Sourcing and balance by outlet group (news stories)

What this dataset measures

Coverage of I-638 was collected from outlet search pages, sitemaps, RSS feeds, web search and syndicated copies for three outlet groups fixed before coding (mainstream, conservative, progressive/advocacy), plus labeled fact checks. Any story that describes the measure or quotes someone on it was included; republished copies were coded once under the original. The same codebook was applied to every outlet so that omissions and errors can be compared in both directions.

Geography

Washington State, United States

Unit of observation

One news or opinion story about I-638

Update cadence

One-time audit; corrections published as new versions

Curator

FactHub Research Team

Curated data tables

Sourcing and balance by outlet group (news stories)

Average advocates per side, first and last word, missing rebuttals, stories with unchecked claims and outlet-voice errors.

Unit of observation: Outlet group

sourcing-balance.csv text/csv · 344 bytes

All 3 rows 12 columns Download the file

All 3 rows

group	news_stories stories	avg_yes_sources people per story	avg_no_sources people per story	first_word_yes stories	first_word_no stories	last_word_yes stories	last_word_no stories	no_rebuttal stories	stories_unchecked_no_claim stories	stories_unchecked_yes_claim stories	stories_with_outlet_voice_error stories
mainstream	77	1.34	1.96	44	24	11	55	18	31	14	22
conservative	87	1.87	0.63	70	12	57	11	58	9	9	18
progressive	29	0.79	2	7	20	4	20	24	15	3	16

12 field definitions

Field	Type	Meaning	Units	Value type
group	string	Outlet group.	—	observed
news_stories	integer	News stories read.	stories	derived
avg_yes_sources	number	Mean Yes advocates per story.	people per story	derived
avg_no_sources	number	Mean No advocates per story.	people per story	derived
first_word_yes	integer	Stories where the Yes side is quoted first.	stories	derived
first_word_no	integer	Stories where the No side is quoted first.	stories	derived
last_word_yes	integer	Stories where the Yes side is quoted last.	stories	derived
last_word_no	integer	Stories where the No side is quoted last.	stories	derived
no_rebuttal	integer	Stories with no response from the other side to the lead claim.	stories	derived
stories_unchecked_no_claim	integer	Stories repeating at least one No-side claim with no context.	stories	derived
stories_unchecked_yes_claim	integer	Stories repeating at least one Yes-side claim with no context.	stories	derived
stories_with_outlet_voice_error	integer	Stories with a factual error in the outlet's own voice.	stories	derived

Supporting files

Source snapshots, manifests, quality reports, documentation, and other materials used to produce or review this release.

• codebook.md Documentation · text/markdown · 8 KB (8,153 bytes)

  The codebook fixed before coding: ground truth from the text, sample rules, every field, claim codes and facts.

• derive.js Transformation manifest · text/javascript · 4.7 KB (4,849 bytes)

  Node.js script that recomputes the three derived tables from stories.csv (node derive.js).

Sources

1. Washington Secretary of State: Initiative Measure No. IL26-638, filed text Ground truth for coding facts and errors: Sec. 3(1) and 3(2) and the Sec. 1 findings

   Retrieved October 23, 2026. Public record (Washington State government). Linked, not redistributed

   Reliability: Government. The text as filed with the Secretary of State by the sponsor. It is the measure voters decide on; its Sec. 1 findings are the sponsor's own statements.

2. Washington Secretary of State: 2026 General Election voters' pamphlet, IL26-638 statements Source of the campaign claim codes (each side's own pamphlet wording)

   Retrieved October 23, 2026. Public record (Washington State government). Linked, not redistributed

   Reliability: Government. The Secretary of State prints each campaign's argument and rebuttal as the campaign wrote it; the state does not check their accuracy.

3. WIAA: Health and wellness forms (medical history, physical examination and medical eligibility forms) Ground truth for fact F12 and the YDOC claim code: the history form that stays with the doctor asks the family to report sex assigned at birth; the forms sent to schools have no sex field, and no form has the provider verify sex

   Retrieved October 23, 2026. Publicly posted form. Linked, not redistributed

   Reliability: Other. The recommended pre-participation forms posted by the Washington Interscholastic Activities Association, which runs high-school sports. WIAA marks the history and physical examination forms as staying with the doctor and the medical eligibility form as going to the school.

4. The 262 coded news and opinion stories (one URL per row of the stories table) Coded material. Each row's url is its own source; this entry links one example (KUOW, June 4, 2026).

   Retrieved October 23, 2026. Copyright of each outlet. Only URLs, headlines and short quotes for analysis are included; story text is not redistributed

   Reliability: News report. News and opinion outlets of every leaning, grouped before coding as mainstream, conservative or progressive/advocacy. Their own positions on I-638, where stated, are recorded in the opinion rows.

Limitations and reproducibility

Partially reproducible: Every coded story has its URL, and the codebook, the coded table and the script that computes every derived table (derive.js) are published, so anyone can recode a story and recompute every count. Coding is human judgment and was done by one coder per outlet group; no blind second coder has checked it. Some pages have since changed or are paywalled, so a recoder may see different text.

• No blind second coder has checked this coding; agreement is unmeasured. (A similar audit of I-645 coverage, with the same design, found 85-100% agreement.)
• The sample was found by search, not drawn at random. KING 5, KREM, KGW and McClatchy papers block automated reading and are under-represented; some of their stories were read through archived or syndicated copies, and 8 of 262 rows could not be read (their fact fields are blank).
• TV and radio stories were coded from their web text, not the broadcast.
• Including a fact is not the same as getting the story right, and omitting one is not by itself an error: short event stories cannot carry every fact.
• Counts of advocates do not weigh how much space each side got.
• Names of minors, and of parents whose names would identify them, are replaced with bracketed descriptions in every field. One story's URL (row P28, a Spokesman-Review op-ed of Oct. 11, 2026) is withheld because its slug contains a minor's name; its outlet, date, headline and codes are kept.
• Outlet groups are our classification, fixed before coding; The Stranger, Axios and Cascade PBS are counted as mainstream.
• F12 was reworded on 2026-10-23 after we checked WIAA's forms: the history form that stays with the doctor asks the family to report sex assigned at birth, but the forms sent to schools have no sex field and no form has the provider verify sex. Stories coded 1 said current physicals or forms do not record or verify sex, which the corrected wording still covers; no codes changed.