Every congressional hearing is a guest list. Across 32,676 of them, the guest list turns out to track who holds the gavel — and a small class of professional witnesses gets invited no matter who does.
A congressional hearing looks like fact-finding and works like theater. Someone decides who sits at the witness table, and that someone is the committee's majority — the party that won the last election. So a natural question has a measurable answer: when a chamber flips from one party to the other, does the kind of person Congress calls to testify flip too?
We can watch it happen. Every one of 107,142 listed witnesses was sorted into one of seven affiliations — federal government, state or local government, a for-profit company, an industry trade association, an advocacy group or think tank, academia, or a private individual. Federal officials dominate the table (35% of all witnesses); advocacy groups (19%), companies (15%), and trade associations (12%) fill most of the rest. The interesting part is not the average mix. It is what the mix does when control changes hands.
Comparing each committee to itself across the Congresses when it was Republican-run versus Democratic-run — so the committee's permanent character is differenced out — one shift is unmistakable. Advocacy and NGO witnesses lose about 4.5 points of the table under Republican control, a difference far too large and too consistent to be noise. Democratic majorities bring the advocates in; Republican majorities show them out. The business side moves the other way but more quietly: industry and trade-association witnesses gain about 2 points overall, not quite distinguishable from zero across all committees at once.
Two of the pre-registered guesses were simply wrong, and the ways they were wrong are the story. We expected government-witness share to be flip-invariant — oversight as bipartisan ritual. It isn't: government witnesses rise about 4 points under Republican majorities. For most of the window a Republican Congress faced a Democratic administration, and Republican-run committees hauled in more federal officials to answer for it. Oversight is not neutral; it points at whoever runs the executive branch.
The business-witness effect that everyone expects — more industry under Republicans — is real but narrow. It lives almost entirely in the House, and almost entirely after 2010. In the Senate, majority control barely moves the business share at all (−0.4 points, indistinguishable from zero). Before 2011 there is no effect in either chamber; from 2011 on, a Republican House adds about 4 points of industry and trade witnesses. The witness table polarized on a schedule.
If control decides which kinds of people testify, a different force decides which specific people: a professional witness class that is invited regardless of who holds the gavel. Two hundred and eighty-eight individuals testified ten or more times. At the top sits Douglas Holtz-Eakin — 44 appearances across fourteen different committees — the former Congressional Budget Office director turned think-tank president, a one-man institution of expert testimony. Behind him are the people whose jobs are to answer to Congress: Federal Reserve and FDIC chairs, the FBI director, successive Comptrollers General who run the GAO.
None of this makes hearings fraudulent. Expertise is scarce, and the people who have it get asked back. But the two patterns together describe a system with a stable spine and a partisan surface: the same specialists cycle through regardless of control, while the majority quietly tunes the rest of the table — trading advocates for officials and, in the modern House, for industry. The guest list is evidence of who won.
Source: the CoCoHD hearing-details corpus (gtfintechlab), 32,676 hearings 1998–2022, via GovInfo. Committees are canonicalized across historical renames from the raw committee_mapped field; hearings assigned to their primary committee. Witnesses (107,142) were exploded from each hearing's listed roster and classified into seven affiliation types with GABRIEL (gabriel.classify, gpt-4o-mini) — validated against a gpt-4o gold sample at macro-F1 0.91, resolvable 99.5%. Estimates regress a witness-type's share of the committee-Congress table on a Republican-majority indicator with committee fixed effects and standard errors clustered by committee (pyfixest); cells with fewer than five witnesses are dropped. The panel is truncated at Congress 117 (2022) because later hearings publish with a multi-year lag.
Witness affiliations are model-classified; the trade-association ↔ advocacy boundary is the fuzziest category (think tanks, unions, professional associations). The "who testifies most" network resolves clean personal names for 59% of witness rows — recognizable repeat witnesses come through, but the count understates. Hearings without a listed witness table (37%) are excluded from composition. Majority control is coded by chamber and Congress; a hearing's chair, not a raw count, sets the table, and chair and majority coincide here by construction.
Hearing records are U.S. public domain. Corpus: CoCoHD (gtfintechlab). Article text and figures: CC BY 4.0.