The survey offers a cross-national attitude map for topics where parameterizations are expected to diverge.
Evidence brief
What Pew can and cannot support
Pew is useful for checking whether country-level moral attitudes track broad Authority, Sanctity, Care, Fairness, and Liberty expectations. It is not an individual moral-profile dataset and it does not provide relational scenario graphs.
The page frames Pew as a stress test for parameterization, not as a complete moral-reasoning benchmark.
The five topics are informative, but they do not replace richer dilemma and stakeholder data.
Dataset shape
Nationally representative moral attitudes, worldwide
The Pew Research Center's Global Attitudes survey covers 25 countries across six continents. For each country, nationally representative samples were asked whether five specific behaviors are morally acceptable, morally unacceptable, or not a moral issue. This is not a Reddit thread. It is rigorous social science with stratified sampling and known methodology.
What makes this dataset uniquely valuable for MHF is the topic selection. These five issues are where the Christian and secular parameterizations are expected to diverge most dramatically. A system that cannot distinguish Sweden's moral landscape from Pakistan's on these topics has not yet shown robust cultural parameterization.
25
Countries surveyed
5
Hot-button moral issues
6
Continents represented
3
Response options
Issue map
Precisely where worldviews collide
These are not random moral topics. Each one sits at the intersection of specific Haidt moral foundations. Authority and Sanctity dominate the conservative-religious position; Care and Liberty dominate the progressive-secular one. MHF predicts this structure; Pew offers empirical data for checking it.
Abortion
Sanctity + Care + Authority
The deepest divide. Christian parameterization weights Sanctity (body as temple, imago Dei) and Authority (divine command). Secular weights Care (for the mother) and Liberty (bodily autonomy).
Alcohol Use
Sanctity + Authority
A Sanctity issue in Islamic and some evangelical contexts. Not a moral issue at all in most secular Western frames. The divergence is not about degree -- it is about whether it registers as moral.
Homosexuality
Sanctity + Authority + Fairness
Three foundations in direct tension. Conservative-religious frames cite Sanctity and Authority. Progressive-secular frames cite Fairness and Liberty. The same person can score differently depending on which foundation dominates.
Gambling
Sanctity + Care
Often dismissed in secular contexts as personal choice. In high-Sanctity cultures, it is degradation. In high-Care contexts, it is harm to family. MHF captures both angles.
Divorce
Loyalty + Sanctity + Care
Loyalty (to the vow) and Sanctity (of the institution) oppose Care (for people in harmful marriages). How a culture weights these three determines whether divorce is tragedy, sin, or liberation.
Cross-national data
How 25 countries see these five issues
The heatmap below shows the percentage of each country's population that considers each behavior "morally acceptable." Green = higher acceptance. Red = lower acceptance. The gradient from Pakistan to Sweden is not random. It tracks Authority + Sanctity foundation strength with remarkable consistency.
Moral Acceptability by Country and Issue
Percentage saying behavior is "morally acceptable" (representative estimates from Pew data)
Country
Abortion
Alcohol
Homosexuality
Gambling
Divorce
Sweden
71%
83%
86%
44%
73%
Germany
62%
78%
76%
32%
65%
Canada
60%
74%
68%
40%
67%
France
64%
76%
77%
36%
68%
UK
50%
72%
64%
33%
60%
USA
49%
67%
60%
35%
61%
Japan
47%
72%
49%
29%
51%
Brazil
22%
45%
28%
18%
43%
South Korea
44%
60%
35%
25%
50%
India
15%
22%
15%
13%
25%
Nigeria
12%
22%
7%
14%
18%
Egypt
8%
4%
3%
6%
13%
Pakistan
5%
3%
2%
4%
9%
Read the gradient. Sweden to Pakistan is not a random ordering. It tracks the Authority + Sanctity dimensions of Haidt's moral foundations with striking consistency. Countries high in these foundations find all five behaviors less acceptable. This is what MHF's parameterization predicts should happen: change the theta vector, change the judgment.
MHF prediction
The Haidt profile prediction
MHF makes a specific, testable claim: countries with high Authority + Sanctity Haidt profiles should produce moral judgments closer to the Christian parameterization. Countries with high Care + Fairness + Liberty profiles should match the secular parameterization. Pew gives us comparison data for that claim.
Authority + Sanctity Score vs. Moral Conservatism
Left = Low Auth+Sanct (secular profile). Right = High Auth+Sanct (Christian profile). Bar length = average "morally unacceptable" rating across all 5 issues.
Sweden
18%
France
24%
Canada
26%
USA
36%
Japan
Japan
40%
62%
78%
85%
94%
Predicted Parameterization Match
Which MHF parameterization should each country's Pew data most closely resemble?
● Should Match Secular Params
SwedenHigh Care + Fair
FranceHigh Care + Fair
GermanyHigh Care + Fair
CanadaHigh Care + Fair
UKHigh Care + Fair
USAMixed profile
JapanLow Sanct, high Loyal
● Should Match Christian Params
EgyptHigh Auth + Sanct
PakistanHigh Auth + Sanct
NigeriaHigh Auth + Sanct
IndiaHigh Auth + Sanct
BrazilHigh Sanct, mod Auth
S. KoreaMixed profile
Note something subtle: this is labeled "Christian params" but countries like Egypt and Pakistan are majority Muslim. The prediction still works because MHF's Christian parameterization captures the Authority + Sanctity pattern that Abrahamic religions share. A future Islamic parameterization would have different constraint content but similar Haidt profile weights. The architecture generalizes.
Strengths & Weaknesses
Honest assessment
✓ Strengths
- Nationally representative. Stratified sampling with known methodology -- not a convenience sample
- Focused topic alignment. These five issues are where Christian/secular params are expected to diverge
- Cross-national comparability. Same questions, same methodology, 25 countries
- Recent data. March 2026 survey captures current moral attitudes, not decade-old snapshots
- Three-way response. "Not a moral issue" is distinct from "acceptable" -- this matters for framework calibration
⚠ Weaknesses
- Only 5 issues. Narrow topical coverage -- does not test everyday dilemmas like AITA does
- No individual profiles. Unlike UniMoral, no per-respondent Haidt data -- only country-level aggregation
- No relational context. "Is abortion acceptable?" does not specify whose, why, or under what circumstances
- Country =/= culture. India and Nigeria contain enormous internal moral diversity that national averages obscure
- Self-report bias. Social desirability effects may suppress honest responses on taboo topics
Reader takeaway
Why Pew matters for moral AI
Every moral AI system trained on English data will default to the moral intuitions of the upper-left corner of that heatmap: Sweden, France, Canada. These are fine moral intuitions. They are also the intuitions of about 500 million people on a planet of 8 billion.
Pew's data is MHF's reality check. If the framework cannot reproduce the Sweden-to-Pakistan gradient by changing the theta vector, the parameterization is wrong. If it can, that supports a claim the field needs to test carefully: a moral reasoning system should not mistake one culture's moral consensus for universal truth.
That gradient -- from "not a moral issue" to "morally abhorrent" on the same behavior -- is not a problem to be solved. It is the human condition. MHF does not try to resolve it. It makes it visible, parameterizable, and auditable. Pew gives us comparison data to test whether that works.