The Rise of Income and Wealth Inequality in America: Evidence from Distributional Macroeconomic Accounts

Why it matters, in detail. C3 rests on the 2018 top-400 effective tax rate in the October-2019 (TOI) vintage: 23.0% against 27.7% for all included adults on the Figure-5 basis (pre-tax income plus pure capital gains), a 4.7-point gap (5.3 points against the 28.3% macro rate). The authors' September-2020 DINA outputs, the vintage behind Figures 1-3 and 6 and the abstract's 29% macro rate, give 27.2% against 28.3% (a 1.1-point gap), and the February-2022 outputs give 29.3% against 28.7%, i.e. the top 400 above average; in Sept 2020 the top 0.001% (29.8%) is also above average. The mechanism is the upward revision of corporate plus business-property taxes allocated to the top 400 (10.0%, 13.6% and 15.2% of their income across the three vintages), consistent with the revision of 2018 corporate taxes from 1.25% to 1.60% of national income, while their capped individual income tax barely moves (9.2-9.5%); a ranking test on the Oct-2019 microfile rules out a selection artefact. Figure 5 thus uses the vintage with the largest gap, and the abstract pairs it with the later vintage's 29% macro rate; on the authors' own later data the 'below the macro rate' gap shrinks to about a quarter of its size or reverses, and only the decline of rates within the top 0.01% ('regressive at the very top end') holds in all three vintages. Because administrative evidence (Balkir et al. 2025: 23.8% for the top 0.0002% against 30.2% for the population in 2018-20) supports the qualitative claim, the issue weakens the paper's demonstration rather than refuting C3.

Evidence. Paper: JEP abstract; Figure 5 (p. 20, source 'Saez and Zucman 2019b'); p. 21. Oct-2019 vintage (TOI tables, SZ2019AppendixTables.xlsx): TC8(2018) row 152 (0.2304) and row 155 (0.2770); taxrates_wide 2018 taxT400 = 0.2304, corpT400 = 0.1002; DataAF7 2018 corporate tax = 0.0125. Sept-2020 vintage (PSZ2020 Appendix Tables II, archive linked from gabriel-zucman.eu/usdina): sheet 'taxrates' 2018 taxtop400 = 0.2723, taxall = 0.2834, taxtop0p001 = 0.2981, corptaxtop400 = 0.0721; TG1 2018 col L = 0.2981. PSZ2020 Appendix I TSA12b 2018 col E = 0.0160. Feb-2022 vintage (PSZ2022 Appendix Tables II): sheet 'taxrates' 2018 taxtop400 = 0.2931, taxall = 0.2874. Code, read as text: outsheet_dina.do (full-population tax-rate section, uses ditax_fixed400 and a peinc_kg denominator) and taxrates.do. Scripts: top400_vintages.py, top400_ranking_test.py, adjudication_results.py.

Why this severity. In the vintage the paper uses for its income and wealth figures and for the abstract's macro rate, the gap between the top 400 and the average falls by about 80% (from 4.7 to 1.1 points on a comparable basis), and it reverses in the next vintage. The paper's own evidence for C3 is therefore fragile. Later administrative evidence (Balkir, Saez, Yagan & Zucman 2025: 23.8% vs 30.2% in 2018-20) supports the qualitative claim, so this weakens the paper's demonstration rather than refuting C3.

The authors' response. None found on this point. Later work by the authors with co-authors (Balkir, Saez, Yagan & Zucman 2025, NBER w34170, Table 2) matches the Forbes 400 to IRS data. It finds 23.8% for the top 0.0002% in 2018-20, against 30.2% for the whole population, and about 30% for both groups in 2010-17.

