C33 Workplace training
alive Untested: no attacks yet · serves Q1 · last change 11 November 2026
Organised workplace training for adults raises what they learn and how they behave on the job by about half a standard deviation or more on average, compared with untrained comparison groups.
Go deeper
Where it holds
All limits (6)
Inside the circle: the core of the claim. Outside: every bound and carve-out in the claim file, numbered, with the attack that forced or added it where the record shows one.
forced by a narrowing attackin the original wordingwritten at creation, or added without a narrowing
Warm cards in the list were added by later attacks; plain cards were in the original draft or written at creation.
Formal training programmes in organisations, judged by learning tests and on-the-job behaviour; not reactions or company results alone.
Written when the claim was created (29 October 2026); not forced by an attack
Evidence on file: Lacerenza, Reyes, Marlow, Joseph & Salas 2017 (Journal of Applied Psychology, DOI 10.1037/apl0000241; abstract only): leadership training improved learning (δ = .73) and transfer to the job (δ = .82); stronger with practice, feedback and spaced face-to-face sessions.
Written when the claim was created (29 October 2026); not forced by an attack
Arthur, Bennett, Edens & Bell 2003 (Journal of Applied Psychology, DOI 10.1037/0021-9010.88.2.234; full text read 2026-11-07): sample-weighted d = 0.63 for learning (k = 234, N = 15,014) and 0.62 for behavioural criteria (k = 122, N = 15,627); supports the claim's size (not an attack).
Written when the claim was created (29 October 2026); not forced by an attack
Wording changed later: Added on 7 November 2026 without a matching attack in the record.
Weak spots of this meta-analysis (corrected 2026-11-11, from "many evaluations are pre-post or self-report"): single-group before-and-after designs were excluded (p. 237) and self-reported reactions are a separate criterion (k = 15), but whether studies were randomised was never coded, there is no publication-bias check, nearly all effects come from journal articles of 1960–2000, and the intervals printed as "95% CI" (learning −0.53 to 1.79; behaviour 0.05 to 1.19) show how widely effects spread across studies, not how precise the averages are (Table 1, p. 239).
Added with attack #1 (corrected; it did not narrow the claim) · Arthur, Bennett, Edens & Bell 2003 · 11 November 2026
Where the same 17 studies (N = 839) measured both, job behaviour came out smaller than learning (d = 0.44 vs 0.66).
Added with attack #1 (corrected; it did not narrow the claim) · Arthur, Bennett, Edens & Bell 2003 · 11 November 2026
Without a study-quality filter it cannot test the kill line either way.
Added with attack #1 (corrected; it did not narrow the claim) · Arthur, Bennett, Edens & Bell 2003 · 11 November 2026
0 of 6 bounds and carve-outs were forced by a specific attack.
How it changed
There is nothing to replay yet: C33 has not been attacked, and only one wording exists, the one it was given on 11 November 2026.
Organised workplace training for adults raises what they learn and how they behave on the job by about half a standard deviation or more on average, compared with untrained comparison groups.
What would change this
What would kill it: A meta-analysis of workplace-training evaluations finds learning and on-the-job behaviour effects near zero (below about 0.2 SD) once study-quality filters are applied. Added as a claim on 29 October 2026, promoted from a candidate belief in the evidence library.
Attack record
No attacks yet. This claim is untested: its wording has not been exposed to counter-evidence in this project.
Corrections to our notes
These fix a number or a fact in our notes. They are not attacks and do not count toward the floor of 3 survivals.
- — C33's limits name the weak spot of the evidence as "many evaluations are pre-post or self-report", which is not accurate for Arthur et al. 2003, the main full-text source.
Details
What was checked: accuracy of the Bounds' weak-spot sentence for Arthur et al. 2003; not a kill-condition test (no study-quality filter was applied, so the meta-analysis cannot show effects below about 0.2 SD "once study-quality filters are applied"; its unfiltered means, 0.63 and 0.62, support the claim's size).
Why: Nightly-attacks 2026-11-11 Europe/Stockholm. p. 237: "studies based on single group pretest–posttest designs were excluded from the data." Table 1 (p. 239): reaction d = 0.60 (k = 15, N = 936); learning 0.63 (k = 234, N = 15,014, corrected SD 0.59, interval −0.53 to 1.79); behavioural 0.62 (k = 122, N = 15,627, SD 0.29, interval 0.05 to 1.19); results 0.62 (k = 26); within studies measuring both learning and behaviour (k = 17, N = 839): learning 0.66 vs behaviour 0.44. The printed "95% CI" bounds equal the mean ± 1.96 × corrected SD (0.63 ± 1.96 × 0.59 = −0.53 to 1.79), so they describe the spread of effects across studies, not the precision of the mean. Randomisation or other design quality was not among the coded variables, and no publication-bias analysis is reported; 393 of the 397 effects come from journal articles (pp. 236–237). Claim text, Status and kill-condition unchanged. Not a survival; does not count toward the floor; does not change C33's fruitless streak.
Evidence
4 linked sources: 2 supports, 2 limits (first-pass draft, pending human review). supports: Lacerenza et al. 2017, Arthur et al. 2003 · limits: Skylar Powell & Yalcin 2010, Taylor et al. 2005