Literature survey: does community size drive polarization?

Essay: Larger Communities Breed Polarization — “A peaceful world is a disconnected one.” Compiled: 2 September 2026 Companion model: influence.jl (bounded-confidence opinion dynamics; contrasts random interaction with nearest-neighbour / homophilous interaction).

0. The short version

The essay’s thesis — that a larger pool of potential associates lets people sort into like-minded clusters, and that those clusters then push their members to the extremes — is a recombination of three well-established research literatures:

1. Opportunity → homophily. When people have more choice of associates, they exercise finer-grained preferences for similarity. Friendships on large campuses are more ideologically and behaviourally matched than on small campuses, even though the large campus is objectively more diverse (Bahns, Pickett & Crandall 2012; Bahns et al. 2017). Same mechanism in racially diverse schools (Moody 2001) and immigrant enclaves (Borjas — see §5).
2. Homophilous discussion → extremity. Deliberation among like-minded people produces group polarization: members move further in their pre-existing direction and become more confident and more internally uniform (Sunstein 2002, 2009; Schkade, Sunstein & Hastie 2007). Cross-cutting exposure does not reliably fix this and can backfire (Bail et al. 2018).
3. Formal opinion-dynamics models show that whether a population converges or fragments depends on interaction structure and openness — and, directly relevant here, that the number of stable opinion clusters grows with population/group size (Kou et al. 2012; Hegselmann & Krause 2002; Flache et al. 2017).

Two framing caveats worth engaging with in the essay:

• Individual-level polarization is rarer than system-level polarization. On Reddit the 2016 shift was driven by new users sorting in, not by existing users radicalising (Waller & Anderson 2021); in US survey data most issue attitudes move in parallel rather than pulling apart (Baldassarri & Bearman 2007). The “engine” is often compositional sorting, not conversion — which is exactly what a size-and-choice argument predicts.
• The “Big Sort” geographic-sorting claim is contested (Bishop 2008 vs. Abrams & Fiorina 2012), though newer big-data work revived it (Brown & Enos 2021).

1. The direct precedent — choice, opportunity, and friendship similarity

This is the cluster the half-remembered Financial Times piece belongs to.

The FT article

Harford, T. (2024). “Why friends are always right — no matter their views.” Financial Times, 23 February 2024. → papers/harford-2024-why-friends-are-always-right-FT.md (full text saved).

Harford’s “two-part engine of polarisation”: (1) given the choice, we seek out people like us; (2) being surrounded by them makes us more extreme and more certain. He builds it from the Sunstein Colorado experiment (§2) and the Bahns campus-size study (below). Opens by citing John Burn-Murdoch’s FT data journalism on the young men / young women ideological divergence — a live, international example of a large connected population sorting apart.

Bahns, A. J., Pickett, K. M., & Crandall, C. S. (2012). Social ecology of similarity: Big schools, small schools and social relationships. Group Processes & Intergroup Relations, 15(1), 119–131.

→ papers/bahns-pickett-crandall-2012-social-ecology-similarity.pdf · DOI 10.1177/1368430211410751

The keystone paper. “Free-range dyad harvest”: approached pairs of people already socialising at the University of Kansas (~25,000 students) vs. small Kansas colleges (~100–500 students), measured attitudes, values, prejudice, and health behaviours. Friend dyads at the large university were significantly more similar than dyads at the small colleges, across attitudes, beliefs and behaviours. Large-campus students also reported greater “relational mobility” (ability to enter and leave relationships).

Interpretation: a bigger choice set lets latent similarity preferences express themselves at finer grain. This is the empirical core of the essay’s thesis.

Bahns, A. J., Crandall, C. S., Gillath, O., & Preacher, K. J. (2017). Similarity in relationships as niche construction: Choice, stability, and influence within dyads in a free choice environment. Journal of Personality and Social Psychology, 112(2), 329–355.