Why it matters, in detail. About 40% of national income is missing from individual tax returns and is allocated with partly unvalidated keys (underreported business income and the noncorporate capital-consumption adjustment in proportion to positive reported income; retained earnings split 50/50 by dividends and realized gains, which the paper calls 'far from ideal'; pension keys revised in Sept 2020 after the Auten-Splinter critique). Auten & Splinter (2024) find a 1979-2019 rise in the top-1% pretax share of 4.4 points (9.4% to 13.8%), against about 8.0 points (10.8% to 18.7%) in the Sept-2020 PSZ series, i.e. roughly half the rise; their 2014 level gap of 6.0 points splits into underreporting 2.0, retirement income 1.0, other taxes 0.7, corporate income 0.7 and tax-return corrections 0.4. They also find bottom-50% real pretax income up by more than a third over 1979-2019, against +9.5% in PSZ, so bottom-half growth would be several times higher than in C4. Independent assessments put the truth in between: Gale, Sabelhaus & Thorpe (2023) find that PSZ's choices push top shares up and AS's push them down, and Iselin & Reck (2025) place PSZ's misreporting treatment near the top of the plausible range but find AS's implied decline in relative misreporting not credible; a midpoint reading would still cut the top-1% rise by about a quarter. The sign of C1 is not in doubt; confidence is only plausible because the dispute is unresolved and rests on confidential data, and the JEP neither engages the pre-tax critique nor reports uncertainty ranges.

Evidence. JEP pp. 13-14, 20 and fn. 9. PSZ2020 Appendix II: TB1 col G (1979 0.1077, 2019 0.1873); TB3 col D (bottom 50%: $18,017 in 1979, $19,731 in 2019). Auten & Splinter (2024), JPE 132(7): p. 2182 (+4.4 pp; bottom 50% up by more than a third), Table 4 (p. 2210), Table 5 (p. 2216). Iselin & Reck (2025), working paper, abstract. Gale, Sabelhaus & Thorpe (2023), Brookings. PSZUpdates Sept 2020, s.1.2.1 (pension revision, -0.6 pt in 2016).

Why this severity. If AS's allocations were right, the 1979-2019 rise in the top-1% share would be roughly halved and bottom-half growth would be several times higher. A reading midway between the two series would still cut the rise by about a quarter. The direction of C1 is not in doubt.

The authors' response. Saez & Zucman (2020, NBER w27921, s.3) argue that AS treat legally exempt income such as bonus depreciation as evasion and misread the random-audit evidence. They accepted the pension-rollover point and corrected it in Sept 2020. Piketty, Saez & Zucman (2024; 2026 comments) attribute 41% of the 1979-2019 trend gap to the allocation of partnership excess depreciation and S-corporation exempt income, and they contest AS's linked-owner estimates.

Why it matters, in detail. The text's C4 statistics (12.5% share in 2018; $17,500 to $18,500 per adult, 'only 0.2 percent' a year; post-tax +25%; transfers offsetting 'roughly one-third' of the growth gap; disposable cash income +16% over 1980-2016) match the Oct-2019 vintage (12.67% share, about 0.21% a year), not the Sept-2020 series plotted in Figures 3 and 6. On the plotted data the 2018 share is 13.40% and pre-tax income rises from $17,246 to $19,586 (+13.6%, 0.34% a year), so the text understates bottom-half growth by more than half; post-tax income rises 27.0%, so transfers offset 23.7% of the gap with the +70.1% average rather than a third; disposable cash income rises 17.6% to 2016 and 24.8% to 2018; and Medicare and Medicaid account for 57% of the post-tax gain to 2016 but only 46% to 2018. Part of the 2016-18 gain is a projection seam (bottom-50% share +0.5 point and bottom-50% income of the 65+ +42% in 2017, in both the Sept-2020 and Feb-2022 vintages); with a 2016 endpoint the Sept-2020 data give +6.2% (0.17% a year) and a 31% offset, and WID (Oct 2026) gives +8.3% to 2016 and +15.4% to 2018. Near-stagnation far below the 70% average therefore holds in every vintage; what fails is the precision and internal consistency of the stated numbers, which keeps this moderate rather than major.

