Earnings

Why it matters, in detail. Pre-tax labor income nets out social-insurance and pension contributions, which rose over time and are larger for men, and adds benefits split equally within couples, which shifts part of husbands' pensions and Social Security to wives. Both narrow the measured gap. The 2014 ratio is 1.757 on the plotted concept vs 1.792 (published factor labor income) or 1.809 (public-file wages + fringes + 70% of mixed income), i.e. 56.9 vs 55.3-55.8 cents. For medians the contribution wedge widens over time (factor vs pre-tax male median: 8.7% in 1964, 14.2% in 2014 in the micro-files). The plotted male median falls 0.3% over 1964-2014; on the stated concept it rises 9-11% (about 0.2%/yr). That is still stagnation, but not 'the same'.

Evidence. Chapter p. 20 (definition, key findings), Figs. 1-2. PSZ (2018) QJE 133(2), Figure VI note p. 591 ('Pretax labor income is factor labor income plus pensions, Social Security, and unemployment insurance benefits, minus the corresponding contributions. Pensions and Social Security benefits are split 50/50 between spouses') and Figure VII note p. 593. Adjudicator digitisation of the chapter's vector Figure 1: RMSE 0.020 against TF1 col [18] (pre-tax, ages 20-64) vs 0.097 against col [14] (factor, 20-64) (scripts/pdf_chart_dump.py, digitise_figures.py). TF1 (Nov 2017) 2014: col [18] 1.7568 (56.9 cents) vs col [14] 1.7920 (55.8 cents). Public DINA micro-file 2014 (Oct-2019 vintage), ages 20-64 incl. zeros: wages + fringes + 70% of mixed income (flemp+flmil) 1.809 (55.3 cents) vs plinc 1.760. Male median in 2014 $ (PSZ national-income deflator), micro-files: stated concept 36,718 (1964) to 40,605 (2014), +10.6%; factor +9.4%; pre-tax +2.8%; published TB13 (pre-tax) 35,796 to 35,700, -0.3% (output/micro_gender_stats.txt, medians_deflated.txt). PSZ2017Programs/build_usdina.do: plpco = -wagpen - oacont - oacont_se; ploco = -dicont - uicont - dicont_se; pensions and benefits halved for married couples; plinc = flinc + plcon + plbel. MainData Data!GT ('Memo: average pre-tax labor income of men 20-64 / women 20-64') feeds Fig. 1, and Data!FW-FY, labelled 'Median factor labor income', equal the pre-tax TB13 medians. Saez's companion slides (May 2018, slide 3) also define 'Earnings = wages earnings + payroll taxes + fringe benefits + self-employment income' while the chart is labelled 'average pre-tax labor income'.

Why this severity. The mislabel changes the headline only modestly (57 cents becomes 55.3-55.8; '75 percent higher' becomes 79-81%, about 5-7% of the premium), and it goes the way of a larger gap, so it does not flatter the thesis. But it is an undisclosed inconsistency in the chapter's central definition, and it turns the secondary claim that the male median was 'the same' in 1964 and 2014 into a rise of about 10%.

The authors' response. No response located. The concept is stated correctly in the notes to PSZ (2018) Figures VI-VII, but not in the chapter.

Why it matters, in detail. The chapter conflates s (women's share of earners) with r_f/r_m (the relative probability of earning). With women about 52% of adults, equal participation would put s near 0.52; the observed s = 0.468 (1993) and 0.4765 (2014) imply r_f/r_m = 0.82-0.84. In the CPS, the 1.17 employment-rate ratio (ages 25-64) accounts for about 29% of the log of the 1.75 comprehensive ratio, so participation is still a material component of C1. For C2 the benchmark barely matters: women's 16.5% share of the top 1% vs 47.65% of all earners (rather than 50%) is an under-representation factor of 2.9 instead of 3.0.

