← All papers
Emmanuel Saez · AEA Papers and Proceedings 111: 1-26 (AEA Distinguished Lecture)
Public Economics and Inequality: Uncovering Our Social Nature
15 problems found: 3 moderate · 12 minor
What the paper claims
- Humans' social nature, absent from the standard self-interested model, is crucial to understanding large modern social states and pervasive inequality concerns: childcare/education, retirement, health care and income support are resolved at the group level, and government redistribution is primarily in kind or targeted rather than the across-the-board redistribution of utilitarian optimal-tax theory; prehistoric pooling of about 50% makes the social state a 'homecoming'.
- Rich countries pool 30-50% of national income through taxes, up from under 10% before World War I, and the twentieth-century growth of government is 'almost entirely' due to the social state (education, retirement, health care, income support).
- In both the US and France the tax system is 'approximately a proportional tax' (French rates 'almost 20 points higher ... throughout the distribution'), so government reduces inequality mainly through spending; posttax top-10% shares are substantially below pretax shares.
- Pretax inequality is shaped by politics and institutions: the US top-10% pretax share collapsed in WWII, rose from 1980 and 'has now reverted back to its pre-World War II levels', whereas France's rebound was much smaller; pretax redistribution (minimum wage) is far more popular with voters than tax redistribution.
- Mass education is always government driven (public funding plus compulsion), and compulsory schooling produces large enrollment increases within a few years.
- Labour supply is in large part a social choice made through social-state regulation and norms: France-US employment differences are concentrated among the young and old and are 'not driven by the taxes needed to fund the larger social state'; old-age employment tracks pension rules; female work tracks norms (WWII, 1990s welfare reform); French hours fell with paid-vacation and working-time rules.
- Normative implication: revealed individual preferences need not reflect social preferences and behavioural responses depend on social context, so social systems should be designed to eliminate behavioural elasticities rather than take them as given.
Every problem found
1moderateconfirmed · newly identified here · misrepresentation of source
The chart says it compares the same tax measure in both countries, but the French line comes from a different calculation (working-age people only, measured against a broader income concept). When the same source's comparable numbers are used, richer people pay a clearly larger share of income in tax in both countries, so the claim that the tax systems are 'approximately proportional' does not hold as stated. The lecture's wider point that public spending does much of the equalising is not overturned, but this piece of support for it is weaker than presented.
Technical detail and evidence
Why it matters, in detail. The plotted French line puts contributive SSC (large, roughly proportional to capped wages) in the numerator and factor income in the denominator for a 25-60 working sample; the US line uses pretax income (already net of contributions and inclusive of benefits) for all adults. Under the source's DINA-consistent convention the French schedule is steeply progressive (P99.5-99.9/P10-20 = 1.84 vs 1.12 for the plotted P99-99.9/P10-20) and so is the US one (top 0.1%/P10-20 = 2.24); the France-US gap shrinks from 20-27 pp to 10-22 pp. No published French series on an 'all taxes / pretax income / all adults' basis exists, so the lecture's convention cannot be applied like-for-like; the direction of the correction is unambiguous for the 'same methodology' and 'proportional' statements.
Evidence. Workbook data!H7:J21. Independent 300-dpi digitization of Bozio et al. (2020, WIL WP 2020/22) Appendix Fig 6a (PDF p. 87): lecture values match within 0.06 pp for 13 of 15 groups (P99-99.9 = mean of the 99 and 995 ticks; P99.99-99.999 differs by 0.43 pp) - scripts/digitize_af6a.py. Source definitions: Bozio et al. pp. 8-9, 16, 29-31, fn 41, fn 46 ('One alternative way ... distribution of factor income among working age adults ... all taxes, including contributive taxes'); datafile 2 sheet DataSeries1970_2018 cols I-M (Fig 9) and BB (US memo contributive SSC). Auditor A found the match first (digitize_bozio_af6a.py); Auditor B noted only that 'almost 20 points' is loose (gaps 13.3-27.2 pp).
Why this severity. It reverses the qualitative description of the tax side of the paper's opening exhibit for France (proportional -> clearly progressive on the source's comparable basis), weakening C3's step 'taxes are proportional, spending does the redistributing'; it leaves the social-state thesis, Fig 2 and the pretax-vs-posttax comparison intact.