Evidence. JEP pp. 14-15, 22-23; Figure 6 (p. 24). Workbook DataF3(Income): G111 = 0.13400; memo col I (TOI vintage) = 0.12669. Workbook DataF6 (1980 / 2016 / 2018): pre-tax C23, C59, C61 = 17,246 / 18,317 / 19,586; post-tax D23, D61 = 22,231 / 28,223 (1980 / 2018); disposable cash B23, B59, B61 = 15,656 / 18,405 / 19,545. PSZ2020 Appendix II: TB1; TB3; TB7b (65+ bottom-50% income $19,421 in 2016, $27,660 in 2017); TC3b (Medicare+Medicaid $1,189 / $4,264 / $3,927 in 1980 / 2016 / 2018). WID bulk data, Oct 2026: aptincj992 p0p50. Script: adjudication_results.py.

Why this severity. The inconsistency is confirmed. On the paper's own plotted data, the stated bottom-half growth is understated by more than half and the 'one-third' offset is overstated. But near-stagnation far below the 70% average holds in every vintage, and part of the 2016-18 jump is a projection seam, which keeps this below major. Claude rated it major; Codex rated it moderate.

Why it matters, in detail. The p. 15 counterfactual is meant to show the change in each group's average pre-tax income under equally shared 1980-2018 growth, which requires income-weighting subgroup changes. DataAF8 instead takes unweighted means of decile percentage changes for the bottom 50% and P50-90 (C4 = AVERAGE(B4:B8)) and population-weighted means for P90-99 and the top 1%, on TOI's restricted, gains-inclusive population. Income-weighting the sheet's own inputs gives +50.8%, +13.5%, -9.4% and -47.9% (-50.1% with the top-400 row), and full-adult Sept-2020 shares (s1980/s2018 - 1) give +49.8%, +12.4%, -12.0% and -45.3%, against the stated +57/+16/-8/-36%. The stated figures thus overstate the bottom-50% gain by about 6-7 points and understate the top-1% loss by about 9-14 points, with the middle groups off by 1-4 points. The direction and ordering of the redistribution, and so C6's message, are unchanged; the error is confined to an explicitly illustrative, secondary result.

Evidence. JEP p. 15. Workbook DataAF8(Equitable): formulas in C4, C9, C13, C15; inputs D4:D18 equal TOI TC8(2018) C138:C152. PSZ2020 Appendix II TB1, cols C, D, L, G. Script: adjudication_results.py.

Why this severity. The illustration's numbers are off by 6-14 points, most of all for the top 1% (-36% against -45% to -50%). It is an explicitly illustrative, secondary result, and the direction of the redistribution is unchanged.

Why it matters, in detail. Footnote 3 and Figure 1 are the paper's main external validation (C5), but the claimed near-identity with the Fed DFA is partly mechanical or parameter-driven. (a) The SZ top-1% share (tax units, Figure 1b) rises 7.6 points over 1989-2018 (30.30% to 37.91%), not 'also' 10 points like the adjusted DFA (+10.0). (b) The SZ top-10% values for 2017-19 are the 2016 value grown at the DFA's own rate; with the authors' published Sept-2020 value for 2018 (75.70% rather than 77.55%) the 2018 gap is 2.3 points rather than 0.5, and the 1989-2018 rise is +7.6 for SZ against +9.5 for the DFA. (c) The household-to-tax-unit conversion assumes a Pareto alpha of about 1.3, whereas the DFA's own tail implies about 1.5-1.64, which would put the DFA's tax-unit top-1% share at 39.9-40.7% rather than an 'identical' 38%. (d) The Sept-2020 revision calibrated top-1% housing and pension wealth to the SCF and DFA, so part of the agreement is built in. (e) The current DFA vintage, harmonized identically, still gives a consistent top-1% share (36.2-38.3%) but a top-0.1% share of 14.7-15.6% against 19.4% for SZ. In (a)-(c) the corrected DFA is at least as concentrated, or rises at least as fast, as SZ, so they do not suggest that SZ overstate the top-1% level; what fails is the claimed closeness of levels and trends and, with current Fed data, the agreement at the top 0.1%.

