.0% with doubling vs 25.7% without; extra revenue at a 50% rate $116bn vs $85bn. TB1 top 0.1% (31.35%) matches undoubled weights, while full2018/TC8 use doubled weights. The top-400 record supplies $1,941bn of the $3,129bn wealth base above $1bn (full2018det). top400.do: min(0.45, ratio); public-file raw ratio 0.94. Port: no cap +6.4 pp; cap at 55% +1.5 pp. TB4 col Q formulas (taxrates_wide_internal for 1963-2012, taxrates_wide for 2013-2018); taxrates_wide_internal 2014: taxT400 0.1288, inctT400 -0.0575. Online Appendix pp. 6-9, 21, 25-27. Components, BSYZ (2025) Table 2 vs TOI: income+payroll 11.0 vs 9.6; corporate 8.9 vs 5.1; property+sales 3.4 vs 7.3; estate 0.6 vs 1.1.

Why this severity. No bias in the top-400 rate itself is demonstrated: the doubling is rate-neutral and the cap is corroborated. But the headline cell is fragile, its documentation is incomplete, and secondary claims (the top-0.001% rate and the $100bn figure, the wealth-tax base) rest on undisclosed calibrations.

The authors' response. Online Appendix pp. 6-9: the 45% ratio 'could be refined in future work', and excluding foundation income 'tends to bias our estimated top tax rates upwards'. BSYZ (2025) find a fiscal/economic income ratio of 42%.

7moderateconfirmed · newly identified here · speculative assumption

The chart showing that the rich cheat more is built mostly from assumptions applied equally to everyone (for example, that half of estate tax is evaded), combined with the fact that the rich pay more of those particular taxes. Change those assumptions and most of the gap between the rich and the middle class disappears.

Technical detail and evidence

Why it matters, in detail. Recomputed from DataF3 components, top-400 evaded/owed = (0.40+2.55+1.01+0.17+0.03)/30.47, or about 13.6%, with corporate and estate evasion at 7.5%, vs 24.4% published (P50-60: 13.3%). The only group-specific measured input is income-tax evasion at the top (GLRRZ), and that input is itself contested (Auten-Splinter).

Evidence. SZ2019MainFigures.xlsx DataF3: cols N-S and AA-AF; assumptions block D29-D33. Book Ch. 3 'Tax avoidance versus tax evasion: a flawed debate'; notes to Figs. 3.1-3.2.

Why this severity. Substantially affects a secondary claim: with corporate and estate evasion set at the property-tax rate, the top-400 evasion share falls to about 14%, close to the middle class. The core tax-rate findings use taxes paid and are unaffected.

The authors' response. Book Chs. 3-4: profit shifting violates the economic-substance doctrine and is better described as evasion; random audits miss sophisticated evasion at the top.

8moderatelikely · previously known · methodology

Poor households pay sales tax when they spend government benefits. The book counts that tax as theirs but does not count the benefits as their income. The authors themselves later said this portion should be netted out, and doing so lowers the working class's rate by about one point.

Technical detail and evidence

Why it matters, in detail. Sales tax is split in proportion to the transfer share of disposable cash income (SZ 2026 eq. 7 with linear taxes), on the coarsened public file. The magnitude is uncertain (about +/-0.3 pp) because of the overlap between dicab and dicred and the ITEP bin structure, but the sign is unambiguous.

Evidence. Online Appendix pp. 2-3 (ITEP allocation by disposable cash income plus income taxes minus SNAP) and pp. 4-5 (rationale for the population restriction). SZ (2026) online appendix A, 'Tax rates and transfers' (eq. 7); BSYZ (2025) Table 2, fn 12. Public 2018 file (scripts/transfers_salestax.py): cash transfers excluding SNAP are 9.3% of the bottom half's disposable cash income (14.0% including refundable credits). That implies $22-33bn of sales taxes on transfer-financed consumption, which lowers the bottom-50% rate by 0.77-1.17 pp.

