# Research method: who funded the incoming House?

Working snapshot: November 7, 2026. This research describes financial backing and documented affiliations. It does not infer donor motivation, coordination, control, or a promise of a future vote.

## Universe and time

- Cover the 435 voting House seats, including reelected incumbents and first-time winners. Delegates and the resident commissioner are outside this universe.
- Freeze the downloaded election roster rather than silently following changing calls. Its snapshot contains 413 called races and 22 counting. No winner is assigned to counting races. Calls are unofficial; statewide certification is a separate event.
- Match candidates by state, district, party, and normalized name. Treat Minnesota DFL as the Democratic party for matching. Treat at-large districts as FEC district 00. Candidate IDs do not encode the current district reliably after redistricting.
- Nickname matching and shared-campaign matches are provisional, not independent identity verification. `seat-roster.csv` preserves the method and result source. Three explicit campaign-name matches are J. D. Ford, Mike Kelly (FEC name George J. Kelly), and Jake Ellzey (John Kevin Ellzey).
- Bob Onder, Mark Harris, and Rob Menendez each have multiple candidate IDs linked to the same principal campaign committee. Preserve aliases and count the campaign only once. Do not sum their duplicate candidate-summary rows.
- The campaign summaries for all 413 called winners end October 14, 2026. Transaction discovery is limited to receipts/contributions dated January 1, 2025, through November 3, 2026, as available in the retrieved files. These windows differ and must remain visible.
- Primary and general activity are included in the cycle-level summaries. Do not describe the totals as general-election-only money. Transaction files retain election designations for future separation; debts and redesignations require review.
- Post-general reports are due December 3 and cover activity through November 23. They are not final full-cycle reports. Later amendments and year-end reporting will also change the data.

## Campaign summaries and denominators

Use FEC `candidate_summary_2026.csv` for the reported funding mix. Confirm that itemized individuals + unitemized individuals + other committees + party committees + candidate contributions equals total contributions, allowing at most two cents of rounding. Keep total receipts, loans, authorized-committee transfers, and refunds in separate fields.

The funding-wall and scatter charts use **gross total campaign contributions** as their denominator. They are not percentages of all receipts, outside support, or money spent in the race. Other-committee contributions include nonparty political committees and are not synonymous with corporate donations. Itemized/unitemized are disclosure categories, not perfect measures of large/small donors.

Candidate fundraising transfers can double count when authorized committees send money to each other. The contribution mix avoids using total receipts as its denominator. Transfers still need examination before a final net fundraising total is reported. FEC explains this issue in its [all-candidates file description](https://www.fec.gov/campaign-finance-data/all-candidates-file-description/).

Five pilot campaigns have 25 financial fields checked against processed OpenFEC totals. The CSV comparison is saved in `data/api-validation.csv`. The API and bulk download are two delivery paths for the same underlying disclosures, not independent external corroboration. One additional API totals request encountered the public demo-key rate limit; the failure is recorded in `api-sources.json`.

## Direct committee contributions

Start with the FEC PAS2 file. Its rows mix direct contributions, independent spending, coordinated spending, and other transaction types. Never sum the file wholesale.

- Direct discovery uses codes 24K and 24Z only, and requires the recipient to be an authorized campaign committee linked to the winner.
- Exclude memo items from this cash/contribution discovery view while retaining them in the source extract for separate attribution work. This is not a universal rule to discard memo items: earmarks and joint-fundraising attributions often need their own treatment.
- Preserve donor committee ID, recipient ID, transaction ID, report ID, source row ID, amendment flag, election designation, date, and filing image link.
- Deduplicate only exact FEC source-row IDs automatically. Flag repeated transaction/report keys for review; do not assume equal dates and amounts are accidental duplicates.
- These amounts are donor-reported, before campaign refunds, and not reconciled net recipient receipts. They are discovery leads and preliminary network diagrams, not publication-ready donor rankings.

The OTH file supplies some recipient transfers, memo attributions, in-kind entries, and refunds, but does **not** supply a complete ordinary contribution receipt ledger for these campaigns. Its observed receipt subset must not be mistaken for all contributions. Only one campaign's observed committee subtotal matches its summary in this first extract. `committee-reconciliation.csv` makes the gaps explicit. Completing recipient verification requires processed Schedule A receipts and relevant Schedule B refunds, including original filings where needed.

Never add donor-side and recipient-side records together as two gifts. Identify leadership PACs and candidate/party transfers separately from industry-backed PACs. A PAC's sponsor is not the same thing as a corporate treasury contribution.