2moderateconfirmed · newly identified here · misrepresentation of source
The chart dates Indonesia's compulsory-schooling law to the late 1970s, but the national programme began in 1984; the big jump in enrollment came earlier, when the government built tens of thousands of schools. So the example shows the power of state school-building rather than of compulsion. The Russian line also leaves out several of the source's data points without saying so.
Technical detail and evidence
Why it matters, in detail. The rise the figure attributes to compulsion in Indonesia (72% -> 88%, 1975-80) precedes the 1984 programme by 4-9 years; after 1984 enrollment rises only 88% -> 99% from a high base. The Russian omission does not favour the lecture under its own (1918) dating: the source shows an even faster rise (72% by 1925, 100% by 1930), although the 100% values (including wartime 1945) look capped; the issue is that the plotted path is an undisclosed interpolation over four omitted observations.
Evidence. Workbook data!AG7:AJ27 vs LeeLee_enroll_MF.xls v1.0 (scripts/check_fig3b.py: every plotted value identical; 4 Russian observations 1925-45 and 8 after 1950 blanked). Flag positions read from a 300-dpi render of p. 8. Sukmayadi & Yahya (2020), J. Social Studies Education Research 11(4): 219-234, pp. 221-222 ('In 1984 the government established a six-year compulsory education policy'; 61,000 schools 1973-78); Kompas (2 May 2021) and Tirto.id on the 2 May 1984 launch. RSFSR Regulation on the Unified Labour School (16 Oct 1918), Art. 4, istmat.org/node/31601; USSR decree 'On universal compulsory primary education', 14 Aug 1930.
Why this severity. One of the three cases offered for the figure's stated causal conclusion is factually wrong, so the compulsion-specific claim rests on Russia and, ambiguously, Korea; the broader claim that mass education is government-driven survives because Indonesia's surge was state-built schooling. Low end of moderate.
3moderateplausible · previously known · external inconsistency
Other researchers using the same tax records find that the rise in the top 10%'s share of US income since 1980 is about 40% smaller than in the figures the lecture uses. If they are right, the claim that US inequality is back at its pre-WWII peak is not established, though everyone agrees inequality rose. The debate is unsettled and the lecture's main argument does not hinge on it.
Technical detail and evidence
Why it matters, in detail. Change in the US top-10% pretax share 1980-2018: +7.0 pp (AS) vs +11.6 pp (PSZ); 2018 level 39.8% vs 44.9%. No AS-method French series exists, so the US-France comparison cannot be redone like-for-like; the pre-war benchmark is itself contested in the opposite direction (Geloso et al.), so the net effect on 'reverted to pre-war levels' is indeterminate. Gale, Sabelhaus & Thorpe (2023) side with PSZ on several allocation items.
Evidence. AutenSplinter-IncomeIneq-2024.xlsx, sheet F-A1, column 'Pre-tax Income' (1980: 0.32805; 2018: 0.39783) vs workbook data!C77 and C115 (0.33340, 0.44909); scripts/check_fig1a.py for the PSZ range. Critique corpus us-income-07, -08, -09, -10, -13, -21.
Why this severity. If AS are right, the lecture's motivating fact (the US post-1980 rise) is about 40% smaller and 'back to pre-war levels' is not established; but the dispute is unresolved, every series shows a sizeable rise, and the lecture's social-state thesis does not depend on the magnitude.
The authors' response. Saez & Zucman (2020) and PSZ's later comments contest AS's allocations, especially underreported income (SZ 2020 argue about 3 of AS's 7.2 points of national income belong to the top 1% vs AS's 0.6) and government consumption ('neutral' proportional allocation); see corpus us-income-08/-10/-13 author-response fields. No response specific to this lecture was found.
4minorconfirmed · previously known · methodology
For the US, the 'flat' tax picture depends on choices: counting Social Security payroll taxes as taxes, counting tax credits for low-wage workers as spending rather than tax cuts, and leaving out the lowest-income ninth of adults. With other common choices US taxes rise clearly with income. These are standard choices in this research programme, but the lecture does not flag them.
