Contribution
The paper (LIT-496) gives morality-as-cooperation its own instrument. It copies the MFQ’s two-part format: a Relevance scale (“when you decide whether something is right or wrong, to what extent are the following considerations relevant”) and a Judgement scale (agreement with moral statements). It writes items for the seven domains, and in three online samples tests whether seven factors emerge and whether the MFQ’s five do. It also states the theory’s structural prediction precisely. Each type of cooperation gives a distinct factor, and “behaviour not tied to a specific type of cooperation will not constitute a distinct moral domain” (p. 9). Care and purity are named as the domains expected to fail (pp. 10–11).
Key insight
The paper converts a claim about content (cooperation is moralized) into a claim about structure (endorsement of the seven varies independently). The structural claim is the one a questionnaire can test, and the one that distinguishes MAC from MFT on the domains they share (group, deference, fairness). The content claim, that nothing else is moralized, needs items the MAC-Q does not contain.
Assumptions
- Items from the theory. The authors wrote 57 Relevance and 97 Judgement items “based on a comprehensive review of the game-theoretic … literature” and focused on “the principal moral value in each of the seven hypothesised domains” (p. 18). They then kept three items per subscale, chosen for theoretical consistency and item-total correlation within the subscale (p. 19). A seven-factor solution is partly built into this selection.
- Virtue-term wording. The items describe the behaviours as virtues or their violations: “acted heroically”, “showed love for their family”, “kept their promise”. This is the same feature NOTE-034 found in LIT-047’s codebook.
- Samples. UK and US online panels, paid £0.5 and $0.75. Exclusions were heavy. In Study 1, 2,396 people opened the survey and 1,392 were kept; the listed exclusions (544 inattentive, 151 incomplete, 305 spoiled) sum to 1,000, not 1,004. In Study 3a, 688 of 1,157 were excluded, 537 of them for inattention.
- Fit criteria. RMSEA below .05 good, CFI above .90 acceptable, SRMR below .08 good (endnote x).
Key results
- Study 1 (UK, n = 1,392, September 2015).
- The EFA gives seven factors on both scales, all items loading highest on their own factor.
- CFA, Relevance: RMSEA .03, CFI .98, SRMR .03.
- CFA, Judgement: RMSEA .06, CFI .87, SRMR .07. The authors’ own criteria rate that CFI “unacceptable”.
- Combined (Table 9): the seven-domain “Simple Domains” model has CFI .73 and RMSEA .07. The “Different but Related” model, with fourteen correlated latent variables (seven per scale), has CFI .93 and RMSEA .04. The multi-trait multi-method model has CFI .90.
- Alphas: .76–.86 on Relevance, .53–.83 on Judgement (Property .53).
- Study 2 (MFQ, UK, n = 1,042, pilot data 2013–2014).
- Relevance: parallel analysis gives three factors (Care/Fairness, Ingroup/Authority, and a single-item Disgust factor).
- Judgement: five factors, but not MFT’s five (Care/Fairness, Purity, Patriotic Authority, Family, Gender Roles).
- No five-factor CFA fits on all criteria.
- Relevance means on the 1–6 scale: Care 4.63, Fairness 4.46, Authority 3.89, Purity 3.80, Ingroup 3.77 (Table 12).
- Study 3a (US, n = 469, June–July 2016).
- MAC Relevance gives seven factors.
- MAC Judgement gives five by parallel analysis (Family, Group/Reciprocity, Martial Virtues, Fairness, Property), and seven only when seven are forced.
- MFQ Relevance gives two factors (Care/Fairness and Ingroup/Authority/Purity).
- MFQ purity Relevance mean 4.19, against Ingroup and Authority at 4.31 (Table 28). MFQ purity Judgement α = .75, the highest of the MFQ Judgement subscales.
- Criterion validity is mixed. On MAC Relevance only Family and Deference were the best predictors of their own criterion scales, and Property did not correlate with its criterion (p. 35).
- Incremental validity: the MAC-Q predicts better for family, group, heroism and property; the MFQ for deference and fairness. For self-reported cooperative actions, the MAC-Q was better in 10 cases, equal in 3 and worse in 1.
- Study 3b (retest after one month, n = 137). MAC-Q .79–.89 on Relevance and .66–.87 on Judgement; MFQ .70–.80 and .46–.75.
- Data. Not shareable, because the consent form “inadvertently precluded data sharing” (endnote vii). Materials and syntax are on OSF (osf.io/w5ad8).
