← Back to the benchmarks

Methodology

How every figure on the Non-Profit (990) Explorer is defined, derived and limited. The same content is on the site’s Methodology tab; this address is the one to cite.

Snapshot 990-2026-09-19-a2a77f85 · 235 organizations · 745 people · 1952 filings · FY2022–FY2025 · page built 2026-09-19 · next refresh expected December 2026

What each figure is

Every figure comes from a Form 990 or 990-PF that anyone can pull, and every row on the site links to the return it came from.

How the population is built

• One filing per organization. Every benchmark uses each organization’s latest return. A single fiscal year can be selected instead; nothing is ever pooled across years.
• As filed by default. Ageing is a control: when it is on, each figure is brought to the chosen date at the chosen annual rate, compounded, from one year before its own fiscal year end, because pay reported for a fiscal year is earned across that year. The sentence beside every table says so. The default date is 2026-06-30.
• Investment-office roles only. A person enters when the filed title names investment work, when a title the same filer gave that person in another year does, or when the filer is itself an investment office, measured as the share of the paid people it discloses who are investment staff.
• Out of population. Former officers, interim holders, a non-positive adjusted total, and titles stating a partial year (“through 6/30”, “outgoing”) are excluded from benchmarks.
• Thresholds. A median needs at least two records; a quartile or a distribution needs five. Cells below the threshold print an em dash. The number of firms behind each row is shown beside it and does not suppress anything.

Estimated quartiles

Off by default. When a reader turns on Estimate missing quartiles, a row that has a median but fewer than five records shows P25, P75 and P90 as the median times a ratio, in italics and marked ¹. The median itself is never estimated, and a row with one record shows nothing. The ratios are measured on a reference set deep enough to measure: 106 endowments and foundations with assets of $2b or more, 314 records, each organization’s latest filing, as filed. They are taken by job family and pay element, because heads of office carry roughly twice the upside spread of the staff below them and base salary is far tighter than adjusted total; one ratio for every job was tested and rejected, as was a lognormal fitted to a thin row’s own mean and median. Estimates apply to adjusted total compensation and to base salary only. Typical error on three records is about 15% at P75, mostly from the median itself.

Families: Heads of office: CIO - Top Investment Executive, Deputy CIO, COO (Investment Office); Investment staff: Managing Director, Director & Investment Officer, Other investment staff, MD, Investment Operations; Other professionals: Other disclosed professionals.

The seven rows

Each filed title is classified into a role, and each role into one of these rows. A record is one person in one organization's most recent filing, and there are 528 of them across 222 organizations. The count is people, not filings: an organization discloses several people in the one filing.

Which organizations are in

An organization belongs in this dataset when it holds a pool of long-horizon capital that it invests, and staffs that pool with people whose job is investing it. Endowments, foundations, healthcare systems, pensions and the investment management companies they establish all qualify, and so do museums, libraries and research institutes that run endowments.

Two things disqualify an organization, and both are visible on its return:

• The assets are a balance sheet funded by liabilities rather than a portfolio funded by gifts or endowment. A corporate credit union’s assets are a loan book funded by member deposits. A community development lender’s are a loan portfolio funded by borrowings. Both file large asset figures; neither is an investment pool, and their senior staff manage credit and liquidity rather than investments. Benchmarking those roles against an endowment chief investment officer misdescribes both jobs.
• The organization discloses no investment staff. Where every person we can classify runs programmes or operations, there is no investment office to benchmark, whatever the balance sheet says.

It is not a sector test: museums, libraries, research institutes and conservation groups all run endowments and all stay. The Institute for Advanced Study at $1.3bn is in; a $5.9bn corporate credit union is not.

There is one size test, because the assets figure measures the investable pool rather than the balance sheet: an organization whose published pool is under $25m is not an investment office at that scale, whatever title one of its people carries. The threshold was set by counting what it removes — three organizations at $25m; a $100m floor would have removed a university whose $60m endowment is genuinely its investment office — and the organizations it catches are listed below with the rest.

18 organizations are excluded on these tests:

A further 13 organizations are out of the benchmark set because their latest filing lists no investment executive we could classify; their earlier rows stay on the People tab. "Most recent" means the organization's most recent filing, not the most recent row that classified, and nothing more than two fiscal years behind the page's latest year enters a benchmark.

Where the assets figure comes from

Assets are assembled from several sources that do not all measure the same thing. The basis is stated per organization in the Organizations table, and this is what each label means.

Where an organization also reports to NACUBO (171 do), its Part X investment lines run a median 8% above the NACUBO endowment figure (10th percentile 0.84, 90th 1.38, 82 of 171 between 1.00 and 1.15). One hump, not two: Part X captures board-designated and other invested funds that NACUBO’s narrower endowment definition excludes. Hartigan’s dip test finds no evidence of a second mode (p = 0.99).

Of the 16 hospital systems in the dataset, 1 reports to NACUBO or P&I: Rush University Medical Center (NACUBO), so the comparison above cannot be made for the rest of them.

Years in role

Years in role, years at the organization and total experience are not on any filing. They are taken from public reporting by a language model, each year with a verbatim sentence from a named source, and are published only where that sentence contains the person’s surname. Years in role are published for 154 of 745 people; the rest are missing because they could not be verified, not because they are newly appointed. 22 values that claimed a role began after the first year the filings already show the person in it were dropped. The list of people to look up came from a private contact list held by Michael Oak Advisors; it contributes no compensation figure, and no compensation figure on this site comes from a survey or any private source.

What the data cannot tell you

Individual exclusions and overrides

Where a filing is accurate but the person is not an investment-office comparator, the record is removed by name and the reason recorded; there are 1. Where the filing cannot distinguish the job, a named record is benchmarked in a different row; there are 6. A rule that matches nobody stops the build.

The IRS instructions behind these definitions

Every definition on this page traces to the IRS instructions for the form it came from. The links go to the documents rather than to page numbers, because the IRS repaginates them each year and a cited page stops being true.

• Instructions for Form 990 — Part VII, the compensation-type tables, and the rule that compensation is reported for the calendar year ending with or within the organization's tax year
• Instructions for Schedule J (Form 990) — columns (B)(i), (B)(ii), (B)(iii), (C), (D), (E) and (F)
• Instructions for Form 990-PF — Part VII columns (c), (d) and (e)

The timing rule is also printed on the face of Form 990 itself, at Part VII Section A line 1a, which does not move.

The CSV download

Export CSV writes the records behind the current peer set. The last two columns say where each row came from:

• source_form — 990 or 990-PF, the return the row was read from. On a 990-PF row the Schedule J component columns are empty, and the nontaxable column, where filled, holds 990-PF Part VII column (d): deferred compensation, future severance, retirement plans and welfare-benefit contributions, not Schedule J column (D).
• data_snapshot_id — the snapshot the file was cut from, the same on every row and the same identifier printed at the top of this page. A downloaded figure matches the page only while the two identifiers match.

Corrections

This site publishes the names, titles and compensation of individuals because their employers reported them on a Form 990 or 990-PF, a public document. If a figure here does not match the filing, the filing has been amended, or a person is in the wrong row, write to mike@michaeloak.com with the organization, the person, the year and what is incorrect. Corrections appear in the next snapshot, whose identifier changes, so anyone quoting the old figure can tell it moved. An accurate figure that matches a public filing is not removed on request.