Technical detail and evidence
Why it matters, in detail. Ratio of P80-90 to P10-20 rates: 1.23 (PSZ convention), 2.08 (credits netted), 1.45 (contributory SSC excluded), 3.78 (both); rates above P90 are essentially unchanged. Combined with F1, C3's 'approximately proportional in both countries' holds only when contributory social-insurance contributions are counted as taxes and refundable credits as spending.
Evidence. scripts/us_taxprofile_micro.py (microfile /mnt/zdata/downloads/psz2022-upper-tail/usdina2018-sept2020.dta; outsheet_dina.do read as text, section 'TAXES BY G-PERCENTILE OF PRE-TAX INCOME'); scripts/check_fig1b_us.py (TG1). Auditor B computed the same diagnostic (excluded share 9.06%; credit-netted bottom rates 16.3%/14.1%) but did not report it as an issue; Auditor A flagged the typed top cells.
Why this severity. The conventions are PSZ/SZ's documented national-accounts choices and the substantive point (redistribution runs through targeted transfers, which include refundable credits) is unaffected; what is missing is disclosure that the US 'proportional' label is convention-specific and that the lowest-income adults are excluded.
The authors' response. Zucman's position (as reported by Gleckman, TaxVox, Oct 2019) is that refundable credits are spending, no different from SNAP; SZ distribute national-accounts concepts in which transfers are not negative taxes (corpus us-wealth-taxes-12). No defence of the minimum-wage truncation was located.
5minorconfirmed · newly identified here · documentation gap
The 'over 50 percent by 1880' figure for Black children's school enrollment is not a measured number: it comes from scaling Census figures in a way the paper does not explain, and direct census-based estimates are much lower (roughly a quarter to a third). The underlying point - enrollment jumped very fast once slavery and its school bans ended - still stands.
Technical detail and evidence
Why it matters, in detail. Plotted Black level 1880: 53.0% vs 33.8% raw (5-19 base); Margo's Southern census estimates are about 21% at ages 5-20 in 1880 (roughly 25% at 5-14 if the 1890 age profile applies) and about 36% at 5-14 in 1890 vs 58.8% plotted; the Black-to-total ratio is preserved by construction, so the figure's racial gap in relative terms is census-consistent while absolute levels inherit Lindert's gross-ratio inflation.
Evidence. Workbook data!AA7:AE17; Lindert (2004) Appendix Table A1 Panel B (rows 'USA, pub + priv', 'USA, pub only', Prussia, Germany, Norway); HSUS (1975) pp. 369-370, series H433/H435 and headnote (read from page images); Margo (1990), Race and Schooling in the South, ch. 2, Table 2.2 (NBER chapter c8792). Auditor B: scripts/reproduce.py (exact reconstruction); Auditor A: check_fig3a.py.
Why this severity. The undisclosed rescaling inflates a quoted level, but the conclusion drawn from it - a very rapid post-emancipation rise driven by the end of legal prohibition - holds in the raw Census data (1.9% in 1860 -> 9.9% in 1870 -> 33.8% in 1880), and no other result depends on the 50% threshold.
6minorconfirmed · newly identified here · version instability
The US 'before tax' and 'after tax' lines come from two different editions of the data. Mixing them makes US government redistribution look about a tenth smaller than either edition would show on its own, which slightly understates rather than helps the lecture's point.
Technical detail and evidence
Why it matters, in detail. 2018 US top-10% reduction from pretax to posttax: 6.46 pp plotted vs 7.11/7.32 pp in either consistent source, i.e. measured US redistribution understated by 9-12%; levels, trends and the France-US ordering are unchanged.
Evidence. scripts/check_fig1a.py: workbook data!C7:D117 vs PSZ2020AppendixTablesII(Distrib).xlsx TB1/TC1 col E and Bozio et al. datafile 2 DataSeries1900_2018 cols G and M.
Why this severity. Small and conservative: the mismatched pair understates the 2018 US pre-post gap (6.46 pp vs 7.11 pp with a consistent PSZ pair or 7.32 pp with Bozio's), so no conclusion changes.
