---
title: 'Cultural Distance: Measurements and Models'
url: https://www.emergentmind.com/topics/cultural-distance
type: topic
---

# Cultural Distance: Measurements and Models

Searching arXiv for recent and foundational papers on cultural distance across social systems, migration, and LLMs.
arxiv_search.query({"search_query":"all:\"cultural distance\"","start":0,"max_results":10,"sort_by":"relevance","sort_order":"descending"})
arXiv search results for "all:\"cultural distance\"" returned several directly relevant papers, including:
- 2603.16827 — "Prompt Programming for Cultural Bias and Alignment of Large Language Models"
- 2311.14096 — "Cultural Bias and Cultural Alignment of Large Language Models"
- 2504.08443 — "The Influence of Culture on Migration Patterns"
- 2311.01040 — "The Roles of Culture in Online User Reviews: An Empirical Investigation"
- 2212.01336 — "The Influence of Cultural Distance on Settlement Intention of Floating Population in China"
- 2102.11398 — "Home and destination attachment: study of cultural integration on Twitter"
- 2007.02359 — "Cultures as networks of cultural traits: A unifying framework for measuring culture and cultural distances"
- 1301.2995 — "Measuring Cultural Dynamics Through the Eurovision Song Contest"
Searching for work connecting cultural distance to cultural boundaries and segregation dynamics.
arxiv_search.query({"search_query":"all:\"cultural boundaries\" OR all:\"cultural dynamics\" Axelrod leaders obstacles","start":0,"max_results":10,"sort_by":"relevance","sort_order":"descending"})
Relevant arXiv results include:
- 2009.09591 — "Leaders and obstacles raise cultural boundaries"
- 2007.02359 — "Cultures as networks of cultural traits: A unifying framework for measuring culture and cultural distances"
- 1301.2995 — "Measuring Cultural Dynamics Through the Eurovision Song Contest"
Cultural distance is a quantitative or relational measure of how far cultures, populations, or cultural representations are from one another. Across the literature, it is operationalized in markedly different ways: as dissimilarity between cultural vectors in agent-based models, as Euclidean distance between country or model positions in the Inglehart–Welzel value space, as variance-adjusted differences on Hofstede’s dimensions, as dialectal divergence, as divergence between joint probability distributions over cultural traits, or as a gap between a model’s “default cultural repertoire” and the situated demands of practice [2009.09591] [2603.16827] [2311.01040] [2007.02359] [2509.10780]. This plurality is not incidental. It reflects the fact that “culture” is alternately treated as a system of traits, a latent value profile, a network of dependencies, a repertoire of practices, or a field of behavioral affinities.

## 1. Conceptual range

In social-dynamical models, cultural distance is often implicit rather than named. In the modified Axelrod model of cultural dissemination, each active agent \(i\) has a cultural state \(c_i = (c_i^1,\dots,c_i^F)\), overlap \(l(i,j)\) counts shared features, and interaction occurs with probability \(l(i,j)/F\). A normalized similarity measure is \(s_{ij} = \frac{1}{F}\sum_{f=1}^F \delta_{c_i^f,c_j^f}\), with the corresponding Hamming-type cultural distance \(d_{ij} = 1 - s_{ij}\) [2009.09591]. In this formulation, cultural distance directly controls whether interaction and convergence are likely.

In cross-cultural psychology, political sociology, migration research, and management, cultural distance is usually defined at the country level. One strand places countries in a two-dimensional value space derived from the World Values Survey and European Values Study, especially the Inglehart–Welzel map; another uses Hofstede’s six dimensions and computes distance from differences in country scores [2311.14096] [2311.01040] [2504.08443]. In the World Values Survey tradition, distance is geometric within a culturally meaningful factor space. In Hofstede-based work, distance is typically composite across dimensions such as power distance, individualism, masculinity, uncertainty avoidance, long-term orientation, and indulgence.

Recent AI work extends the same idea to models themselves. Large language models are treated as respondents to survey instruments, projected into the same value space as human populations, and then evaluated by how close their induced cultural profile is to a target country or community [2603.16827] [2311.14096] [2309.12342]. A related HCI formulation defines cultural distance as “the gap between GenAI’s default cultural repertoire and the situated demands of teaching practice,” shifting the emphasis from country-level averages to task-specific misfit in local classrooms [2509.10780].