Evidence. JEP p. 9; fn. 3 (pp. 8-9); Figure 1 (p. 10). Workbook DataF1-F2(Wealth): I89 = 0.3030, I118 = 0.3791, R89 = 0.2810, R118 = 0.3814; formulas in H117-H119; header H7. PSZ2020 Appendix II TE9: tax units, 2018 = 0.7570. DFA, 22-Sep-2020 vintage, 2018Q3: top 1% 32.39%, top 10% 69.86%. DFA 2026Q2 detail file, 2018Q3: script dfa_harmonize.py. PSZUpdates Sept 2020, s.1.1.6-1.1.7.

Why this severity. This is the paper's main validation argument: that official data 'paint the same picture'. Agreement on the top-1% level survives. But the claimed near-identity of levels and trends is partly mechanical or depends on a parameter choice, and at the top 0.1% it does not hold with current Fed data.

Why it matters, in detail. The p. 7 statement that the Sept-2020 revisions 'do not significantly affect the level of top wealth shares nor their trend' fails on the authors' own before/after series (July-2019 internal estimates, close to the Oct-2019 TOI series, against Sept 2020): the 1989-2016 rise in the top-1% share (tax units) falls from +11.42 to +8.49 points (-26%), the 1989-2016 rise in the top-0.1% share (equal-split) from +8.13 to +6.35 (-22%), and single-year levels move by about 3 points (top 1%, tax units: 27.2% to 30.3% in 1989; 39.4% to 36.4% in 2010); the pension revision alone lowers recent top-1% shares by 1-1.5 points. The statement holds only for 2016 levels and the abstract's 1978-2016 window (top-0.1% rise +12.09 to +11.34, -6%), so C2's headline is barely affected. But the post-1989 trend, the window used for the DFA comparison behind C5, was cut by a fifth to a quarter, an instability of that size that the paper describes as insignificant.

Evidence. JEP p. 7. Workbook DataF1-F2(Wealth): cols C vs I, D vs J, G vs M (rows 78, 89, 114, 116) and row 124. TOI TA1 wealth shares. PSZUpdates Sept 2020, s.1.1.7.

Why this severity. The abstract's C2 window is barely affected. But the revision cut the post-1989 trend, which is the window the DFA comparison uses, by a fifth to a quarter, and the paper describes this instability as insignificant.

Why it matters, in detail. Every 2018 number in the JEP (C1-C4) is a provisional projection: per the Sept-2020 Updates note, 2017 is built from the 2012 public-use file scaled component by component to SOI aggregates, and 2018 grows each component with nominal GDP and recomputes taxes with TAXSIM; the JEP gives no caveat. The projection leaves seams between 2016 and 2017 (bottom-50% income share +0.5 point; SZ top-10% wealth share -1.6 points, smoothed away with DFA growth, see F5; bottom-50% income of the 65+ +42%), and the bottom-50% seams feed into the C4 growth figures (F3). For C3 the projection was decisive at the time of writing: in the TOI series behind Figure 5 the top 400 paid more than average in every observed year 2013-2016 (2016: 30.7% against 29.9%), and below-average rates appear only in the projected post-TCJA years 2017-18. The damage is limited ex post: the Feb-2022 vintage confirms the 2018 values for C1 and C2 (19.26% and 18.04%), and later administrative work (Balkir et al. 2025) supports C3.

Evidence. PSZUpdates_Sept2020, s.1 and s.2.3. TOI tables TB1, cols B and K, 2010-2018. PSZ2020 Appendix II: TB1, TB7b, TE9.

Why this severity. This is an important undisclosed caveat on every 2018 number, and at the time of writing it decided the sign of C3. The damage is limited: later vintages confirm the 2018 values for C1 and C2 (19.26% and 18.04% in Feb 2022), and later administrative work supports C3.

The authors' response. The Updates note (Sept 2020, s.2.3) documents the projection method and promises revisions as new data arrive. Balkir et al. (2025, w34170) later found, with administrative data, a top-400 rate of 23.8% against 30.2% for 2018-20.