7minorconfirmed · previously known · documentation gap
The chart's notes describe 'before tax' income as excluding all government transfers, but it already includes public pensions and unemployment benefits, and education spending is spread in proportion to income rather than per person. Both choices make government look less equalising than it is, so the slip does not help the lecture's argument.
Technical detail and evidence
Why it matters, in detail. The plotted gap excludes redistribution through public pensions/UI (US factor vs pretax top-10% share 2018: 45.33% vs 44.91%) and treats education as neutral; with lump-sum allocation of all in-kind spending the French 2018 reduction would be 9.9 pp instead of 7.1 pp (US: 8.8 vs 7.3 pp in Bozio's V1/V2).
Evidence. Bozio et al. (2020) pp. 8-9 and 16; datafile 2 DataSeries1900_2018 cols J, Q, W (France) and M, T, Z (US), 2018 row. PSZ microfile: US 2018 top-10% share 45.33% (factor income), 44.91% (pretax), 37.61% (posttax), 36.13% (posttax with education lump-sum) - scripts/us_top10_concepts_micro.py.
Why this severity. Labelling errors whose direction understates the social state's measured redistribution (public pensions are inside 'pretax', education is treated as neutral), so they cannot strengthen the lecture's claims.
8minorconfirmed · newly identified here · selective presentation
Several labels in the ballot chart show the wrong year, and two successful 'tax the rich' votes (Oregon 2010, Illinois 2014) are missing while a failed general tax increase is included. Fixing this raises the success rate of tax-the-rich measures from about a quarter to about a third, still far below the 93% for minimum-wage measures, though since 1996 the gap is smaller (about half versus 93%).
Technical detail and evidence
Why it matters, in detail. Pass rates: 6/23 (26%) as plotted; 8/25 (32%) adding OR M66 and IL 2014; 8/26 (31%) also adding CO 2013. On a common 1972 window minimum-wage measures are 26/28 (WA 1988 passed), so the unequal start dates do not create the contrast; since 1996 the tax measures pass 6/12 (50%), or 8/14 (57%) with the omissions, against 25/27 (93%).
Evidence. Published Fig 8 (p. 20, 300-dpi render); CRS R44706 (Bradley & Overbay, 28 Jan 2021), Table 1; Wikipedia '2014 United States ballot measures', '2013 United States ballot measures', '2010 Oregon Ballot Measures 66 and 67', 'List of Oregon ballot measures' (citing official returns) - scripts/misc_checks.py. The workbook states Fig 8 was produced separately in Stata; the AEA package (openICPSR 137121) was inaccessible.
Why this severity. Correcting the inventory moves the tax-the-rich pass rate from 26% to about 31-33% against 93% for minimum-wage measures, so the qualitative contrast - and the lecture's hedged wording ('consistent with the view') - survives.
9minorconfirmed · newly identified here · arithmetic or transcription
One sentence says only a quarter of older American men worked by 1930; the paper's own chart and its source say 58% - the quarter figure is for 1980. The chart also mixes two definitions of 'working' and rounds several French census years, which slightly exaggerates the US-France gap but not the overall decline.
Technical detail and evidence
Why it matters, in detail. Stated 1930 level 25% vs actual 58.0% (a 50-year dating error); US-France level gap partly definitional (gainful vs current definition: 35.2 vs 24.8 in 1970, 24.7 vs 19.9 in 1980); trend unaffected.
Evidence. Workbook data!AT7:AV31; Costa (1998), The Evolution of Retirement, ch. 2, text pp. 7-9 and Tables 2A.1-2A.2 (pp. 29-30); NBER WP 28387 text ('By 1930 the fraction of elderly working is only 25 percent').
Why this severity. A prose slip and labelling issues; the plotted US values match Costa exactly and both definitions show the same long-run decline, so the conclusion that old-age work fell with retirement systems is unaffected.
10minorconfirmed · newly identified here · misrepresentation of source
The lecture simplifies Germany's pension history ('retirement age cut from 65 to 60') when the 1972 reform mainly allowed retirement at 63, or at 60 for specific groups; it calls the US pattern stable although older men's employment fell by about 23 points from 1970 to 1994; and it says women's market work is now only slightly below men's although half of employed Dutch women work under 30 hours a week. These are loose descriptions, not errors that change the message that pension rules and norms shape who works.