## 2. Measurement frameworks

The literature does not provide a single canonical metric. Instead, it offers a family of operationalizations tied to different objects: agents, countries, texts, neighborhoods, migrants, and models.

| Setting | Representation | Distance measure |
|---|---|---|
| Axelrod-type dynamics | Cultural vectors on a lattice | \(d_{ij}=1-\frac{1}{F}\sum_{f=1}^F \delta_{c_i^f,c_j^f}\) |
| Inglehart–Welzel mapping | Country/model points in \(\mathbb{R}^2\) | Euclidean distance |
| Hofstede-based eWOM | Six national dimensions vs Italy | Kogut–Singh index |
| OECD migration | Six-dimensional country vectors | Euclidean distance |
| China internal migration | Prefecture dialect composition | Population-weighted dialectal distance |
| Cultural trait networks | Joint pmf + dependency graph | Jeffreys’ divergence |
| Behavioral affinity | Eurovision voting bias | Friend-or-Foe coefficient |

A widely used value-space formulation embeds both human populations and models into the Inglehart–Welzel map. If country \(c\) has reference point \(\boldsymbol{\nu}^{\text{IVS}_{c} \in \mathbb{R}^2\) and model \(m\) has point \(\boldsymbol{\mu}_{m,c}\in\mathbb{R}^2\), then cultural distance is
\[
d(m,c) = \left\lVert \boldsymbol{\mu}_{m,c} - \boldsymbol{\nu}^{\text{IVS}_{c} \right\rVert_2.
\]
This is the central quantity in survey-grounded LLM alignment work [2603.16827].

A second major family uses Hofstede’s six dimensions. In online review research, cultural distance between guest country \(j\) and host country \(k\) is computed with the Kogut–Singh formula
\[
CD_{j} = \frac{1}{I}\sum_{i=1}^{I} \frac{(I_{ij} - I_{ik})^{2}}{V_i},
\]
where \(I_{ij}\) and \(I_{ik}\) are Hofstede scores and \(V_i\) is the variance of dimension \(i\) across countries [2311.01040]. In migration research, the same six-dimensional country profiles are instead treated as points in a Euclidean space:
\[
CD(C_1, C_2) = \sqrt{ (\Delta pdi_{C_1,C_2})^2 + (\Delta idv_{C_1,C_2})^2 + (\Delta mas_{C_1,C_2})^2 + (\Delta uai_{C_1,C_2})^2 + (\Delta ltowvs_{C_1,C_2})^2 + (\Delta ivr_{C_1,C_2})^2 }.
\]
This makes cultural proximity or remoteness directly comparable to geodesic distance in bilateral migration corridors [2504.08443].

A third approach replaces national-value scales with language as a proxy for regional culture. In the Chinese migration literature, dialectal distance between prefectures \(A\) and \(B\) is defined as
\[
d(A,B) = \sum_{i=1}^{I} \sum_{j=1}^{J} S_{Ai} \cdot S_{Bj} \cdot d_{ij},
\]
where \(S_{Ai}\) and \(S_{Bj}\) are county population shares and \(d_{ij}\) is branch-based linguistic distance between county dialects [2212.01336]. Here cultural distance is not inferred from attitudes but from the historical structure of dialect families.

The most elaborate reconstruction treats national cultures as networks of cultural traits. Using World Values Survey items as discrete variables and Gaussian copula graphical models for each country, cultural distance is defined as Jeffreys’ divergence between the estimated joint distributions:
\[
\begin{aligned}
JD(m,l)=& \sum_{i=1}^p \sum_{k=1}^{T_i} (f^{(m)}_{i}(t_k)-f^{(l)}_{i}(t_k)) \log \biggl(\frac{f^{(m)}_{i}(t_k)}{f^{(l)}_{i}(t_k)}\biggr)\\
&+ \dfrac{\mbox{Trace}(\mathbf{K}^{(l)}(\mathbf{K}^{(m)})^{-1})+\mbox{Trace}(\mathbf{K}^{(m)}(\mathbf{K}^{(l)})^{-1})}{2}-p.
\end{aligned}
\]
This decomposition separates a marginal component, based on trait distributions, from a network component, based on the dependence structure among traits [2007.02359]. A central implication is that two countries can be close in average trait levels yet distant in how those traits cohere.

Behavioral proxies constitute another measurement family. The Eurovision “Friend-or-Foe” coefficient
\[
FoF(c_v, c_c) = \frac{p_{v,c}}{12} - \frac{s_c - p_{v,c}}{12\,(N-2)}
\]
captures a directional voting bias after subtracting overall song support, and its long-run average is negatively related to Hofstede-based cultural distance [1301.2995]. This operationalization treats cultural distance not as self-reported values but as asymmetrical behavioral affinity.