Evidence. Chapter p. 21 and key-findings box p. 20. TF1 col [2] (Nov 2017): 38.0% (1962), 46.2% (1990), 46.8% (1993), max 47.7% (2012), 47.65% (2014). Female population share p = (mean_men - mean_all)/(mean_men - mean_women) from TA5: 0.515 (1962), 0.518 (1993), 0.520 (2014); Census V2016: 51.4% of adults 20+ in 2014. r_f/r_m = [s/(1-s)][(1-p)/p] = 0.578 (1962), 0.818 (1993), 0.839 (2014) (output/claims_check.txt). BLS CPS labor-force participation, 16+, annual means of FRED LNS11300001/LNS11300002: men/women 82.0/37.9 (1962), 75.4/57.9 (1993), 69.2/57.0 (2014) (output/bls_deflator_checks.txt). CPS-ASEC 2014 earner-rate ratio, men/women aged 25-64: 1.17 (output/cps_pinc03_age.txt). PSZ (2018) p. 594: 'women are almost as likely to work as men today ... converged to close to 50% during the 1970s and 1980s: women have closed the participation gap'.

Why this severity. A substantive secondary claim is wrong on every standard measure: a 17-18-point participation gap remained in 1993. It also sits uneasily with the first key finding, which counts participation differences as part of the 57-cent gap. It does not change the C1 or C2 numbers, and the glass-ceiling contrast survives with the correct benchmark.

Why it matters, in detail. The 1960s baselines (3.7x, under 5% at the top) are model-based: CPS shares for 95% of couples plus 1969 tax tabulations carried back for the top 5%. The most fragile number is the 1962 top-0.1% female share: 2.04% (Nov 2017), 2.51% (Oct 2019) and 3.08% (Sep 2020, Feb 2022) in the published tables, and 0.94% in the public micro-file. The quoted changes are robust to about +/-1 point. The override lowers the 1960s female share in top cells; single filers are at most about 3% of the top 0.1%, so it can only slightly raise measured progress. The 1962 ratio is 3.63-3.71 across vintages.

Evidence. Chapter p. 20. PSZ (2018) QJE p. 590 (W-2 'from 1999 onward. Prior to 1999, we rely on IRS tabulations ... top 5% ... and on similar tabulations ... in the CPS for the bottom 95%'). PSZ Data Appendix (Nov 2017) section B.2, p. 21: W-2 'only since 1999 and a few years isolated years before 1999 (namely 1969, 1974, and 1982-6)'; 'Before 1979, we assume that the split of the labor portion of self-employment income is the same as in 1979'. Section A.2: impute.do 'imputes gender of single filers over the period 1962-1978'. PSZ2017Programs/sharefbuild.do: lines 8-9 ('TO DO3: CPS and tax data are not fully consistent for bottom 95% ... creating slight discontinuity in 1998 to 1999'); lines 159-160 ('secondary is always defined as the person with lowest wages so it introduces a bias downward for wives and secondary earnings are capped at 30K'); lines 445-449 (interpolation schedule); lines 581-589 ('excess number of women in 1969 for high wages (relative to 1974, not clear why, probably data issue)', then 1969 top-cell values replaced by 1974 values). TF1 (Nov 2017): pre-tax ratio 1.963 (1998), 2.013 (1999), 1.998 (2000); women's top-10% share 21.3%, 20.1%, 21.2%. Sep 2020: 1.946 to 1.947; Feb 2022: 1.940 to 1.940 (output/vintage_compare.txt). 1962 public file: singles are 7.8%/4.8%/3.1% of the top 10%/1%/0.1%; married women 1.72%/0.85%/0.21% (output/top_composition.txt).

Why this severity. The disclosure gap is real; the 1969 override is not documented even in PSZ's appendix. But its demonstrated effect on the numbers the chapter cites is small. The 2014 levels rest on W-2 data, and the 1999 seam moves the ratio by about 0.05 (2.5%) in one year. Extreme assumptions on single filers' gender bound women's 1962 top-1% share within [0.85%, 5.6%]. Across PSZ's own vintages and files, the 1962-2014 changes stay within +13.1 to +14.1 points (top 1%) and +9.0 to +10.9 points (top 0.1%). Downgraded from Auditor A's 'moderate' for this reason.