Why it matters, in detail. Figure 5 is computed on a different basis from the one the JEP describes. Its denominator adds 'pure' realized capital gains (gains above 3% of national income: 2.05% of national income, or 40.6% of realized gains, in 2018). On the stated national-income basis the 2018 top-400 rate is 25.8% rather than 23.0%, the comparison average is the 28.3% macro rate rather than the Figure-5 'All' rate of 27.7%, and the top-400 shortfall against the macro rate halves from 5.3 to 2.5 points but stays positive. The population also excludes adults with pre-tax income below half the annual minimum wage (26.3 million, 10.8% of adults in 2018), so 'P0-10' is roughly P11-P20. Including them raises the bottom-decile rate from 24.6% to 49.6% in the Oct-2019 microfile (mean income about $1,900, mean taxes about $1,000), so the truncation does not favour the authors' conclusion, but it misdescribes the population and hides how sensitive bottom-group rates are to the treatment of transfers. Net effect on C3: in the vintage the paper uses, its own stated definition roughly halves the top-400 gap without reversing it.

Evidence. JEP pp. 18-20 and the Figure 5 note. Workbook DataF5(Taxes) B7-B8. TOI tables: TC8(2018) rows 152, 155, 157 and M152; TA3 K114; TA1 U115. TOI appendix p. 5 and pp. 9-11. outsheet_dina.do: 'keep if peinc > minwage/2'. Scripts: adjudication_results.py, microfile_check.py.

Why this severity. On the stated basis the C3 gap halves (from 5.3 to 2.5 points) but stays positive. The bottom-decile label is wrong, though correcting it would not help critics of the top-400 comparison.

The authors' response. The TOI appendix (pp. 5, 9-11) and the Updates note (s.2.1.4) document and justify both choices: adding pure gains aligns capital-gains taxes with their base, and very low-income adults have unstable tax-to-income ratios. The JEP does not carry either definition over.

Why it matters, in detail. The sign of the top-400 result depends on one triangulated parameter: top400.do caps the top 400's fiscal income at 45% of their pre-tax income and scales their income taxes down accordingly, and the variables needed to check it (top400, ditax_fixed400, fiinc_fixed400) are dropped from the public files. In the Oct-2019 online microfile the top 400's raw fiscal/pre-tax income ratio is 0.94; uncapped, their 2018 rate is 30.1%, above the 27.7% average, against 21-23% with the cap; caps of 30%, 60% and 70% give 20.0%, 26.1% and 28.2%, and the break-even cap is 70.6%. The public file reproduces published 2018 statistics only approximately (top-1% income share 20.05% vs 20.53%; bottom-50% share 13.43% vs 12.67%; top-0.1% wealth share 20.82% vs 19.25%), and several article numbers (pre-1962 group tax rates, the 2017-19 SZ top-10% wealth values, the hard-coded DFA and SCF series, the people's-growth constant) cannot be reproduced from it at all, so the p. 4 claim that the microfiles reproduce 'all our findings' is incorrect and C3 is not externally verifiable. The parameter is supported ex post: Balkir et al. (2025) find fiscal income of about 40% of economic income for the top 400, comfortably below the 70.6% break-even in the Oct-2019 file, which keeps this moderate rather than major.

Evidence. JEP p. 4. Code, read as text (Sept 2020): top400.do, outsheet_dina.do. TOI appendix pp. 21, 26-29. PSZUpdates s.2.1.3. TOI TA1, 2018. Scripts: top400_ranking_test.py, microfile_check.py, microfile_shares_frac.py.

Why this severity. The abstract's third headline cannot be verified from public data, and its sign depends on one triangulated parameter. It stays moderate rather than major because later administrative data support the cap: Balkir et al. (2025) find fiscal income of about 40% of economic income for the top 400.

The authors' response. The Updates note (s.2.1.3) says the 45% estimate 'could be refined' and that linking the Forbes 400 to tax data would be desirable. Balkir et al. (2025) later did this.