→ papers/bahns-etal-2017-similarity-niche-construction.pdf · DOI 10.1037/pspp0000088

11 samples, 1,523 interacting pairs. Tests why friends are similar: selection (“niche construction”) vs. social influence. Finds similarity is mostly selection at the point of formation — it does not grow with closeness, discussion, or relationship length. Important nuance for the model: in this dataset the homophily is front-loaded (who you pick), not an ongoing convergence process. influence.jl’s influence step is the second mechanism (§2), which other work supports more strongly for attitudes under group discussion.

McPherson, M., Smith-Lovin, L., & Cook, J. M. (2001). Birds of a feather: Homophily in social networks. Annual Review of Sociology, 27, 415–444.

→ papers/mcpherson-etal-2001-birds-of-a-feather.pdf

The canonical homophily review. Distinguishes baseline homophily (induced by the composition of the pool) from inbreeding homophily (preference beyond what the pool forces). Key line for the essay: homophily in larger, more differentiated settings is limited less by opportunity and more by preference — so bigger, more heterogeneous populations permit more preference-driven sorting, not less.

Moody, J. (2001). Race, school integration, and friendship segregation in America. American Journal of Sociology, 107(3), 679–716.

JSTOR: https://www.jstor.org/stable/10.1086/338954 (paywalled — no local copy)

Friendship segregation by race peaks in moderately diverse schools and, crucially, rises again once a minority group is large enough to form a self-sufficient community: students then find enough same-group friends and stop forming cross-group ties. This is the size-threshold (“critical mass”) version of the essay’s argument, with race standing in for ideology. Also: larger schools show more homophily at equal preference (opportunity effect).

2. Homophilous discussion → extremity (group polarization / echo chambers)

Sunstein, C. R. (2002). The law of group polarization. Journal of Political Philosophy, 10(2), 175–195.

→ papers/sunstein-2002-law-of-group-polarization.pdf (text = the 1999 U. Chicago Olin working paper, identical content)

States the “law”: deliberation moves a like-minded group toward a more extreme point in the direction it was already leaning. Two mechanisms — (a) limited argument pools (a like-minded group hears mostly arguments for one side) and (b) social comparison (people adjust toward the group’s perceived valued position). Explicitly extends the argument to the internet and self-selected media.

Schkade, D., Sunstein, C. R., & Hastie, R. (2007). What happened on Deliberation Day? California Law Review, 95, 915–940.

California Law Review: https://www.californialawreview.org/print/what-happened-on-deliberation-day (no clean local PDF)

See also Schkade, Sunstein & Hastie (2010), “When deliberation produces extremism,” Critical Review, 22(2–3), 227–252.

The Boulder / Colorado Springs experiment Harford cites. Liberals from Boulder and conservatives from Colorado Springs discussed climate, affirmative action, and civil unions in politically homogeneous groups. After ~15 minutes: both groups moved to the extreme, became more internally uniform, and the gap between the two towns widened. A clean demonstration that grouping like with like is itself polarising.

Sunstein, C. R. (2009). Going to Extremes: How Like Minds Unite and Divide. Oxford University Press.

→ papers/sunstein-2009-going-to-extremes-ch1-polarization.pdf (Chapter 1)

Book-length treatment; Ch. 1 walks through the Colorado study and the general theory in accessible prose. Good source for a quotable framing.

Bail, C. A., et al. (2018). Exposure to opposing views on social media can increase political polarization. PNAS, 115(37), 9216–9221.

→ papers/bail-etal-2018-exposure-opposing-views-polarization.pdf

Field experiment: paying Twitter users to follow a bot retweeting the other side made Republicans more conservative and (non-significantly) Democrats more liberal. The naive fix — “just expose people to the other side” — backfires. Relevant to the essay’s implied policy corollary (that forced mixing may not de-polarize).

Iyengar, S., Lelkes, Y., Levendusky, M., Malhotra, N., & Westwood, S. J. (2019). The origins and consequences of affective polarization in the United States. Annual Review of Political Science, 22, 129–146.

→ papers/iyengar-etal-2019-affective-polarization.pdf

Review of affective polarization (inter-party animus rather than issue extremity). Identifies partisan social identity + a media/geographic environment that lets partisans avoid the other side as key drivers. Useful for separating “opinions spread apart” from “people come to dislike each other,” which the essay may want to keep distinct.

