Emmanuel Saez · Statistical Journal of the IAOS 33(1): 219-221 (Festschrift for Arthur Kennickell)

Comments on 'Lining up: Survey and administrative data estimates of wealth concentration'

2 problems found: 2 minor

What the paper claims

Unlike income, net wealth is essentially absent for about half of US families ('While almost all US households have some income ... about half of US families have essentially zero wealth in net').
The top decile (ranked by wealth) owns over two thirds of total wealth.
The top percentile owns over one third of total wealth (numbers 'come from wealth statistics that Arthur has helped produced', i.e. the SCF).
Measuring wealth is hard because (i) no administrative data record US wealth systematically and comprehensively (no comprehensive wealth tax), unlike income and earnings, and (ii) wealth is much more concentrated than income, so random surveys capture few top respondents and top records are prone to reporting/recording error.
The SCF dual-frame design addresses these challenges: the area-probability sample measures wealth within the bottom 99%; the list sample, built from individual income-tax data, oversamples high-net-worth units '(with net worth estimated capitalizing their asset income)', as explained in 'Lining up' [8].
Substantive assessment of Kennickell's 'Lining up' and of reconciling survey (SCF) and administrative/capitalized-income (Saez-Zucman) estimates of wealth concentration. NOT OBTAINED; indirect evidence only: Kennickell's Afterword (SJIAOS 33(1): 263) says 'Both Barry Johnson and Emmanuel Saez note the importance of clarifying the relationship between income and wealth'; Crossref-deposited references include Kopczuk & Saez (2004) and Saez & Zucman (2016).

Every problem found

1minorconfirmed · newly identified here · overclaiming

Taken together, the poorer half of American families owns only about 1% of all wealth, so in that sense the comment is right. But it is not true that half of all families individually have nothing: the typical family had about $81,000 in 2013, and only about one family in eight had zero or negative wealth. The wording is loose, but it does not change the comment's point that wealth is extremely concentrated.

Technical detail and evidence

Why it matters, in detail. Under the per-family reading the prevalence of negligible net worth is overstated by a factor of about 2-4: 13% of families <= 0, 21% <= $5k and 26% <= $10k against 'about half'. On a vehicle-excluded (Saez-Zucman-like) concept the factor is about 1.5-2.3. Under the aggregate reading the bottom-50% share of 1.1% (-0.01% excluding vehicles) makes the statement essentially exact. Direction: rhetorical overstatement of zero-wealth prevalence. No other number on p. 219 is affected; top-10% (75.3%) and top-1% (36.3%) shares are unchanged.

Evidence. Saez (2017) p. 219, Section 1, col. 1, 2nd paragraph; fn 2. SCF 2013: bottom-50% share 1.1% (Kennickell 2017 FEDS Table 3, p. 32: -0.7+0.2+1.6; adjudicator re-computation from the public summary extract 1.08%; -0.01% excluding vehicles). Per family: net worth <= 0 13.0% (Kennickell Table 2, p. 31; own 12.9%); <= $5k 20.8%; <= $10k 26.2%; 40th percentile about $38.8k; median $81,200 (Federal Reserve Bulletin 2014, p. 8). Excluding vehicles: 21.8% <= 0, 31.7% <= $5k, median about $64k. Scripts: final/scripts/scf_checks.py, scf_lowwealth_exveh.py. Same aggregate shorthand in Saez & Zucman (2016) QJE p. 554 and fn 45 ('the bottom half of the distribution always owns close to zero wealth on net', citing the SCF's 1.1%).

Why this severity. A framing statistic, not the comment's contribution. Read as an aggregate share, the evident intended sense and the authors' usual wording, it is accurate to within about 1 pp. The concentration conclusions (C2-C4) do not depend on it.

2minorconfirmed · newly identified here · other

Two small typing mistakes; they change nothing in what the comment says.

Technical detail and evidence

Why it matters, in detail. None: no figure, source attribution or argument is affected.

Evidence. p. 219, col. 2, 3rd paragraph, and fn 2 (publisher first-page preview, /mnt/zdata/downloads/saez2017-comment-claude/saez_summary.png).