Why this severity. Using only the authors' own later convention, the C1 margin shrinks from 1.2 points to 0-0.4 points. The effect on the bottom-half level is modest (~4%) and there is none at the top.

The authors' response. TOI's online appendix (pp. 4-5) justifies the half-minimum-wage restriction on exactly this ground. SZ (2026, online appendix A) and BSYZ (2025, fn 12) later adopt the netting concept.

9moderatelikely · previously known · speculative assumption

The estimate that a wealth tax would raise about 1% of GDP assumes the rich would hide or shelter only 15% of their wealth. That figure comes from a few foreign studies of much smaller taxes; with plausibly more avoidance, revenue could be a fifth to a half lower.

Technical detail and evidence

Why it matters, in detail. 30% avoidance gives 0.95% of NI; rate-scaled semi-elasticities of -8 and -13 give 1.10% and 0.94%; no avoidance gives 1.36%. The Summers-Sarin estate-tax benchmark (about $25-75bn) implies much more avoidance, but it imports estate-tax enforcement.

Evidence. scripts/wealth_tax_check.py on full2018det.xlsx. Book Ch. 8 ('about 1% of GDP each year'); Fig. 9.2 (1.2% of NI). PWBM (Dec 12, 2019); Bruelhart et al. (2022, AEJ:EP); Kopczuk (BPEA 2019).

Why this severity. Under defensible avoidance assumptions the policy score is uncertain by roughly -20% to -50%. This is secondary to the book's empirical core.

The authors' response. Saez & Zucman, 'Response to Summers and Sarin' (June 25, 2019); SZ BPEA (2019); 'How would a progressive wealth tax work?' (Feb 2019): avoidance is a policy choice (comprehensive base, third-party reporting, exit tax).

10moderatelikely · previously known · overclaiming

The book's recommended top tax rate of about 75% follows from one particular estimate of how much the rich change their behaviour. Other credible estimates imply recommendations anywhere from the mid-40s to the low 80s, so the 75% figure is far less certain than the book suggests.

Technical detail and evidence

Why it matters, in detail. With a=1.5 and e between 0.15 and 0.8, the revenue-maximizing rate ranges from 82% down to 45%. At a 75% rate, the incremental revenue (3.95% of NI) falls to about 2.7% with e=0.4 or 1.5% with e=0.57 (Codex).

Evidence. Book Ch. 7 'Taxing the rich' and notes 3, 8, 9 and 16; Diamond & Saez (2011, JEP); Neisser (2021, EJ); Badel, Huggett & Luo (2020, EJ). Arithmetic verified: 72.7%; (2/3)x75 + (1/3)x30 = 60%; about 3.95% of NI.

Why this severity. Affects normative policy calibrations only; the core empirical findings are unaffected.

The authors' response. Book Ch. 7 conditions the 75% on curbing avoidance first ('Hiking top tax rates without any other change to the tax code or to enforcement would be a bad idea').

11minorconfirmed · previously known · documentation gap

The 'bottom half' in the charts leaves out the poorest tenth of adults. Including them would actually make the working class's tax rate look higher, so the exclusion does not help the authors' case, but readers are not told clearly who is in the group.

Technical detail and evidence

Why it matters, in detail. The full-population bottom-50% rate (book vintage) is 24.97%, vs 24.21% restricted. The restriction partly substitutes for the transfer/consumption-tax issue in F4.

Evidence. tax-by-gperc.do lines 14-17 and 52. TC8/full2018: 216.69m retained adults vs 243.02m in TA1 col. 20 for 2018. Book Ch. 1 and n. 9; Online Appendix p. 5. Public file: 23.3m adults (9.6%) excluded (Codex; adjudicator port).

Why this severity. A labelling and coverage problem disclosed in a note. Removing the restriction raises the bottom-50% rate (+0.6-0.8 pp), so the restriction is conservative for C1.