3. Formal models of opinion dynamics (what influence.jl is doing)

Hegselmann, R., & Krause, U. (2002). Opinion dynamics and bounded confidence: Models, analysis and simulation. JASSS, 5(3), 2.

→ papers/hegselmann-krause-2002.pdf

The HK bounded-confidence model: agents average the opinions of others within a confidence radius ε. Large ε → consensus; small ε → multiple stable clusters (“polarization or fragmentation”). influence.jl’s theta(x,k) = 1/(1+(k-x)^2) is a smooth version of the same idea (weight decays with opinion distance instead of a hard cutoff).

Deffuant, G., Neau, D., Amblard, F., & Weisbuch, G. (2000). Mixing beliefs among interacting agents. Advances in Complex Systems, 3(1–4), 87–98.

→ papers/deffuant-etal-2000-mixing-beliefs.pdf

The other founding bounded-confidence model (pairwise random encounters + partial adjustment). High threshold → one central cluster; low threshold → several. Directly comparable to the use_closest_neighbors = false branch of influence.jl.

Kou, G., Zhao, Y., Peng, Y., & Shi, Y. (2012). Multi-level opinion dynamics under bounded confidence. PLoS ONE, 7(9), e43507.

→ papers/kou-etal-2012-multilevel-opinion-dynamics.pdf

Most directly on “size.” With heterogeneous confidence levels, the number of final opinion clusters is approximately a linearly increasing function of group size, holding the confidence bound fixed. i.e. bigger populations fragment into more distinct camps — a formal statement of the essay’s claim. (Note: a separate strand finds pure sample size negligible when agents are redrawn from the same distribution — the size effect needs either a fixed population with fixed neighbourhoods or heterogeneity.)

Flache, A., Mäs, M., Feliciani, T., Chattoe-Brown, E., Deffuant, G., Huet, S., & Lorenz, J. (2017). Models of social influence: Towards the next frontiers. JASSS, 20(4), 2.

→ papers/flache-etal-2017-models-social-influence.pdf

Authoritative review of assimilative, similarity-biased, and repulsive influence models, and why homogeneous influence tends toward consensus while similarity-biased influence (exactly influence.jl’s nearest-neighbour branch) sustains persistent clustering. Best single citation to situate the companion model in the literature.

Mäs, M., Flache, A., & Helbing, D. (2010). Individualization as driving force of clustering phenomena in humans. PLoS Computational Biology, 6(10), e1000959.

→ papers/mas-flache-helbing-2010-individualization.pdf

Adds a “need for distinctiveness” noise term. Produces a metastable regime: consensus within clusters, diversity between them — the structure the essay describes.

Mäs, M., & Flache, A. (2013). Differentiation without distancing: Explaining bi-polarization of opinions without negative influence. PLoS ONE, 8(11), e74516.

→ papers/mas-flache-2013-differentiation-without-distancing.pdf

Shows you can get bi-polarization from positive influence + homophily alone — you do not need people to actively repel opposing views. Supports a gentler reading of the essay’s mechanism.

Baldassarri, D., & Bearman, P. (2007). Dynamics of political polarization. American Sociological Review, 72(5), 784–811.

→ papers/baldassarri-bearman-2007-dynamics-political-polarization.pdf

Model + GSS data. Explains the “absence and presence” paradox: takeoff polarization on a few salient issues, parallel movement on most. Interaction structure (who talks to whom about what) governs whether an issue takes off. Good counterweight against overclaiming.

Axelrod, R. (1997). The dissemination of culture: A model with local convergence and global polarization. Journal of Conflict Resolution, 41(2), 203–226.

JSTOR: https://www.jstor.org/stable/174371 (paywalled)

Classic: local assimilation on a lattice still leaves stable distinct cultural regions. Larger / more feature-rich spaces sustain more regions. Historical anchor for the modelling section.

DellaPosta, D., Shi, Y., & Macy, M. (2015). Why do liberals drink lattes? American Journal of Sociology, 120(5), 1473–1511.