## Individuals, employers, and industry

The individual bulk file is a **subset** of itemized donations: its inclusion rule generally requires the donor's aggregate to exceed $200 for the relevant reporting period rule. It is not every publicly itemized small contribution. [FEC file definition](https://www.fec.gov/campaign-finance-data/contributions-individuals-file-description/)

Read the archive's `itcont.txt` member once; its `by_date/` members repeat the same data in partitions and must not be added again. Filter to the winners' authorized campaign committees before parsing donor details. Preserve source records in a local SQLite database. Deduplicate only identical source IDs.

The first employer-group extract uses nonmemo types 15 and 15E, retaining signed amounts. Keep other types and memo entries in the local database. This does not complete joint-fundraising, conduit, tribe, partnership, or refund attribution. In particular, OTH contains substantial 15J memo attribution, which needs a separate donor-origin pass. Unitemized totals have no donor identities to classify.

The completed archive pass scanned 35,327,429 source rows and selected 2,207,601 campaign records, with no duplicate selected source IDs. It produced 154,355 seat/employer groups and 8,237 leading reported name/state/employer labels. These are groups and records, not counts of unique donors. In 249 campaigns the latest selected nonmemo 15/15E transaction is dated June 30; the latest across the extract is October 11. These observed date windows are not guarantees of completeness and cannot be substituted for the October 14 summary coverage. Each profile and `individual-coverage.csv` exposes the difference.

Group reported employer names conservatively using uppercase and whitespace trimming. Never automatically merge people solely because their names match. The top-individual file groups reported name, state, and employer; those groups are not established unique people and must not be published as resolved donor identities. The database retains filing links for verification.

Employer affiliation supports wording such as “individual donors reporting Google as their employer.” It does not support “Google gave the candidate this money,” nor any claim of employer direction. Employment status labels such as retired, not employed, self employed, homemaker, and information requested remain unclassified unless further evidence supports attribution. An employer's sector is not proof of an individual donor's political motive.

Maintain two evidence tables:

- `classifications.csv`: committee sector, organization type, issue affiliation, evidence basis, and confidence. Medium confidence entries need additional organization-source checks. FEC registered organization types also support structural classifications for otherwise unclassified donors.
- `employer-classifications.csv`: 15 exact reported employer labels with checked business-sector sources, including technology/AI, investment/trading, crypto services, defense technology, and government-relations firms. Related spelling labels remain separate rather than being silently merged. All other employers remain unclassified, not zero or interest-free.

Economic sector, organization type, and issue affiliation are different dimensions. Do not add their overlapping totals together. Labor PACs can represent workers in several sectors. Do not force a diversified company into a single narrow business line. Do not infer ideology from an individual donor's name, faith, ethnicity, occupation, or nationality.

Any industry-share graphic must state its denominator and show the unclassified remainder. In the first pass, sector totals are a classified subset of committee contributions, not a full industry map of all campaign money. API employer charts show selected named employers from top-20 results and exclude employment-status labels; they are illustrative leads, not complete employer rankings or the same cutoff as the bulk summary.

## Outside spending

Keep independent expenditures and campaign contributions in different ledgers. Preserve support and opposition as separate fields, and distinguish spending about the winner from spending about other candidates. Primary spending against another candidate is not automatically general-election support for the eventual winner.

The independent-expenditure bulk file contains original and amended notices together. FEC explicitly warns of duplicated transactions. `outside-spending-review.csv` identifies reports superseded through explicit `prev_file_num` links, but that is not complete deduplication. No spending totals are calculated in this first pass. [FEC file definition and duplication warning](https://www.fec.gov/campaign-finance-data/independent-expenditures-file-description/)

Before publication, reconcile notice amendments, periodic Schedule E reports, estimate/actual updates, unpaid obligations, and repeated transaction IDs. Never simply add notices to the PAS2/periodic-report totals. Retain the named candidate and support/oppose direction. If investigating a spending group's donors, describe disclosed funding of that group separately; do not assign its whole donor pool proportionally to a specific seat without evidence.

## Publication requirements

The research package is a dated working snapshot. Before a final article: verify winner identities/result calls, reconcile the donor leads used in prose, complete the joint-fundraising attribution needed for any donor-origin claims, review medium-confidence classifications used in findings, refresh reporting, and disclose the remaining unknown share. Only publish outside-spending numbers after their separate reconciliation passes.

All graphics retain their actual scope and cutoff in their captions. SVG and PNG versions are generated from the same Plotly figure data as the interactive versions. No invented donors or demonstration amounts are used.