The authors' response. Online Appendix p. 5 and n. 9: adults with very low pre-tax income live on transfer income, on which sales taxes produce very high rates; the restriction avoids that problem.

12minorconfirmed · newly identified here · internal inconsistency

About a third of the '$100 billion' from the proposed minimum tax on multinationals would come from raising taxes on profits US companies already report at home, not from tax havens.

Technical detail and evidence

Why it matters, in detail. The offshore-only yield, $61.5bn in 2016, is about two-thirds of the headline; the stateless treatment moves it by up to about +/-$15bn. The 2018 projection is $66.5bn at 4% growth vs $69.1bn at 6%.

Evidence. SZ2019AppendixTables TD3: rows 17, 28, 30, 32 and 34, cell C36. Online Appendix p. 28. Book Ch. 6 'Each country must police its multinationals'. IRS 2016 CbC Table 1A reproduced cell for cell (Codex).

Why this severity. The domestic component is logically part of a country-by-country minimum (the book's 'seven points, from 20% to 27%' refers to global profits). This is a framing and documentation issue in a secondary estimate.

The authors' response. Online Appendix pp. 28-29: the aggregate computation is a lower bound because effective tax rates vary across firms within each country.

13minorconfirmed · previously known · misrepresentation of source

A few vivid statements about tax havens do not match the government data the book cites. For instance, in 2016 US firms did not report more profit in Bermuda and Ireland than in Britain, Japan, France and Mexico combined.

Technical detail and evidence

Why it matters, in detail. The 'Bermuland' claim fails by $13bn on its own cited source. The stateless line partly duplicates profits reported in the owners' jurisdictions. Treating booked profits as shifted overstates US shifting by an unknown margin.

Evidence. IRS SOI 16it01acbc Table 1A (profit before income tax), rows 8, 40, 56, 87, 122, 128 and 151. IRS Instructions for Form 8975 (Revised 12/2020): fiscally transparent U.S. business entities 'should be reported on one Schedule A ... that indicates the tax jurisdiction "stateless"', and the owner aggregation rule. Foley et al. (2020, Tax Notes Int'l 99(9):1163-65). Book Ch. 4 'Forty percent of multinational profits shifted to tax havens' and n. 10. Fuest et al. (2022).

Why this severity. These are rhetorical overstatements in a supporting chapter. The core facts about corporate tax and havens (large haven profits, low tax rates there) stand.

14minorconfirmed · newly identified here · overclaiming

The book treats employer health-insurance premiums as a hidden tax and says insurance is compulsory, but the penalty for being uninsured had been cut to zero by 2019. That framing drives the claim that the system is 'massively regressive'; it does not affect the main tax figures.

Technical detail and evidence

Why it matters, in detail. P50-60: 25.4% in taxes plus 13.0 pp in premiums = 38.4%.

Evidence. Book Ch. 5 'Health insurance: a big but hidden labor tax'; Ch. 9 'Private health insurance: a huge poll tax' and Fig. 9.1; MainFigures DataF9.

Why this severity. A disclosed framing with an international-comparability rationale, used only in a secondary figure; the taxes-only profile is unaffected.

The authors' response. Book Ch. 5: even if conservatives succeed in weakening the mandate, 'the situation would not fundamentally change'. The notes to Fig. 9.1 call the premiums 'quasi-mandatory'.

15minorconfirmed · newly identified here · documentation gap

Outside researchers cannot recreate the headline numbers from the files the authors made public; with those files, billionaires come out slightly above the working class.

Technical detail and evidence

Why it matters, in detail. Deltas computed on the public file match both auditors' sensitivities to within 0.01 pp, which supports using the file for sensitivity analysis but not for levels.

Evidence. readme.txt; import-dina.do. scripts/check_corp_keys.py: in 2018, corptax = corptax100 (corr. 1.00000 with C-corp equity + 0.52 x pension wealth) and corptax0 has corr. 0.99917 with non-housing wealth; in 1962, corptax has corr. 0.99999 with non-housing wealth. scripts/toi_port.py: top 400 23.71% vs bottom 50% 23.39%. Book Conclusion.