The authors' response. PSZ Data Appendix section B.2 states that the interpolation 'delivers results that are close to the results obtained for years when we have exact wage split (1969, 1974, and 1999-2014)'. No response on the 1969 override; the 1999 seam disappears in the Sep-2020 vintage, and PSZUpdates gives no stated reason.

Why it matters, in detail. Top pass-through income is male-dominated. Reclassifying it as labor raises men's top labor incomes if it is attributed to husbands (lowering women's shares), or raises wives' incomes if it is split like capital (raising them). The attribution rule matters more than the labor share. PSZ's own Sep-2020 revisions moved the 2014 top-0.1% share by +1.3 to +1.6 points. The contested scaling of mixed income for under-reporting (us-income-10) is bounded by the no-mixed-income case (17.8% of the top 1%).

Evidence. Adjudicator re-run (scripts/passthrough_check.py; Oct-2019 2014 file; scorinc is split equally across spouse records in all 21,047 married units). Women's shares of the top 1%/0.1%: baseline 15.88%/11.87%; +75% S-corp split 50/50, 18.03%/15.73%; all to husband, 14.85%/10.47%; by couple's mixed-income split, 16.34%/12.73%. Excluding mixed income entirely: 17.80%/12.18% (output/mixed_income_bound.txt). TF1 2014 by vintage: 16.46/11.49 (Nov 2017), 16.29/11.51 (Oct 2019), 17.30/12.82 (Sep 2020), 17.27/13.12 (Feb 2022). PSZUpdates section 2.2.2 (partnership mixed income 50% capital, sole proprietorships 20% capital); section 2.1.5 (heterogeneous S-corporation shares used for wealth: 'labor income is disregarded when estimating S-corporation wealth'). Feb-2022 build_usdina.do: flinc = flemp + flmil + flprl + flsubl, with flmil = flmil_s + flmil_p (no S-corporation labor component).

Why this severity. Plausible alternative classifications move the top-1% share by about +/-1-2 points and the top-0.1% share by -1.4 to +3.9 points. Women remain a small minority at the top under all of them, so the glass-ceiling conclusion is unaffected.

The authors' response. Saez & Zucman (2020), 'Trends in US Income and Wealth Inequality: Revising After the Revisionists', NBER WP 27921, section 4: about half of pass-through profit is capital income, not a quarter. PSZ Updates (Sep 2020, section 2.2.2) re-split mixed income (partnerships 50/50, sole proprietorships 80/20 labor/capital).

Why it matters, in detail. With PSZ's published 20-44 ratio of 1.60 (factor income; 1.54 pre-tax in the micro-file), a 20-34 ratio of 1.3 implies a 35-44 ratio of roughly 1.85-2.0, above the CPS's 1.73. The CPS 55-64 value (1.80) is below 'almost 2'. Concept differences could explain part of the gap: tax data capture top earners (mostly men aged 45-64) better, PSZ count incarcerated adults as zeros, and benefits are split within couples. The magnitudes therefore remain undocumented.

Evidence. PSZ Data Appendix (Nov 2017) Figure S.7 ('Average labor income of all men aged 20-64 / all women aged 20-64, by age group': All, 20 to 44, 45 to 64; 1979-2014). TF1 cols [15]-[16], 2014: 1.599 (20-44), 2.020 (45-64). PSZ codebook: age is 'restricted to three age categories'. CPS-ASEC 2015 PINC-03, 2014 earnings incl. non-earners, men/women mean per person: 1.56 (25-34), 1.73 (35-44), 1.87 (45-54), 1.80 (55-64); PINC-05: 1.36 (15-24) (output/cps_pinc03_age.txt, cps_pinc05_check.txt).

Why this severity. A secondary descriptive claim that is unverifiable rather than shown to be wrong. The qualitative pattern (the gap grows with age through mid-career) is corroborated.