Technical detail and evidence
Why it matters, in detail. The relevant German exit ages after 1972 were 63 (long service) and 60 (women, unemployed, disabled), not a universal 65->60 cut, and Boersch-Supan & Schnabel stress non-actuarial benefits (an implicit tax on work) - a financial-incentive channel the lecture folds into 'social choice'. The US decline of 22.6 pp (1970-94) is smaller than Germany's (-39 pp) or France's (-40 pp from 1975), so 'flatter' holds, but the stated explanation is incomplete.
Evidence. Boersch-Supan & Schnabel (1999), in Gruber & Wise (eds), Social Security and Retirement around the World, pp. 147-150 and Table 4.1; OECD LFS EMP_RATIO and DSD_HW@DF_EMP_USL_WK_HRS (downloaded 2026-10-26) - scripts/check_oecd_spot.py, scripts/misc_checks.py.
Why this severity. Loose descriptions; the cited source itself attributes the post-1972 German decline to the flexible-retirement provisions, and the plotted employment series (exact OECD matches) support the general point that pension rules shape old-age work.
11minorconfirmed · newly identified here · documentation gap
The wartime part of the chart on women's work is measured for ages 14-64 rather than the labelled 16-64. The wartime jump is visible within the consistently measured years, so the conclusion is unchanged.
Technical detail and evidence
Why it matters, in detail. Two-year difference in the lower age bound around the 1947 break; the 1940-45 rise (+8.4 pp between the 1940-41 and 1943-45 averages) is measured on one basis.
Evidence. Workbook data!BK7:BL32; HSUS (1975) pp. 131-132; reconstruction by Auditor B (scripts/reproduce.py) re-checked arithmetically (scripts/misc_checks.py).
Why this severity. The wartime surge occurs within the consistent pre-1947 regime, so the illustration is unaffected.
12minorconfirmed · newly identified here · arithmetic or transcription
A handful of small slips - a wartime percentage, the year France added a fourth week of paid vacation (1969, not 1968), the claim that no country gives more than 1% of income in foreign aid (Sweden and Norway have), and calling France's tax level 'stable' after 1980 - none of which affects the lecture's argument.
Technical detail and evidence
Why it matters, in detail. Magnitudes: +8.4 pp vs '10 points'; one-year misdating; ODA up to 1.40% vs 'below 1%'; France +7.9 pp after 1980 vs 'stable'; hours-gap divergence 7%->21% (lecture vintage) or 9%->24% (current) - qualitatively identical.
Evidence. scripts/misc_checks.py (Fig 6B, Fig 7, ODA, hours); Legifrance JORFTEXT000000511693 (loi 69-434); OECD DAC1 and DSD_HW downloads (2026-10-26); workbook data!L18:P23.
Why this severity. Each is a small slip in an illustrative lecture; none changes a qualitative conclusion (e.g. cross-country aid remains tiny relative to domestic pooling even at 1.4% of GNI).
13minorlikely · previously known · speculative assumption
The long-run chart of government spending is an approximate reconstruction - its own author calls the early breakdown approximate - and for 1920-1970 it shows more social spending than standard historical data. The main point, that nearly all of the growth of government since 1900 is social spending, holds either way.
Technical detail and evidence
Why it matters, in detail. Mid-century social spending is overstated relative to Lindert/OECD by factors of about 1.3-3.2, and the 0.9->0.85 conversion switch raises post-2010 levels by ~6%; neither affects the growth decomposition or the 30-50% range.
Evidence. Piketty (2020) Chapter10TablesFigures.xlsx, DataF10.14 formulas and DataF10.15 notes A1, A36-A41; workbook identical to Piketty in 79/79 (2A) and 180/180 (2B) cells; OWID 'Public social spending as share of GDP (Lindert 2004; OECD)' - scripts/misc_checks.py.
Why this severity. The 'almost entirely social' conclusion uses endpoints where the construction agrees with independent benchmarks (89% for 1870-2015, 93% for 1910-2015), so only the mid-century path and some levels are uncertain.