Why this severity. Standard confidentiality limits plus an undisclosed vintage mismatch: this impedes verification but does not by itself bias the results.

The authors' response. Online Appendix pp. 29-30 and readme: the online files are coarser, and complete external files can be shared with researchers who have PUF access.

16minorconfirmed · newly identified here · misrepresentation of source

The book says nobody had measured all taxes together since 1985. In fact a research group whose data the book itself uses publishes such estimates every year, and under different conventions it finds the system moderately progressive.

Technical detail and evidence

Why it matters, in detail. ITEP's more progressive profile reflects different conventions: transfers counted in income, credits netted, and narrower top groups.

Evidence. ITEP, 'Who Pays Taxes in America in 2019' (April 2019) and '... in 2018'; Book Ch. 1 and n. 7.

Why this severity. A misstatement about the literature. ITEP has no long series and no top-400 estimates, so the book's claim of novelty for the long-run series is fair.

17minorconfirmed · newly identified here · speculative assumption

The very high tax rates shown for the richest in the early 1950s are estimates projected backwards from 1962, not measurements.

Technical detail and evidence

Why it matters, in detail. The 1950 top-400 value cannot be verified without micro data; the upgrade factors add several points to early-1950s top rates by construction.

Evidence. SZ2019AppendixTables TB4: cols O-Q formulas referencing taxrates_wide(_internal) row 1962 and TA3 ratios; cols T-V upgrade factors. Online Appendix pp. 25-26. Book Ch. 7.

Why this severity. The method is disclosed in the appendix. It affects only the earliest benchmark years, not 1962-2018.

The authors' response. Online Appendix pp. 25-26: macro conditions and inequality were stable over 1950-62, so the approximation is accurate to a first order.

18minorconfirmed · previously known · selective presentation

Some comparisons pick starting points that make trends look larger. In the book's own figures, working-class tax rates are no higher today than in 1980; the change at the top is what drives the story.

Technical detail and evidence

Why it matters, in detail. Between 1980 and 2018 the bottom-50% rate fell 1.5 pp in TB1 (and rose only 0.5 pp in the Feb 2022 full-population series), so the post-1980 'crossover' story is driven entirely by the top.

Evidence. MainFigures DataF5b; Appendix Tables A1 and B1 (col D). Book Introduction ('The Triumph of Injustice'); Ch. 1 'Plutocracy'; Ch. 5. PSZ update notes (Feb 2022). Corpus us-wealth-taxes-06 and -08; us-income-08 to -11.

Why this severity. Framing and endpoint choices; the qualitative messages (rising concentration; bottom rates higher than in the 1950s) are unaffected.

19minorconfirmed · newly identified here · arithmetic or transcription

One line (a small estate-tax item) is missing from a table in the book, so its numbers do not add up to the printed total.

Technical detail and evidence

Why it matters, in detail. The archived figure restores the missing row and reconciles the total.

Evidence. Book Fig. 9.2 (ebook-derived PDF page image, visually checked). MainFigures sheet F9.2. TablesFigures/PDF/SZ2019MainFigures.pdf, Figure 9.2.

Why this severity. A presentation omission of 0.1% of NI; the package total is correct.

20minorconfirmed · newly identified here · documentation gap

Two of the authors' own files give somewhat different sizes for the proposed national income tax base, and the difference is not explained.

Technical detail and evidence

Why it matters, in detail. Base ratios differ by 3.7 points of NI; at 6%, revenue differs by 0.22 points of NI.

Evidence. SZ2019AppendixTables TD4 (D5, D13:D23); taxsimulator.xlsx calculator!AV8:AV22 (Codex); Book Ch. 9 'Funding the social state in the twenty-first century'.

Why this severity. About 0.22% of NI (~$40bn) at common scale, between files for different years; documentation, not an established overestimate.