Why it matters, in detail. Equal splitting averages couples' labor income, which raises the pooled median when wives earn little (1962: 21.6k vs 16.0k individualised; 2014 public file: 32.3k vs 28.1k). Plotting it next to individualised sex-specific medians makes the three lines inconsistent. The undisclosed Dec-2016 to Nov-2017 switch also leaves PSZ's own text inconsistent with its table.

Evidence. Micro-file recomputation in 2014 $ (output/medians_deflated.txt): equal-split pooled median 21,670 (1962), 23,102 (1964), 33,543 (1999) vs TB13 col [7] 21,588 / 23,144 / 33,611. Individualised pooled median 15,952 / 17,349 / 29,066. Dec-2016 TB13 col [7]: 15,556 (1962), 25,071 (1989), 27,600 (2014); Nov 2017: 21,588, 30,138, 31,450; men's and women's columns unchanged. Digitised Fig. 2 middle line: about 21,890 (1962) to 31,560 (2014), matching TB13 col [7] (RMSE about 300). PSZ (2018) p. 594: '$27,500' and '10% higher' than 1989 (Nov-2017 table: +4.4%). MainData Data!FW-FY labelled 'Median factor labor income' although they equal the pre-tax medians.

Why this severity. Presentation only: the text discusses only the men's and women's medians, which are correct. But the plotted pooled line sits about 35% above an individualised pooled median in 1962 and about 12-15% above in 2014.

Why it matters, in detail. Before 1999 the public files give every wife a tabulated share of the couple's self-employment income. This creates millions of tiny 'earners' and leaves very few wives at the top; the published pre-1979 computations evidently differ, contrary to the appendix. Post-1979 public files have about 3m too few working-age men. Outsiders therefore cannot recompute the chapter's series, which rest on confidential data.

Evidence. Adjudicator re-implementation of outsheet_dina.do lines 385-424 (cumul flinc if flinc>0; rank >= .9/.99/.999) on the Oct-2019 public files vs the Oct-2019 TF1. 1962: all earners 40.07% vs 38.01%; top 10% 4.73% vs 5.12%; top 1% 2.50% vs 3.15%; top 0.1% 0.94% vs 2.51%. 1964: 39.63/37.71, 3.53/4.74, 2.60/2.59, 0.92/2.55. 2014: 49.42/47.70, 29.82/27.21, 15.88/16.29, 11.87/11.51 (output/psz_fracfemale_exact.txt). In the 1962 file, 3.53m women (3.25m married) are earners only through mixed income (zero employee compensation; median factor labor income $263 nominal); excluding them gives 37.50% (output/zero_wage_women_1962.txt). PSZ Data Appendix A.3.3: 'Before 1979, there are no internal files so external and internal computations are identical before 1979'. Public 2014 files: ages 20-64 are 51.94% (Oct 2019) and 52.10% (Feb 2022) female vs 50.33% in Census V2016 (output/micro_gender_stats.txt, census_sex_share_2014.txt). TF1 1979 cols [2]-[12] identical to 1978 in the Nov-2017, Oct-2019, Sep-2020 and Feb-2022 tables.

Why this severity. This affects outside verifiability, not the published values. The chapter's numbers come from the internal-data tables, whose 2014 earner share (47.65%) is close to BLS (women were 47.4% of people who worked during 2014: 75.4m of 159.1m, Work Experience of the Population, Table 1). Public-file values differ by at most about 1.5 points at the top and 2 points for the all-earner share.

The authors' response. PSZ codebook (Feb 2022): the external micro-files 'do not exactly reproduce the results published in Piketty, Saez, and Zucman (2018) that were produced using more comprehensive internal-use tax data'. Nothing addresses the pre-1979 gender statistics.

Why it matters, in detail. On the full-time median measure, essentially all of the 1973-1993 narrowing happened after 1979 (gap 37.6% in 1979, 23.9% in 1993); the 1970s change was about 1.5 points. Some adjusted measures do show 1970s gains, so the statement is a common but inaccurate gloss of the series the chapter cites.