Why this severity. Copy-editing slips with no bearing on any claim or number.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

I2: the parenthetical describing list-sample units' net worth as 'estimated capitalizing their asset income' misrepresents 'Lining up', because since 1998 the stratifier (WINDEXM) blends the capitalization index WINDEX0 with a regression index WINDEX1 (more weight from 2004), and survey net worth matches WINDEX1, not WINDEX0. — The facts about WINDEX0/WINDEX1 are correct (Kennickell FEDS 2017-017 pp. 6-7, fn 22, pp. 20-21), but the parenthetical is a fair one-line simplification in the source's own terms. 'Lining up' describes the frame as stratified on a wealth index 'constructed largely from observed income flows' (abstract) and as 'wealth proxies based in part on capitalized income flows' (p. 26). WINDEX1 is itself an income-to-wealth mapping that 'implicitly embeds rates of return' (p. 7). Johnson's commentary in the same issue (p. 215) describes the 1989 index as 'computed by capitalizing the various income sources using appropriate average annual rates of return'. The comment refers readers to 'Lining up' for detail, and no claim on p. 219 depends on the distinction. It would need re-examination only if pp. 220-221 equate the SCF stratifier with the Saez-Zucman method, which could not be checked. Severity: none.
Caveat (not claimed as an issue): 'the top percentile owns over one third' fails under the Fed DFA concept (29.4% in 2013Q3) and is marginal under the BHKS (2016) preferred measure (about 33%). — Not an issue. fn 2 correctly sources the figure to the SCF (36.3% in 2013; Fed Bulletin 2017 p. 10). The DFA was first published in 2019, after the comment. BHKS's preferred measure gives 'about 33 percent' (BPEA 2016 p. 292), essentially one third. The qualitative point holds under every concept (top 10% at least 68.7%). Concept dependence is recorded under checks passed.
Side note: Kennickell's 'Lining up' Table 4 ('Shares of income received by various percentile groups', 'Income percentile group') appears to report income shares of wealth-ranked groups. — Out of scope: it concerns the discussed paper, not Saez's comment. The adjudicator's re-run supports the side note (SCF 2013 income shares of wealth-ranked groups: top 1% 15.2%, P90-99 25.9%, bottom 50% 22.6%, against Table 4's 15.7%, 25.8% and 22.5%; income-ranked: 19.7%, 27.2%, 14.5%), but nothing on p. 219 uses Table 4.
No substantive verdict is possible and no core finding can be identified, because the comment's text could not be obtained. — Partly incorrect. p. 219 is publicly viewable as the publisher's first-page preview (sage.cnpereading.com issue table of contents). Core findings C1-C5 were extracted and verified from it. The point that pp. 220-221 are unassessed stands.

Audit notes

What reproduced. Reproduced from the public SCF 2013 summary extract (five implicates, family weights; adjudicator script final/scripts/scf_checks.py): top 10% 75.0% (published 75.3%), top 1% 35.5% (published 36.3%; the public file suppresses Forbes-level cases), bottom 50% 1.08% (published 1.1%). Per-family shares (rescaled to the 2013 Bulletin mean of $534,600): 12.9% <= 0 (published 13.0%), 26.2% <= $10k, median about $82k (Bulletin $81,200). Income vs wealth: top-1% income share 19.7% (Bulletin 2017: 20.3%) vs wealth about 36%; 0.28% of families report income <= 0. Concept cross-checks: DFA 2013Q3 68.7/29.4/0.9 (top 10/top 1/bottom 50); Saez-Zucman 2016 77.2/41.8 (2012); BHKS preferred top 1% about 33%. All of Auditor A's numbers were re-derived independently and matched to within 0.1 pp. Not reproduced: anything on pp. 220-221 (text not obtainable: publisher 403/Cloudflare and captcha pages, closed access in Unpaywall/OpenAlex/Semantic Scholar, no Wayback capture, Google Scholar's 3 versions are SAGE plus two EBSCO subscription records, no working paper).

Limits of this audit. Two of three pages unread: pp. 220-221, which hold the comment's substantive discussion of 'Lining up', its recommendations and most of its 10 references, could not be obtained legitimately. Attempts by both auditors and the adjudicator are listed in final.md section 3. Non-public file paths, captchas and shadow libraries were deliberately not used. Any error, overclaim or engagement (or non-engagement) with the heterogeneous-returns, fixed-income-capitalization, unit and uncertainty critiques there is outside this adjudication, and the verdict covers C1-C5 only. 'Lining up' was read in its FEDS 2017-017 version (SJI version at abstract level only). Public SCF data differ slightly from the internal data behind published shares (top 1% 35.5% vs 36.3%). The re-released extract was rescaled to 2013 dollars via the published mean. DFA figures are the 2026Q2 vintage (the DFA did not exist in 2017). Semantic Scholar snippet search and scholar.archive.org were rate-limited throughout.