21minorconfirmed · newly identified here · arithmetic or transcription

The book contains a handful of small factual and arithmetic mistakes; none of them changes its conclusions.

Technical detail and evidence

Why it matters, in detail. All are presentation-level; (a) reverses a stated condition, but its numerical example (23% of compensation) is correct.

Evidence. Book Chs. 1, 5, 6, 8 and 9 (quoted passages); Facebook 2019 DEF 14A beneficial-ownership table (12,178,053 Class A + 365,723,786 Class B shares; 2,404,077,866 A + 452,157,146 B outstanding); DataF1b; TA1; tax-by-gperc.do.

Why this severity. Small errors that do not change any conclusion.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

The replication micro-files are absent from the archive (MicroFiles/Online contains only a .DS_Store). — import-dina.do downloads PSZ2017Dinafiles.zip automatically, and readme.txt documents this (the point Codex made). The substantive replication limits are kept in F15.
The appendix's statement that 2018 top-400 wealth is 'exactly the same as the Forbes estimates, $7.2 billion' is a product of the hard-coded doubling. — Not established. The doubled external record holds $4.27bn per equal-split adult after the 0.825 scaling (about $5.2bn before it), not $7.2bn. TA1/TA2's 2018 top-400 wealth ($6.75bn per adult) is extrapolated from 2016 internal-file shares and is the likelier source of the figure. What is confirmed is narrower and kept in F7: the 2018 external data were explicitly scaled 'to match internal files and Forbes', so the 2018 agreement is not independent validation.
The 45% fiscal-income cap was estimated against wealth x macro return but applied to DINA income (~9.7% of wealth), which undermines the top-400 calibration. — Rejected as a bias against the authors. Applying the wealth-based calibration consistently would lower top-400 fiscal income and the top-400 rate, strengthening C1. The 9.7% also uses post-0.825 wealth (8.0% before), and BSYZ (2025) find a 42% ratio. Kept only as a caveat in F7.
In a like-for-like public-file replication, shareholder-only incidence accounts for about half of the 1962-2018 top-400 decline (69.4% vs 44.0% in 1962). — The decomposition is not relied on: the port's 1962 TOI-key level (69.4%) is about 15 points above the published 54.4%, so it is not like-for-like. The 'about half' conclusion is retained in F6 on other evidence, namely SZ (2026)'s own comparison for the top 1% (-13 vs -7 pp over 1950-2021).
Netting all refundable credits gives a bottom-50% rate of about 19.6%. — Downgraded as the primary estimate. The public-file dicred total ($144bn) exceeds the authors' own EITC+CTC total ($91.7bn) and may include other credits, so F1 uses the authors' 21.0%.
The population restriction (bottom groups exclude ~1/10 of adults) is a moderate issue. — Downgraded to minor (F11). It is disclosed in n. 9 and the appendix, and it lowers rather than raises the bottom-50% rate, so it cannot explain the crossover.
The unreconciled national-income-tax bases are a moderate documentation gap. — Downgraded to minor (F20): about 0.22% of NI, between files for different years, with no demonstrated overstatement.
The remedial-tax composition (I11), the booked-vs-shifted conflation (I13) and the health-premium 'poll tax' (I14) are moderate issues. — Downgraded to minor (F12-F14): they concern secondary claims and framing and do not touch the core tax-rate findings.
Later administrative evidence (BSYZ 2025, 23.8%) materially corroborates the book's 2018 top-400 level (Codex checks 3 and 12; also Claude R17). — Qualified. The authors' own 2020 and 2022 DINA releases put 2018 at 27.2% and 29.3%. The corroboration is specific to BSYZ's method, aggregate-only (component shares differ substantially), and covers a 2018-20 average rather than 2018 (F5, F7).
Corpus critique (Kopczuk; Splinter, us-wealth-taxes-12, examined by both auditors): truncating adults below half the minimum wage pushes measured bottom-group tax rates up. — Rejected on the evidence. Including all adults raises the bottom-50% rate: +0.57 pp on the public file (adjudicator port; Codex +0.57), and 24.97% vs 24.21% in the book-vintage series. The truncation is a labelling issue (F11), not a source of bias toward the headline.
Codex leads it rejected itself: (a) the absence of the Forbes simulation Stata program is a replication failure; (b) the 'beyond Laffer' language implies that 10% is past the revenue peak of the wealth tax. — Agree with Codex. (a) The Forbes histories are in the website bundle, and Codex's independent recursion reproduces the published wealth shares and examples closely. (b) On the wealth-tax-only grid, 10% is near the revenue peak; this is not a confirmed error.