Evidence. OECD US gender wage gap, 2017 vintage (via Our World in Data, updated 17 Oct 2017): 38.1% (1973), 39.1 (1974), 37.6 (1975), 37.8 (1976), 38.2 (1977), 39.0 (1978), 37.6 (1979), 36.6 (1980), 28.5 (1990), 23.9 (1993), 18.9 (2015); 100/(1-gap) = 161.6%, 131.4%, 123.3%. Blau & Kahn (NBER WP 21913, section 2): 'After many years with a stable female/male earnings ratio of roughly 60%, women's relative wages began to rise sharply in the 1980s'. Kopczuk, Saez & Song (2010; NBER WP 13345) describe the literature as finding narrowing 'especially during the 1970s and 1980s', and Blau and Beller (1988), as summarised by Blau and Kahn, find 1970s gains on hours- and selection-adjusted measures.

Why this severity. A background statement about the conventional gap; it does not enter any of the chapter's own estimates.

Why it matters, in detail. Ratio-of-means decomposition (2014): fringes -0.04 (health premia are close to lump-sum per covered worker), 70% of mixed income +0.085, pre-tax netting -0.05; net about zero. Bridging the CPS (ages 15-64): 1.27 (FTYR median), 1.41 (median, all earners), 1.50 (mean, earners), 1.71 (mean incl. non-earners). The measure's novelty is conceptual (all adults, means), not a different empirical gap that requires tax data.

Evidence. Micro-files (output/micro_gender_stats.txt). 2014 Oct-2019 file: flwag 1.7645, flemp 1.7246, flemp+flmil 1.8090, flinc 1.8074, plinc 1.7598. 1999: 1.928, 1.888, 1.992, 1.991, 1.961. Feb-2022 2014 file: 1.834, 1.791, 1.857, 1.857, 1.792, so 'equal to a wages-only ratio' holds only approximately and is vintage-specific. CPS-ASEC 2015 tables PINC-05 and PINC-03 (output/cps_pinc05_check.txt, cps_pinc03_age.txt): men/women ages 15-64, full-time year-round median 1.27, median among earners 1.41, mean among earners 1.50, mean per person incl. non-earners 1.71; ages 25-64, 1.747. PSZ Data Appendix section B.3.3: fringe benefits imputed from CPS pension coverage, with contributions proportional to wages capped at P99, plus health-plan amounts.

Why this severity. This concerns the explanation of the headline, not its value. The 1.75 is correct and is reproducible from survey data; tax data matter for C2 (top shares), not C1. The chapter's wording is hedged ('might', 'in conjunction'), hence 'minor' and 'likely'.

Why it matters, in detail. Decomposition of the change in ln(M/F). With a zero-growth benchmark, women's median growth explains 86-100% of the narrowing from any start year 1962-1990. With a balanced-growth benchmark (male median growing at per-adult national income, x1.63-2.16), the male shortfall explains 36-38% from 1962/64 and 57-59% from 1969/73. Only 1999-2014 is male-driven under both (about 98-99%).

Evidence. Chapter p. 21. TB13 cols [8]-[9] (Nov 2017). Male share of the narrowing of ln(male/female median): -2.5% (from 1962), 0.1% (1964), 11.9% (1969), 14.4% (1973), 1.1% (1990), 98% (1999). Male share of the narrowing of the dollar gap: 0.5% (1964), 32.0% (1973), 99.1% (1999) (output/claims_check.txt). Counterfactual with the male median growing like per-adult national income (TA0 col [13]): the male shortfall accounts for 36% (from 1962), 38% (1964), 57% (1969) and 59% (1973) of the log narrowing; counterfactual 2014 ratio 3.2-3.45 vs actual 1.68 (output/median_counterfactual.txt).

Why this severity. This is an interpretive sentence about a secondary result. The facts it summarises (male stagnation, a 7.6-fold female rise) are correctly reported in the same paragraph. Under a balanced-growth counterfactual for the window the sentence names, male stagnation explains 57-59% of the narrowing, so the claim is defensible on one reading and wrong on the plain accounting reading. Downgraded from Auditor B's 'moderate'.