Why it matters, in detail. The 'collapse of progressivity' part of C3 (1950 rates of 'around 25 percent' for P90-99 and 'almost 70 percent' for the top 0.01%) rests on unobserved pre-1962 group rates, built by moving 1962 rates by income bin and tax type with each tax's macro trend and then adjusting top-group corporate rates to match PSZ Table II-G1, and on assigning 100% of the corporate tax to shareholders, a departure from agency practice and from PSZ (2018) that critics show drives much of the measured decline at the very top. Whole-distribution measures of federal progressivity rise after 1979 (Splinter 2020). The size of the measured 1950-2018 decline at the top therefore depends on an unobserved backcast and on an incidence convention that critics show enlarges it. The 2018 top-400 comparison is not exposed to this concern, since alternative incidence assumptions would lower the top-400 rate further.

Evidence. JEP pp. 19-21; Figure 5. Workbook DataF5(Taxes) B10:B24. TOI appendix pp. 25-26 (notes to Appendix Table B4). PSZUpdates s.2.1.2 and s.2.2. Kopczuk (2019), BPEA comment, pp. 523-524. Splinter (2020), NTJ 73(4).

Why this severity. The size of the 1950-2018 decline at the top depends on the backcast and on the incidence convention. The 2018 top-400 comparison does not: alternative incidence assumptions would lower the top-400 rate further.

The authors' response. Saez & Zucman ('Clarifying Distributional Tax Incidence', 2019; NBER w31912, 2023) argue that statutory incidence of the corporate tax on shareholders is the right convention for measuring current burdens.

Why it matters, in detail. C2's level and rise sit at the top of the range of credible estimates. Smith, Zidar & Zwick (2023) put the top-0.1% share at 15.73% in 2016 against 18.42% for SZ (equal-split adults), with a 1978-2016 rise of +8.59 against +11.34 points (24% smaller) and a 1989-2016 rise of +4.63 against +6.35 (16.53% in their Forbes-replacement specification); the current Fed DFA harmonized as in fn. 3 gives 14.7-15.6% for tax units in 2018 against 19.36%; the adjusted SCF in Figure 1c gives 17.30% against 19.79% (2016, tax units); and the authors' own mortality-adjusted estate series rises 7.9 points over 1976-2012 against 12.6 for the capitalized series. If these alternatives are right, the 2016-18 level is 15-24% lower and the rise since 1978 about a quarter smaller, with the direction unchanged. The p. 11 'lower bounds' framing holds only for one-sided omissions (offshore wealth, conservative valuation of private business); allocation and capitalization errors are two-sided, as p. 9 concedes ('might overestimate the rise'), and no uncertainty range is reported. SZ's rebuttals of SZZ (implausibly high top interest rates, missing billionaire equity) narrow the gap but do not close it, and the remaining disagreement, concentrated in the top 0.01%, is still contested.

Evidence. JEP pp. 9, 11; fn. 1; Figure 1c. SZZ supplemental data, TotalWealthShare.xlsx: Baseline col E; Spec 11 col C. Workbook DataF1-F2(Wealth): cols J, M, AN, AJ, AT. DFA 2026Q2 detail file: script dfa_harmonize.py. Smith, Zidar & Zwick (2023), QJE 138(1), Table I, p. 550.

Why this severity. If the alternatives are right, the 2016-18 level is 15-24% lower and the rise since 1978 about a quarter smaller, with the direction unchanged. SZ's critique of SZZ (missing billionaire equity) narrows the gap but does not close it.

The authors' response. Fn. 1 and Saez & Zucman (2020, NBER w27921, s.2): SZZ assume implausibly high top interest rates and capture only 57% of Forbes billionaire wealth. Saez & Zucman (2022, NBER w30396): dividend-based capitalization understates billionaire equity by a factor of about 2 (SEC filings), and the remaining disagreement is concentrated in the top 0.01%. SZZ's 2021-22 replies dispute this.

Why it matters, in detail. The abstract's 29% macro rate is the Sept-2020 aggregate (29.01%), while the body and Figure 5 use the Oct-2019 value (28.30%), so the abstract sets the Oct-2019 top-400 rate (23%) against a Sept-2020 average, widening the apparent gap by about 0.7 point. The JEP's 'federal corporate tax ... 1.5 percent of national income' in 2018, which was '1 percent' in both working papers, matches the authors' Sept-2020 total corporate tax (1.53-1.60%) rather than the federal component (1.19%), and the change is undisclosed. Each discrepancy is below a point and neither changes the direction of C3; both are symptoms of the vintage mixing behind F1.