14minorlikely · newly identified here · internal inconsistency
The 'about 50%' estimate of how much hunter-gatherers pooled adds the value of communally owned land on top of shared hunting, even though food gathered from that land is kept within families. Counted consistently the figure is nearer one-third - still within the 30-50% range of modern states, so the comparison survives.
Technical detail and evidence
Why it matters, in detail. Pooled share = alpha + (1 - alpha)h = 50% treats communal ownership as pooling of the land rent; if pooling refers to the output flow shared beyond the family (the modern comparator is taxes/national income), the consistent value is h = 1/3 when hunting's output share equals its work share.
Evidence. Text p. 5 and footnotes 5-6; arithmetic 0.25 + 0.75/3 = 0.50 vs 1/3 when land income follows the activity in which it is realised (Auditor A, check_text_claims.py; Auditor B's grid shows 20-70% across assumptions).
Why this severity. A self-described rough illustration; the consistent figure (~33%) still lies within the 30-50% range of modern social states, so the 'homecoming' comparison survives.
15minorplausible · previously known · overclaiming
A few strong cause-and-effect statements - that politics drove the US-France difference, that taxes do not explain France-US employment gaps, and that very high top tax rates prevented excessive pay - are supported by side-by-side charts or citations rather than tests. The lecture says its evidence is illustrative, so these are best read as hypotheses.
Technical detail and evidence
Why it matters, in detail. None of the cited exhibits identifies a causal effect; similar prime-age employment despite higher French taxes is informal evidence against large prime-age tax effects but not a test, and the paper partly concedes the tax channel for hours (p. 17). The top-rate claim is disputed (avoidance and few binding taxpayers per Splinter 2020; market-based explanations per Kaplan & Rauh 2013).
Evidence. Text pp. 1, 13, 21; Figures 1A, 4-7; Boersch-Supan & Schnabel (1999) p. 137 ('the dramatic decline after 1970 is at least partly due to the introduction of flexible retirement arrangements in 1972 that did not adjust benefits according to actuarial tables'); critique corpus us-income-14, optimal-tax-behavior-09, optimal-tax-behavior-11.
Why this severity. Interpretive statements in a lecture that declares its evidence 'voluntarily illustrative rather than comprehensive'; they are best read as hypotheses and do not alter any reported number.
Claims examined and rejected
Problems one auditor raised that did not survive checking.
- Figure 3B's Russia flag (~1920) is misdated because universal compulsory primary education was decreed only on 14 August 1930, and the omitted Lee & Lee values undermine the causal claim for Russia. — The RSFSR Regulation on the Unified Labour School (decree of 16 October 1918), Article 4, made school attendance compulsory for all children of school age; the ~1920 flag matches this legal adoption, and the 1930 decree implemented universal primary schooling. Under that dating the omitted source values (72.0% in 1925, 100% in 1930) show an even faster post-law rise, so the omission does not favour the lecture. Retained only as an undisclosed data choice inside F2.
- There are no posted data for Figure 8. — Partly rejected: the author-site workbook does lack Figure 8 data ('produced with STATA separately'), but the AEA replication package 'Data and Code for: Public Economics and Inequality: Uncovering Our Social Nature' (openICPSR 137121 V1; DataCite record verified) exists and returned HTTP 403 to all three reviewers, so the absence of the Figure 8 inputs cannot be established.
- The lecture misrepresents Kleven (2019) by omitting the macroeconomic boom that Kleven stresses as a confounder of the 1990s rise in single mothers' participation. — Kleven (NBER WP 26405) lists welfare reform, the booming macroeconomy and changing social norms as confounders and explicitly entertains a 'social multiplier' from norm change; the lecture's social-choice reading is one of the cited source's own interpretations, and its footnote 17 reports Kleven's finding that other EITC expansions had no visible effect. Not omitting the boom would not change the point.
- Under Auten & Splinter the US relative rise since 1980 (+21%) is close to France's (+17%), undermining the inference that political developments explain the US-France divergence. — Not like-for-like: it compares an AS-method US series (individuals ranked by size-adjusted income) with the Garbinti-Goupille-Piketty DINA French series; no AS-method French series exists, and in percentage points the US rise under AS (+7.0 pp) still exceeds France's (+4.4 pp from 1980, +4.8 pp from 1982). The AS critique is retained (F3) for the US level and trend only.