→ papers/dellaposta-shi-macy-2015-liberals-lattes-LSE-summary.pdf (LSE blog summary) · papers/dellaposta-2018-real-reason-liberals-drink-lattes-contexts.pdf (Contexts, popular write-up). Full article paywalled: https://doi.org/10.1086/681254

Homophily + influence in a large network turns uncorrelated tastes and opinions into a tightly bundled “lifestyle politics” package — “liberals drink lattes because their friends do.” Mechanism for how a large connected population manufactures super-issues.

See also DellaPosta, D. (2020), “Pluralistic collapse,” American Sociological Review 85(3), 507–536 — https://doi.org/10.1177/0003122420922989 (paywalled).

4. Scale of the political unit: size, participation, and democracy

Dahl, R. A., & Tufte, E. R. (1973). Size and Democracy. Stanford University Press.

(book — not in folder)

Foundational: a trade-off between citizen effectiveness (better in small units) and system capacity (better in large units). Larger polities dilute the individual voice, which the essay could tie to disengagement and factionalisation.

Oliver, J. E. (2000). City size and civic involvement in metropolitan America. American Political Science Review, 94(2), 361–373. — and Democracy in Suburbia (Princeton UP, 2001).

APSR: https://doi.org/10.2307/2586017 (paywalled)

Civic and political participation declines as municipality size rises. But there is a twist directly on-theme: homogeneous small towns also depress participation, because there is nothing to argue about. Conflict-free community = disengaged community — a near-restatement of “a peaceful world is a disconnected one,” from the opposite direction.

Centola, D., Becker, J., Brackbill, D., & Baronchelli, A. (2018). Experimental evidence for tipping points in social convention. Science, 360(6393), 1116–1119.

→ papers/centola-etal-2018-tipping-points-social-convention.pdf · DOI 10.1126/science.aas8827

Controlled experiments on naming conventions: a committed minority flips the group norm once it reaches ~25% of the population, and fails below that. Two implications for the essay: (a) norm change is threshold-driven, so a sub-group that reaches critical mass can lock in and enforce its own convention rather than compromise; (b) in a larger community the absolute number needed to reach 25% of any given interaction neighbourhood is larger, so established positions are stickier and coordination on a shared norm is harder.

Dunbar, R. I. M. (1992). Neocortex size as a constraint on group size in primates. Journal of Human Evolution, 22(6), 469–493.

Plus Dunbar (1993, Behavioral and Brain Sciences 16); Hill & Dunbar (2003, Human Nature 14). (paywalled — no local copy)

“Dunbar’s number” (~150): a cognitive ceiling on stable personal relationships. If real, any community >1,000 must be navigated through sub-groups and categories rather than personal acquaintance — a mechanism for why large communities fracture into identity blocs. Note the critiques: Lindenfors, Wartel & Lind (2021), “‘Dunbar’s number’ deconstructed,” Biology Letters 17 — https://doi.org/10.1098/rsbl.2021.0158 — argue the number is statistically unsupported. Cite carefully.

5. Geography, density, and political sorting

Bishop, B., & Cushing, R. G. (2008). The Big Sort: Why the Clustering of Like-Minded America Is Tearing Us Apart. Houghton Mifflin.

(book) The popular statement of the residential-sorting thesis: Americans increasingly live among the politically like-minded, and homogeneous localities drift to the extremes (explicitly invoking group polarization).

Rebuttal to engage: Abrams, S. J., & Fiorina, M. P. (2012). “The Big Sort that wasn’t: A skeptical reexamination.” PS: Political Science & Politics, 45(2), 203–210 — partisan residential segregation at the county level is modest and fairly stable.

Rodden, J. A. (2019). Why Cities Lose: The Deep Roots of the Urban–Rural Political Divide. Basic Books.

(book) Dense places vote left, sparse places vote right, and the gradient has steepened for a century across democracies. Structural (economic-geography) story rather than a choice story, but complementary: population density itself predicts ideology.