Evidence. Abstract (AEA page; w27922 p. 1). JEP pp. 18-19 and 21. w27922 and WIL WP 2020/20: the same sentence with '1 percent'. PSZ2020 Appendix I: TSA12b 2018 (B = 0.2901, E = 0.0160, P = 0.0153); TSA12c 2018 (D = 0.0119). Workbook DataAF7, 2018: B = 0.2830, D = 0.0125.

Why this severity. These are inconsistencies of about 1 point that do not change the direction of C3. They are symptoms of the vintage mixing behind F1.

Why it matters, in detail. C5's 'years of average income' statistic overstates the 1980-2020 rise: the plotted series gives 70.0 years for the top 1% in 1980 (60.8 is the 1978 value), so the increase to about 200 years in 2020 is 2.85x rather than the implied 3.3x. The 2020 points for SZ and the DFA are 2019 shares carried forward and multiplied by a provisional 2020 wealth/income ratio (5.691), and 'highest level ever recorded' for the 570% ratio depends on that provisional value, since 1933 (5.46) exceeds 2019 (5.13) in TB2. The Figure 2 note says the DFA top-1% households are given the same average wealth as the top 1% of adults, whereas the formula equates shares, which differs by a factor of about 1.9. The qualitative point, a near-tripling of top-1% wealth in years of average income, survives.

Evidence. JEP p. 11; Figure 2 and its note (p. 12). Workbook DataF1-F2(Wealth): BB78 = 60.75; BB80 = 69.998; AY120 = AY119; AW120 = AW119; AX120 = 5.691; BA = AW*AX*100. PSZ2020 Appendix I TB2 col B: 1933 = 5.456; 2019 = 5.128; 2020 = 5.691.

Why this severity. The qualitative point survives: years of average income still nearly triple (2.85x rather than 3.3x).

Why it matters, in detail. Since the Sept-2020 update the series is forced to match Forbes at the very top (top-400 wealth in 1982-2005, billionaire wealth from 2006), so the p. 9 statement that Forbes data 'confirm the surge of wealth concentration seen in tax data' partly compares Forbes with itself; the top 400 alone rise from 1.0% to 3.3% of wealth over 1982-2018. The p. 5 validation of the basic capitalization method rests on one year (2016: $3.1tn against $3.0tn for Forbes), while in the same workbook basic-method billionaire wealth is 49% of Forbes in 2000 and 162% in 2010, so the match is a one-year coincidence rather than evidence of accuracy at the top. The effect on C2 and C5 is small, because the independent Forbes evidence in panel D still shows a large rise at the very top and the paper concedes the basic method's flaws elsewhere.

Evidence. JEP pp. 5, 7, 9. PSZUpdates s.1.1.4. Workbook DataF1(Forbes): cols K, P, AB, AC (rows for 2000, 2010, 2016).

Why this severity. The independent Forbes evidence in panel D still shows a large rise at the very top, and the paper concedes the basic method's flaws elsewhere.

The authors' response. The paper concedes that Forbes is 'far from an ideal data source' (p. 9). Saez & Zucman (2022, w30396) use SEC ownership filings as an independent check on billionaire equity.

Why it matters, in detail. Figure 4's 1946 percentile incomes are imputed as each percentile's 1962 level divided by its group's 1946-62 growth factor (H5 = I5/J$142), so every bottom-90% percentile grows at 2.45% a year over 1946-62 by construction, i.e. for 16 of the 34 years of the 'evenly distributed' period, and neither the note nor the text says so. The flatness of the 1946-80 profile across percentiles is therefore partly imposed rather than measured. C6's qualitative contrast survives on group-level data: observed 1962-80 growth was broadly even below the top 1% (p0-10 2.56%, p10-50 1.77%, p50-90 1.86%, p90-99 1.70%, against 0.53% for the top 1%).