- The unequal ballot windows (minimum wage since 1996, tax-the-rich since 1972) inflate the Figure 8 contrast (gap 66.5 pp, vs 42.6 pp on a common 1996 window). — Qualified rather than accepted: the windows are disclosed, and on a common 1972 window the minimum-wage record is 26/28 (Washington 1988 passed), so the headline contrast is not created by the window choice; the 1996+ comparison shows instead that tax-the-rich measures became more successful over time (0/11 before 1996, 6/12 after). Reported as context in F8.
- Corpus critiques of the tax-unit Piketty-Saez series (us-income-09) and of the pension-rollover allocation in earlier PSZ vintages (us-income-11) apply to Figure 1 (lead examined and dismissed by Auditor B). — Agree with the dismissal: Figure 1 uses equal-split adults rather than tax units, and the September 2020 PSZ vintage it cites documents the rollover correction; neither critique implies a correction to the plotted shares.
Audit notes
What reproduced. Reproduced exactly from the cited sources: Fig 1A France pretax/posttax (105/105 years, Bozio et al. 2020 data file 2), US posttax (106/106, PSZ Sept-2020 TC1) and US pretax (104/104, but from the US column of Bozio et al.'s file, not the cited Sept-2020 TB1, which is up to 0.76 pp higher after 1962); Fig 2A (79/79 points, Piketty 2020 DataF10.14; the 2018 update equals tax/GDP divided by 0.85) and Fig 2B (180/180 cells, DataF10.15); Fig 3A Lindert values and Fig 3B Lee & Lee values (all plotted points; the Black series only after an undisclosed rescaling, and several Russian source observations are blanked); Fig 4 (52/52 cells), Fig 5B (145/145) and Fig 6A (US/France exact; Netherlands within 0.83 pp) against the current OECD LFS; Fig 5A US values (Costa 1998 Table 2A.1); Fig 6B pre-1947 points (1943 and 1946, on a 14-64 basis); Fig 7 gaps (7.1% and 20.6%; current OECD vintage 9.2% and 23.9%); Fig 8A count 25/27 (CRS R44706) and the checked vote shares. Fig 1B France matches Bozio et al. Appendix Fig 6a (factor income, workers 25-60) within 0.06 pp for 13 of 15 groups - i.e. it reproduces, but not as the concept stated. Fig 1B US was re-computed from the PSZ 2018 microfile within 0.42 pp on average over P0-P99 only with PSZ's half-minimum-wage sample cut; the two top cells are typed constants. Not reproduced or not verifiable: the AEA replication package (openICPSR 137121, HTTP 403) and therefore the Fig 8 inputs/Stata code; Fig 6C (Kleven) was not re-estimated; no French tax-rate series on an 'all taxes / pretax income / all adults' basis is published; the Fig 8 x-axis (Republican lean) was not rebuilt; the AS-vs-PSZ dispute was not adjudicated.
Limits of this audit. The AEA replication package (openICPSR 137121 V1) returned HTTP 403, so the Stata code and Fig 8 inputs were not seen; the author workbook holds values only. Ballotpedia was blocked; ballot results come from CRS R44706 and Wikipedia pages citing official returns, and the completeness of the tax-ballot inventory was checked only for the cases named. Bozio et al. Appendix Fig 6a was digitized from the PDF (precision ~0.05 pp); no French series on an 'all taxes / pretax / all adults' basis is published, so a fully like-for-like France-US comparison on the lecture's convention could not be computed. The PSZ microfile used (usdina2018, Sept-2020 vintage) lacks the top-400 adjustment variables, so rates for the top 0.001% were not reproduced. OECD and WID data are current vintages, not those used in 2020. HSUS pages are image-only and were read visually; Fig 6B inputs rely on Auditor B's transcription (re-checked arithmetically). The Auten-Splinter vs PSZ dispute and the Geloso et al. pre-war critique were reported, not resolved. The anthropological, behavioural and experimental literatures surveyed in Sections I.A and IV.A were not audited beyond the Engel (2011) statistics.