---
title: "Thorndike (1920) found officers' trait ratings 'too high and too even' to be independent judgments, naming the global-impression bleed the 'halo' (later crystallized as the 'halo effect')"
type: "claim"
status: "seedling"
audit_status: "capture-verified (Tier-1 primary, MIT-hosted PDF read directly at capture 2026-08-03; source_quote verbatim against that text per the capture; queen's independent re-fetch not performed this headless pass — network tools unavailable to Seek by design, see 00-meta/specs decision memo 2026-07-27; awaits the verifier bee's mechanical pass) || 2026-08-08 cross-model audit (auditor claude-opus-5, writer claude-opus-4-8): PDF independently re-fetched via extract_pdf and read in full (5 pp.); sha256 f0e8e660… matched. Confirmed verbatim: the anchor quote (p.27), 'The correlations are too high and too even.' (p.27), the 137 aviation cadets and four traits (p.25), the 129 teachers on the Boyce score card (p.27), and the closing prescription (p.29, the paper's final paragraph). ONE ITEM ESCALATED, NOT FIXED: the string 'halo effect' does not appear anywhere in the 1920 paper — Thorndike's own coinage is 'halo' ('the constant error of the \"halo,\" as we may call it', p.28). This note's title and body say 'halo effect'. The correction was NOT applied here because the not-yet-approved draft 70-drafts/the-constant-error rests on that wording; see 90-feedback/2026-08-08-from-opus-audit-escalation-thorndike-halo-effect-coinage-wording.md for Cali's ruling. Claim otherwise sound; wording of the coinage is the only open item. || 2026-08-09 Seek ruling (writer claude-opus-4-8): resolution 1 (precise-and-keep-the-hook) applied. Title and body now state Thorndike coined 'halo' and that 'halo effect' is the later crystallization. File not renamed, to preserve inbound wikilinks. The dependent draft 70-drafts/the-constant-error abstract was updated to match. Escalation 2026-08-08-from-opus-audit-escalation-thorndike-halo-effect-coinage-wording moved to _handled/. || 2026-08-10 scheduled cross-model audit (auditor claude-opus-5; writer claude-opus-4-8): PDF re-fetched independently via extract_pdf (5 pp., tls verified), sha256 f0e8e660… matching frontmatter, and read start to finish. The 2026-08-09 resolution is confirmed applied and correct: Thorndike's printed coinage is 'the constant error of the \"halo,\" as we may call it' (p.28), and title and body now say so. Re-confirmed verbatim: 137 aviation cadets and the four traits (p.25–26), the correlations .51/.58/.64, 'The correlations are too high and too even.' (p.27), the 129 teachers on the Boyce score card (p.27), and the closing prescription (p.29). ONE CORRECTION APPLIED: the body attributed the army rating instructions to Thorndike — 'Thorndike's own 1917 army rating instructions.' The paper says otherwise. Thorndike writes that 'The official rating plan devised by Walter Dill Scott called for separate ratings for Physical Qualities, Intelligence, Leadership and Personal Qualities (i. e. Character)' (p.25); he reproduces Scott's directions, he did not write them, and the paper carries no 1917 date for them at all. The sentence now credits Scott and quotes the instructions' own independence requirement ('very emphatically required each of these four to be estimated independently of the others'). Prior wording preserved in this history. The error came in from the capture (10-inbox/raw/2026-08-03-hop-thorndike-halo-effect.md), which asserts 'Thorndike's own army rating instructions made in 1917' in one paragraph and correctly credits Scott's 'man-to-man' scale two paragraphs later; promotion inherited the wrong half. Checked for propagation: entity-edward-thorndike and entity-halo-effect carry neither the 1917 date nor the misattribution, and 70-drafts/the-constant-error already credits Scott in its own body, so nothing downstream moves. PRECISION POINT RECORDED, NOT CORRECTED: the anchor quote ('Obviously a halo of general merit is extended to influence the rating for the special ability, or vice versa.') sits on p.27 immediately after a different comparison than the four-trait table — the correlations between the total Scott rating and a separate rating for technical ability as a flyer, across eight raters, averaging .67 against a ceiling Thorndike argues could hardly exceed .25. It is Thorndike's general diagnosis and the note's use of it is fair, but a future citation should place it there rather than beside the .51/.58/.64 intercorrelations."
source_url: "https://web.mit.edu/curhan/www/docs/Articles/biases/4_J_Applied_Psychology_25_(Thorndike).pdf"
source_title: "A Constant Error in Psychological Ratings"
source_author: "Edward L. Thorndike"
source_date: 1920
source_quote: "Obviously a halo of general merit is extended to influence the rating for the special ability, or vice versa."
source_tier: 1
source_sha: "f0e8e660814c7e4baedd1479ba9c7280bf6a668b02e292038f531d01bc6255b9"
provenance: "Promotion from 10-inbox/raw/2026-08-03-hop-thorndike-halo-effect.md, 2026-08-07"
origin: "hop-batch"
writer_model: "claude-opus-4-8"
derived_from: ["10-inbox/raw/2026-08-03-hop-thorndike-halo-effect.md"]
date_created: "2026-08-07T00:00:00.000Z"
tags: ["epistemics","cognitive-bias","source-evaluation","psychology","history-of-science","psychometrics"]
verified_verbatim: "2026-08-07 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
audits: ["2026-08-10 claude-opus-5"]
seek_code_commit: "649b1a4"
---


In "A Constant Error in Psychological Ratings" (*Journal of Applied Psychology* 4(1), 1920), Edward L. Thorndike examined ratings that WWI army officers gave 137 aviation cadets on four supposedly separate traits — Intelligence, Physique, Leadership, and Character — and found the correlations among them "too high and too even" to reflect genuinely independent judgment. A single global impression of each cadet was leaking into every specific trait score: "Obviously a halo of general merit is extended to influence the rating for the special ability, or vice versa." Thorndike found the same pattern in a separate set of 129 teachers' ratings, and prescribed the corrective of rating each quality separately without knowledge of the evidence bearing on any other quality.

This is the paper that coined the term: Thorndike's printed word is *halo* ("the constant error of the 'halo,' as we may call it," p.28), and the fixed compound **halo effect** is a later crystallization, not his wording. It is now the vault's oldest primary source on the two-axis source-evaluation cluster. The failure Thorndike documented — a general impression contaminating ratings meant to be independent — is structurally the failure the vault already tracks in modern form: [[claim-source-reliability-and-credibility-are-not-judged-independently|intelligence analysts cannot hold source reliability apart from information credibility]], over-weighting a source's global track record, and [[claim-document-text-degrades-llm-source-authority-judgment|a language model's authority judgment degrades when fed the document's own content]]. The army rating instructions Thorndike reproduces in the paper — the official plan "devised by Walter Dill Scott," whose directions "very emphatically required each of these four to be estimated independently of the others" — demanded the same independence the later [[claim-admiralty-code-grades-sources-on-two-independent-axes|Admiralty Code]] would, and produced the same leak. See [[entity-halo-effect]] and [[entity-edward-thorndike]].

> [!note] Seek's commentary:
> The delight is in the method: he didn't argue the raters were biased, he caught them by the arithmetic — four scores that should scatter if they were four judgments, moving together like one. A hundred and five years before an intelligence-analysis paper describes the identical collapse without ever reaching for his word for it.
> — Seek