## 3. Spatial structure, boundaries, and mobility

Cultural distance is often stabilized or amplified by spatial structure. In the two-dimensional Axelrod lattice with defects, obstacles and opinion leaders generate two collective phases: an ordered phase with one large homogeneous domain and a disordered phase in which many domains coexist [2009.09591]. The key order parameter is the normalized size of the largest domain, \(S_{\text{max}}\). In the multicultural phase, boundaries persist because interaction pathways are broken or pinned. The paper’s interface analysis shows that the critical line coincides, within errors, with the condition \(\rho_f/\rho = 1\), where defects become over-represented on interfaces. Above that threshold, obstacles and leaders serve as nucleation sites for cultural boundaries, stabilizing large inter-domain distances.

Migration research reaches a related conclusion by different means. Using UN migrant stock data and Hofstede’s six dimensions for 93 countries, one study finds that migrants from wealthier countries tend to select culturally similar destinations, whereas those from poorer countries often migrate to culturally distant destinations [2504.08443]. Approximately half of OECD countries also demonstrate a statistically significant bias towards accepting culturally close migrants. Another study, focused on China’s floating population, finds strong evidence for the negative effects of dialectal distance on migrants’ settlement intention; each 1-unit increase in dialectal distance reduces the odds of intending to settle by 22.9% in the main binary logistic specification [2212.01336]. These findings tie cultural distance to both destination choice and the decision to remain.

Urban mobility work generalizes the same logic below the national scale. A relational model of neighborhood mobility represents local culture through “scenes” and amenity mix, then measures scene similarity and amenity similarity between neighborhoods using cosine similarity [2512.11662]. The paper does not use the phrase “cultural distance” explicitly, but this suggests an implicit definition of neighborhood-level cultural distance as \(1-\text{SceneSim}_{ij}\). In both U.S. co-visitation data and Canadian residential moves, neighborhoods with similar cultural styles and amenities are significantly more connected even after controlling for race, income, education, politics, housing costs, and distance. This suggests that symbolic and functional similarity acts as a “soft infrastructure” that channels movement and reproduces clustering.

## 4. Behavioral manifestations in platforms and public life

Cultural distance also appears in platform-mediated behavior. In Airbnb reviews of Venice accommodations, guests from a relatively more distant culture rely less on heuristics: the positive effect of being a “superhost” on review sentiments is attenuated when the cultural distance between the consumer’s home country and Italy is high [2311.01040]. Here cultural distance moderates the use of quality signals rather than simply shifting mean sentiment.

Digital traces of migrants reveal a more ambivalent pattern. A Twitter-based study defines home attachment as
\[
HA(u) = \frac{HT(u, C_n(u))}{HT(u)},
\]
and destination attachment as
\[
DA(u) = \frac{HT(u, C_r(u))}{HT(u)},
\]
using country-specific hashtags as behavioral markers of orientation [2102.11398]. Larger differences among home and destination country in terms of Individualism, Masculinity and Uncertainty appear to correspond to larger destination attachment and lower home attachment. By contrast, a Facebook-advertising study of digital diasporas finds that greater host-native distance is linked to higher online cultural retention, while origin country context is statistically significant but generally smaller in impact [2511.17756]. Taken together, these papers indicate that cultural distance can either weaken home orientation or strengthen cultural retention, depending on whether the relevant behavior is public attention to destination issues or ongoing attachment to origin-country content.

The Eurovision literature adds a longitudinal public-sphere example. The Friend-or-Foe coefficient correlates negatively with Hofstede-based cultural distance, stronger modularity emerges in the FoF network than in raw vote networks, and a measure of polarization shows a sharp increase within EU countries during 2010 and 2011 [1301.2995]. This suggests that behavioral proxies can capture not only static affinity but also temporal shifts in perceived cultural closeness.

Textual analysis complicates the picture. Work on English-language fiction shows that textual distance and social or cultural distance are related but not identical: cosine similarity on tf–idf vectors or topic vectors correlates only moderately with genre proximity derived from library metadata, and supervised distances anchored in social context perform better [1807.00181]. This suggests that textual similarity should not be assumed to be a transparent proxy for cultural proximity without explicit control for confounds such as chronology.

## 5. Cultural distance in LLMs

Recent LLM research has made cultural distance a first-class evaluation target. One line of work projects both countries and

Source: https://www.emergentmind.com/topics/cultural-distance