How the FOLK Scores Were Built
Methodology note: the four orientations, the statistical baseline, and the AI-council scoring
The scores were produced in three stages: a statistical stage that defined the four orientations and a starting baseline number for each country; an AI council stage that reviewed the evidence and qualitative literature; and an integrator stage that set the final published score.
Defining the Four Orientations (Statistics)
Five established cultural frameworks were used as inputs: Hofstede, Trompenaars, GLOBE, Schwartz, and the World Values Survey (WVS). These overlap, so the first step was to combine them into one common set of scores.
Between them the five frameworks supply thirty-four cultural indicators. The starting point was the ~25 countries that had all five frameworks fully measured. On these, factor analysis (Multiple Factor Analysis) was run across all thirty-four indicators to find how many distinct underlying orientations were really present. It found four.
Every existing model was built top-down — a team formed a theory, designed measures, then collected data to test it. This was built bottom-up: the data from five independent research traditions was allowed to find its own structure, and what emerged was named afterwards. Each orientation groups the measures that move together, and each was named for the common thread they share.
FOLK Factor Structure Matrix (36 Indicators)
Cross-framework factor loadings across Hofstede (HOF), GLOBE (GLO), Schwartz (SCH), World Values Survey (WVS), and Trompenaars (TRO).
| FW | Attribute / Indicator | D1(Identity) | D2(Expression) | D3(Structure) | D4(Drive) | Assigned | Notes & Mechanism |
|---|---|---|---|---|---|---|---|
| D1 Identity | |||||||
| WVS | Choice | +0.959 | D1 | Self pole | |||
| WVS | Equality | +0.915 | D1 | Self pole | |||
| GLO | GLOBE UA — conformity pressure | -0.904 | D1 | Social pole (Empirical surprise) | |||
| SCH | Embeddedness | -0.892 | D1 | Social pole | |||
| HOF | Individualism | +0.872 | D1 | Self pole | |||
| HOF | Power Distance | -0.872 | D1 | Social pole | |||
| SCH | Affective Autonomy | +0.854 | D1 | Self pole | |||
| SCH | Intellectual Autonomy | +0.841 | D1 | Self pole | |||
| TRO | Ascription-Achievement | +0.841 | D1 | Self pole | |||
| TRO | Particularism-Universalism | +0.775 | D1 | Self pole | |||
| TRO | Diffuse-Specific | +0.756 | D1 | Self pole | |||
| WVS | Voice | +0.675 | +0.461 | D1 | Also Open: Voice = self-expression & open communication | ||
| SCH | Hierarchy | -0.646 | D1 | Social pole | |||
| TRO | Communitarianism-Individualism | +0.640 | D1 | Self pole | |||
| SCH | Egalitarianism | +0.591 | +0.683 | D1 | Also Open: flat social relations → individual autonomy + open expression | ||
| GLO | Gender Egalitarianism | +0.539 | +0.636 | D1 | Also Open: same mechanism as Egalitarianism | ||
| WVS | Defiance | +0.521 | D1 | Self pole | |||
| WVS | Autonomy | +0.501 | D1 | Self pole | |||
| GLO | Institutional Collectivism | -0.510 | +0.568 | D1 | Also Open (Social pole): group-oriented institutions → warm in-group expression | ||
| D2 Expression | |||||||
| HOF | Long-Term Orientation | -0.807 | D2 | Restrained pole | |||
| HOF | Indulgence | +0.797 | D2 | Open pole | |||
| TRO | Future-Past Orientation | -0.811 | -0.807 | D2 | Cross D1 Social pole: past-oriented cultures are both more Social and more Restrained | ||
| GLO | In-Group Collectivism | +0.625 | D2 | Open pole | |||
| GLO | Power Distance (practices) | -0.547 | -0.620 | D2 | Restrained pole. Also Accepting: hierarchical cultures tend toward acceptance of circumstance | ||
| WVS | Scepticism | +0.490 | D2 | Open pole | |||
| GLO | Humane Orientation | +0.455 | D2 | Open pole | |||
| GLO | Assertiveness | -0.494 | -0.573 | D2 | Restrained pole. Also Fluid: assertive cultures comfortable with ambiguity | ||
| D3 Structure | |||||||
| HOF | Uncertainty Avoidance | +0.858 | +0.333 | D3 | Certain pole. Also Striving: high-structure cultures include many high-achievement cultures | ||
| SCH | Harmony | +0.589 | D3 | Certain pole | |||
| GLO | Future Orientation† reassigned | +0.420 | D3 | † reassigned — planning as mechanism for managing uncertainty | |||