Evidence. JEP p. 15; Figure 4 and its note (p. 16). Workbook DataF4(Growth): formulas in H5:H134; rows 141-155.

Why this severity. Group-level data support the qualitative claim; only the flatness across percentiles in 1946-62 is assumed.

Why it matters, in detail. People's growth (the mean of percentile growth rates in C6) is hard-coded at 0.0065 rather than computed. Because p0-p3 have non-positive incomes in 1980 or 2018 their growth rates are undefined, and the 1980-2018 mean depends on the convention: 0.61% with undefined rates set to 0, 0.63% over p4-p99, 0.70% over p5-p99 and 0.83% over p10-p99; for 1946-80 the value is 2.16%, against the text's 2.0%. The stated 0.65% lies inside the reproducible range, and under every convention people's growth stays far below the 1.4% macro growth rate, so C6's contrast is unaffected.

Evidence. JEP p. 15. Workbook DataF4(Growth): F5:F134 = 0.0065. Script: peoples_growth.py.

Why this severity. The stated value lies inside the reproducible range, and under every convention it stays well below 1.4% macro growth.

Why it matters, in detail. The p. 17 argument that BEA personal income 'double counts some forms of income' rests partly on the premise that it 'does not subtract payroll taxes or individual income taxes'. NIPA personal income is defined net of contributions for government social insurance, i.e. payroll taxes, so only the income-tax half holds. No estimate depends on this: it removes one premise of a side argument for the national-income concept behind C1, whose other justifications (international comparability, neutrality to business legal form) stand.

Evidence. JEP p. 17. BEA NIPA Handbook, ch. 2, p. 2-13: 'plus current transfer receipts less contributions for government social insurance (domestic)'.

Why this severity. No number depends on it. It removes one premise of a side argument for the income concept behind C1.

Why it matters, in detail. Six small slips, none affecting a core estimate: the government share of unfunded pension liabilities is 91.2% at end-2018 ($590bn private out of $6,704bn; 97.1% at end-2019), not '99 percent'; 'Federal Reserve Distributional National Accounts' should read Financial Accounts; the SNA treatment of durables is cited to Semega et al. (2019), a Census income and poverty report (the working papers cite an unlisted 'United Nations, 2009'); 22.5% × 2 × $53,000 = $23,850, not 'about $23,400'; the top 0.00001% is 18.5 and 11.1 tax units, not '17 ... and 10'; and the abstract's 18% top-0.1% share is for equal-split adults while Figure 1 shows tax units (19.4%), without the unit being stated. Each is small and none changes a conclusion.

Evidence. Z.1, 21 Sep 2020, tables L.117-L.120: FL573073005 = 590.0; FL593073005 = 6,703.7. JEP pp. 8-10, 15. w27922 p. 6. Workbook DataF1(Forbes) col W; DataF1-F2(Wealth) J118 and M118.

Why this severity. Each slip is small and none changes a conclusion.

Why it matters, in detail. Footnote 9 attributes Auten and Splinter's low post-tax top shares mainly to their per-capita allocation of half of government consumption. AS's published decomposition (JPE 2024, Table 4) attributes 6.0 of the 6.6-point 2014 post-tax gap to pre-tax differences and 1.3 points to government consumption, and for the 1960-2019 trend Gale, Sabelhaus & Thorpe (2023) attribute 2 of 5 points to pre-tax differences and 3 to tax-and-transfer treatment as a whole. The footnote therefore understates how much of the disagreement lies in the pre-tax allocations discussed in F2, which the JEP does not engage. It concerns the characterization of a critic rather than the paper's own estimates, and the December-2019 AS draft the JEP actually cites was not checked.

Evidence. JEP fn. 9 (p. 22). Auten & Splinter (2024), JPE, Table 4 and p. 2211. Gale, Sabelhaus & Thorpe (2023), Brookings. Saez & Zucman (2020), NBER w27921, pp. 32-33.

Why this severity. It characterizes a critic's results, not the paper's own estimates.