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Each of the seven types of cooperation is regarded as morally relevant | moderate | Relevance means run from “somewhat” to “very” relevant in both samples. With virtue-worded items, a low rating was unlikely |
| C2 | Each type of cooperation is a distinct moral domain | moderate for Relevance; weak for Judgement and for the combined scale | seven factors on Relevance in two samples. Judgement gives five in the US sample, and the combined seven-domain model fails (CFI .73) |
| C3 | Behaviour not tied to a type of cooperation (care, purity) does not form a distinct domain | weak | shown only with the MFQ’s items, whose low reliability the paper itself documents. Purity is a distinct Judgement factor in Study 2, and its α is the MFQ’s best in Study 3a |
| C4 | Cooperative behaviour is “the” content of morality; purity’s relevance “remains to be explained” | conceded against | MFQ purity items are rated relevant at the level of loyalty and authority; the authors say this “appear[s] to contradict” MAC (p. 27) |
| C5 | The MAC-Q’s coverage, reliability, structure and fit are superior to the MFQ’s | moderate | true of alphas, retest and Relevance fit; criterion and incremental validity are “broadly comparable” by the authors’ own summary (p. 38) |
| C6 | MAC “identif[ies] for the first time” moral domains of family, reciprocity, heroism and property | moderate as a measurement result; weak as discovery | the domains were written into the item pool, so their recovery is a check on the items, not an independent finding |
Concepts
- Relevance vs Judgement. The MFQ’s two item formats, kept here. The paper’s strongest methodological finding is that they should not be pooled without modelling their difference.
- Simple Domains / Different but Related / MTMM. Three ways to combine the two formats: seven factors; fourteen correlated factors; or seven traits plus two method factors.
- Theory of dyadic morality, side-taking theory. Rivals the paper reviews (pp. 12–13), with relational models, before testing only against MFT.
Connections
- The sixty-society test (LIT-047). NOTE-034’s verdict, that the test is
weakly diagnostic, is not changed by this paper. It is reinforced at one
point and confirmed at another.
- Reinforced: the items, like the codebook, are phrased in virtue terms.
- Confirmed: the paper has no non-cooperative items, so it is no answer to the missing comparison domain. Where non-cooperative items do appear (the MFQ’s purity items), they are moralized.
- The founding statement (LIT-503). That chapter predicted that non-cooperative behaviour would be morally neutral. Studies 2 and 3a are the only place in the programme’s own work where that prediction meets data, through the MFQ’s purity items, and the authors report it failing.
- Purity (LIT-101). The paper’s reply is that purity is not a coherent domain, and that disgust may be a proximate mechanism serving several cooperation problems (p. 40, endnote xv). Endnote xv points to the sixty-society Reply’s sketch of a cooperative sexual morality. Fitouchi, André and Baumard are the worked version of this reply, at least for puritanical norms. They exclude contamination purity by definition. They also reduce what remains to harm and fairness concerns, while this paper argues that harm is not a domain. Their account rescues “morality is about cooperation”. It does not rescue “morality is these seven domains”.
- Moral molecules (LIT-509). The limitations section announces the 21 second-order “moral molecules” (pp. 40–41) that the later paper develops.
- The Turkish validation (LIT-513). An independent team reproduced the main result: seven factors fit Relevance and fail on Judgement and the full form.
- Not held here. The MFQ (Graham et al. 2011), relational models, dyadic morality and side-taking theory are being filed in parallel by other contributions.
Bearing on the record
No ML practice, and the anthology does not hold the paper. For the morality-as-cooperation thread it is the evidence for MAC’s structure. It is not evidence for MAC’s scope. Citing it for “morality is about cooperation” overstates it, since it never asks about anything else. It is also the programme’s own admission that purity is moralized.
Limitations
- Self-built items. The domains were written into the items, which were then selected by within-domain fit, so recovering the seven is partly a check on construction.
- No non-cooperative items, so the content claim is untested.
- Judgement scale. Fit is unacceptable by the paper’s own CFI criterion in both samples, and five factors emerge in the US sample. The Property Judgement items are all reverse-coded, which the authors flag as a possible valence confound (p. 42). Equality is the only fairness measured.
- Two WEIRD samples with heavy exclusions.
- Data cannot be shared, so no reanalysis is possible.
- Overstatement. The Study 1 discussion calls the seven-factor model “a good fit for both scales” (p. 23), but the Judgement CFI is .87.
Open questions
- With harm, purity, honesty and religious-observance items added to the pool, do they load on MAC’s factors, form their own, or scatter?
- Does the seven-factor Relevance structure hold with items written by people outside the programme?
- Why do Relevance and Judgement diverge? The paper leaves it open (p. 40).
Corrections
- none to a seeded skim (there was no seed)
- The citation. The owner’s citation is correct: Curry, Jones Chesters and Van Lissa (2019), Journal of Research in Personality 78:106–124. The published title is hyphenated (“morality-as-cooperation”). The PsyArXiv manuscript’s is not (“morality as cooperation”).
- What the MFQ comparison shows. It is often summarised as the MAC-Q beating the MFQ. On fit and reliability it does. On criterion and incremental validity the authors call the two “broadly comparable” (p. 38).