| D4 Drive | |||||||
| SCH | Mastery | +0.680 | D4 | Striving pole | |||
| WVS | Relativism | -0.644 | D4 | Accepting pole: moral flexibility = individual-level expression of Drive Accepting | |||
| GLO | Performance Orientation† reassigned | +0.425 | D4 | † reassigned — achievement orientation belongs in Drive | |||
| HOF | Masculinity | +0.231 | D4 | Striving pole | |||
| Not Assigned to a FOLK Dimension | |||||||
| TRO | Affective-Neutral | — | Spreads across D1 and D2. Closest to D2 Expression. | ||||
| TRO | External-Internal | — | Spreads across D1 and D2. Internal control = more Self and more Open. | ||||
Only loadings ≥ 0.20 are shown. D3 signs are flipped so Certain = positive throughout. Cross-loadings are included only where there is a clear conceptual reason. Attributes marked † were theoretically reassigned from their strongest statistical factor.
The two halves. The four orientations pair into two groups, and this is what the name FOLK encodes — Four Orientations of Life and Kinship. Kinship covers who you belong to and how you show it: Identity and Expression. Life covers how you organise the world and what you strive for: Structure and Drive.
| Orientation | Low → High pole | What it captures | Why it's called that |
|---|---|---|---|
| Identity | Social → Self | Whether a person's sense of self is anchored in family and community, or in the individual | It locates where identity sits — in the group or the person |
| Expression | Restrained → Open | How openly warmth, emotion and social energy are shown | It is about outward emotional and social expression |
| Structure | Fluid → Certain | How far rules, order and predictability are preferred over flexibility and improvisation | It is about the need for structure |
| Drive | Accepting → Striving | How much ambition, achievement and visible success are emphasised over contentment | It is about achievement drive |
This four-orientation structure, fixed on the complete-data countries, became the reference for everything else. Throughout, poles are written low first — Social to Self, Restrained to Open, Fluid to Certain, Accepting to Striving — so the wording runs in the same direction as the 0-100 scale and the score bars. Earlier framework material lists the high pole first; that ordering is superseded.
Sample Expansion & Framework Coverage
Most countries do not have all five frameworks, so the panel was extended as the minimum coverage requirement was relaxed:
| Frameworks present | Countries scored |
|---|---|
| All five frameworks | ~27 countries |
| At least four | 60 countries |
| At least three | 84 countries |
| At least two | 120 countries |
| At least one | 171 countries |
| No framework data (Extension) | 26 countries (Direct AI Council Evaluation) |
Missing frameworks (for countries with at least one) were filled by imputation (MICE, 40 draws), which estimates the missing values while leaving observed values unchanged; running it 40 times let the resulting uncertainty be measured. At each step the four-orientation structure was re-checked for stability with a similarity score (Tucker's φ): 0.95+ is "green," 0.85+ is "yellow."
Countries with no framework data at all. That statistical process could only reach the 171 countries with at least one framework. The remaining 26 of the 197 countries had no framework measurements to draw on, so they have no statistical baseline. Their scores were produced by the AI council directly — the same panel of expert agents working from evidence, research and qualitative literature, but starting from that evidence rather than from a statistical starting number. (In the data these appear as "recommended-only" records, since there is no baseline or interval behind them.)
Scores were placed on a 0–100 scale anchored so that the world average sits at 50.
A second, independent method (pattern-based regression with jackknife intervals) was run as a cross-check. It agreed strongly on Identity (rank correlation ≈ 0.97) and, once its axis direction was aligned, on Structure (≈ 0.89); agreement on Expression and Drive was moderate (≈ 0.70 each).