Brown, J. R., & Enos, R. D. (2021). The measurement of partisan sorting for 180 million voters. Nature Human Behaviour, 5(8), 998–1008.

https://doi.org/10.1038/s41562-021-01066-z (paywalled; preprint via Harvard DASH)

Individual-level geolocation of ~180M registered voters: the typical voter’s nearest neighbours are strongly co-partisan; partisan isolation is high across densities, not just in cities vs. countryside. Revives the sorting thesis with far better data than Bishop had.

Motyl, M., Iyer, R., Oishi, S., Trawalter, S., & Nosek, B. A. (2014). How ideological migration geographically segregates groups. Journal of Experimental Social Psychology, 51, 1–14.

https://doi.org/10.1016/j.jesp.2013.10.010 (paywalled)

People who feel a values-mismatch with their local community are more likely to move, and they move toward more ideologically congruent places — a behavioural micro-foundation for the sort. The larger and more varied the set of destinations, the easier this is.

Scipioni, M., & Tintori, G. (2021). A rural-urban divide in Europe? An analysis of political attitudes and behaviour. JRC Technical Report EUR 30587 EN, European Commission.

→ papers/scipioni-tintori-2021-rural-urban-divide-europe-JRC.pdf

European Social Survey, 2002–2018: a robust urban–rural gradient in attitudes to immigration, the EU, and trust in institutions, widening over time — the divide is a continuous function of settlement size/density, not a binary. Directly usable evidence that the pattern is not US-specific.

Related (paywalled): Kenny, M., & Luca, D. (2021), Cambridge Journal of Regions, Economy and Society 14(3), 565–582; and “Drifting further apart?” (Political Geography, 2024).

Cramer, K. J. (2016). The Politics of Resentment. University of Chicago Press.

(book) Ethnographic account of rural Wisconsin identity as defined against the metropolis — the felt experience of the density divide.

6. Online communities: size and toxicity

Waller, I., & Anderson, A. (2021). Quantifying social organization and political polarization in online platforms. Nature, 600, 264–268.

→ papers/waller-anderson-2021-nature-published-version.pdf (published) · papers/waller-anderson-2021-polarization-online-platforms.pdf (arXiv 2010.00590 preprint)

Neural embeddings of 5.1B Reddit comments across 10,000 communities, 2005–2018. Reddit polarized sharply around 2016 — but driven by newly arriving users sorting into partisan communities, not by existing users becoming more extreme. Strong support for the compositional sorting reading of the essay: scale enables sorting; sorting, not conversion, produces the aggregate shift.

Further (not downloaded):

• Cinus et al. / Nithyanand et al. and the Reddit-toxicity-vs-size literature — several studies find larger subreddits sustain more toxic and more homogeneous discussion; worth a targeted search if the essay leans on online examples.
• Reuters Institute (2018), Echo chambers, filter bubbles and polarisation: a literature review — https://reutersinstitute.politics.ox.ac.uk/echo-chambers-filter-bubbles-and-polarisation-literature-review — sober summary; the echo-chamber effect is real but smaller and rarer than popular accounts claim.

7. How each strand maps onto the essay + influence.jl

Model note. influence.jl already contains the crucial contrast: use_closest_neighbors = true (homophilous/similarity-biased influence → persistent clusters) vs. false (random mixing → consensus). That is precisely the Flache et al. (2017) distinction. The natural extension for the essay’s argument is to vary num_samples while holding num_closest_neighbors and the θ-kernel width fixed, and show the number of surviving clusters rising with population size — the Kou et al. (2012) result.

8. Gaps / suggested next searches

• More “group size → less compromise / norm-enforcement” experiments beyond Centola et al. 2018 (§4) — especially work that varies absolute group size directly.
• Minority group size and assimilation: Borjas on ethnic enclaves (larger co-ethnic community → slower linguistic/economic assimilation) — the clearest non-ideological analogue of the thesis; only summaries gathered so far.
• LGBTQ / subculture “critical mass” literature — coming-out rates and identity consolidation as a function of local community size.
• Firm / organization size and the formation of internal factions or “cliques.”
• Whether any study operationalises the essay’s specific >1,000-person threshold (most campus work contrasts ~500 vs ~25,000; nothing found at fine resolution).