Audit notes

What reproduced. REPRODUCED. (1) The headline 2018 rates (top 400 23.04%, bottom 50% 24.21%, gap 1.17 pp) recompute exactly from the authors' Stata outputs (full2018.xlsx -> Table B1). (2) An independent port of the authors' Stata code on the public 2018 micro-file reproduces both auditors' numbers and confirms the sensitivities: refundable credits -4.64 pp (all) or -3.18 pp (EITC+CTC, authors' Jan-2020 simulator: 21.04%); payroll add-back -1.66 pp; no pure gains +2.25 pp on the published bins (top 400 25.29%); sales taxes on transfers -0.8 to -1.2 pp; PSZ-key or 25%-to-labor corporate incidence -1.6/-1.3 pp for the top 400; full population +0.6-0.8 pp for the bottom half. (3) The wealth-tax base and revenue, the optimal-tax arithmetic, Table A5's labor/capital rates, the remedial-tax 2016 base and the IRS CbC transcriptions reproduce. NOT REPRODUCED / REVISED. (4) The headline ordering is not reproduced on the public micro-file (top 400 23.71% > bottom 50% 23.39%), and the public 1962 file uses a different corporate-incidence key. (5) With current BEA data, the 'almost half' fall in federal corporate receipts becomes -2.3%, and 2018 corporate taxes are 29-35% higher than the book's input. (6) The authors' own later DINA releases (Sept 2020, Feb 2022) give 2018 top-400 rates of 27.2% and 29.3% vs 25.4% and 25.7% for the bottom half, with no crossover in any year through 2019; their 2025 administrative estimate is 23.8% (2018-20). (7) The historical decline in top rates is 2.5-5 times larger in the book than in the authors' QJE 2018 series for the same years (top 0.1%, 1962-2014: -17.0 vs -3.8 pp). The external DINA files and IRS internal files behind the published levels are not public, so the levels themselves could be checked only against the authors' outputs.

Limits of this audit. The authors' external DINA files and IRS internal files are not public. All sensitivities were run on the coarsened public 2018 file (records aggregated by five), and deltas were applied to the published values; levels on the public file differ by up to ~0.8 pp at the bottom. The sales-taxes-on-transfers adjustment (F4) uses a proportional split by the transfer share of disposable cash income and is approximate. Later-vintage comparisons (F5) are made at the full-population level published by the authors, while the book's sample excludes adults below half the minimum wage; restricted-sample equivalents are inferred from the book-vintage gap (~0.7 pp). BSYZ (2025) report only a 2018-20 average and no bottom-50% rate, so the administrative evidence cannot settle the 2018 ordering. The current-vintage corporate adjustment holds pre-tax income fixed. The 1950-61 top-group rates could not be checked without micro data. The book was read in a third-party ebook-derived PDF (191 pages), so locations are given by chapter, section and figure rather than print pages. The French and German translations were not audited. Several checks reported by the auditors (TPC statutory rates; Codex's Forbes-history recursion; the IRS CbC cell-by-cell match) were not independently re-run. Claims about the inherited income and wealth series (C8) were not re-audited here; they are covered by separate audits of PSZ (2018) and Saez-Zucman (2016).