How the Source Frameworks Map In
All five frameworks are absorbed rather than replaced. Hofstede's six dimensions feed all four orientations — Individualism and Power Distance into Identity, Uncertainty Avoidance into Structure, Indulgence into Expression, and the dimension formerly labelled Masculinity into Drive. Schwartz anchors Identity through Autonomy and Embeddedness, and Structure through Harmony. The World Values Survey is the richest single source for Identity. All four Trompenaars dimensions load cleanly onto Identity. GLOBE spreads across all four.
One Empirical Surprise
GLOBE's Uncertainty Avoidance maps to Identity rather than to Structure — a meaningful distinction, and one that only appears when the frameworks are analysed together.
One Deliberate Override
GLOBE's Performance Orientation groups statistically with Expression, but its meaning — competitive achievement — belongs in Drive. It was moved on judgement. Statistics informed the interpretation; they did not settle it.
Uneven Anchoring
Identity and Expression are richly supported by many independent items across multiple frameworks. Structure and Drive rest on fewer primary anchors, particularly at their Fluid and Accepting poles.
Not a Replacement
The four orientations capture the common signal across the five sources. Anyone needing the granularity of a specific framework should use that framework directly.
The Anchor Score, and the Remaining Issues
Each country ended this stage with two things per orientation: a baseline score and a confidence interval (for example, Bhutan Identity: baseline 45, interval ≈ 28.6–57.2). The interval indicated how much confidence to place in the score. It was originally intended to constrain the final number, but that constraint was ultimately not applied.
The statistical scores had three issues:
Scores clustered near the middle of the scale (50), leaving much of the 0–100 range unused.
Some countries received the exact same value on an orientation.
A few countries had nearly identical scores on all four orientations, caused by heavy imputation.
Further statistical adjustment would have been judgement dressed as maths. Instead, an AI council was used to review and adjust the scores against evidence. This resolved the compression: guided by evidence and no longer held near the mid-scale, scores were spread across much more of the 0–100 range. Identical and flat cases were separated individually on the evidence.
How the Council is Built (Multiple Roles, Multiple LLMs)
Rather than one AI producing one number, the council uses several agents, each assigned a different expert role, run across more than one underlying language model. Using different models avoids a single model simply agreeing with itself.
Kinship, ritual, custom, and everyday practice.
The state, law, religion, and how rules and power operate.
Keeps the score consistent with quantitative data and baseline.
Checks the score against regional peer groups and similar countries.
Supplies qualitative nuance and country-specific detail.
Argues against the emerging answer and finds the weakest point.
Each agent receives the same brief: the statistical scores and their ranges, the source evidence and citations, academic research, and qualitative literature. Each proposes a score per orientation and records its strongest supporting evidence, its strongest opposing evidence, the main weakness in its own case, an alternative score, and a confidence level.
The agents then deliberate; disagreements are logged, and each agent's confidence determines how much its score counts. Two further agents complete the process: an Integrator, which makes the final decision, and a Judge, which checks that decision against the evidence. The design was adapted from a system originally built for stock-movement decisions.
How the Debate Becomes One Number
The council's inputs are converted into a score in a fixed order. Most of it is arithmetic; the integrator LLM acts only at the end.
Two Illustrative Integrator Decisions
Recommended 49.5; the integrator judged Identity lower based on grounded qualitative evidence and set it near 28, which became final.
Recommended 46.4; the integrator set 38 based on deep institutional analysis, which became final.
The Three Key Numbers
Confidence-weighted average of all agent scores.
The deterministic, formula-only number; fully reproducible.
The integrator's decision, published as-is. The confidence interval is retained as a measure of certainty.
34 Indicators • 197 Countries • 1 Universal Language
Cross-framework factor analysis over thirty-four indicators defined the four orientations and a baseline score for each country; a council of AI agents across multiple models reviewed the evidence; and an integrator set the final score. Of the 197 countries, 171 were scored from at least one framework and 26 by the council alone from evidence where no framework data existed.