Pith. sign in

REVIEW 4 major objections 6 minor 23 references

Mapping urban segregation through co-residence network reconstruction

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A network analysis of Vienna's residence registers finds that migrant nationalities sort into two clusters split by income and diversity.

desk verdict A useful, transparent empirical application of a known null-model method to Vienna district data; the income result is solid, but the two-cluster claim is weakened by district-level resolution and the missing multiple-testing threshold. read the letter →

arxiv 2501.15920 v3 pith:E5LLAXFC submitted 2025-01-27 physics.soc-ph cs.SIphysics.data-an

Marc Sadurní , Samuel Martin-Gutierrez , Ola Ali , Ana María Jaramillo , Rafael Prieto-Curiel , Fariba Karimi This is my paper · ORCID
classification physics.soc-phcs.SIphysics.data-an
keywords urbansegregationco-residencenetworkcommunitydetectionViennamigrantintegrationhomophilySimpsonindexstatisticalvalidation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that Vienna's residential landscape is organized into two large clusters of nationalities with opposite co-residence patterns: a majority cluster (Austria, Germany, Ukraine, Russia, Hungary, and others) that lives in wealthier, less diverse districts, and a minority cluster (Serbia, Turkey, Syria, Romania, Poland, and others) that lives in poorer, more diverse districts. The claim is established by building a statistically validated network in which nodes are nationalities and links measure whether pairs of groups share districts more or less often than expected under a null model that preserves district and population sizes, then applying community detection to that network. The paper also shows that district income, district diversity, and nationality homophily are systematically correlated with the cluster structure. If correct, the result provides an interpretable map of urban sorting that complements classical segregation indices and identifies wealth, diversity, and homophily as joint mechanisms of migrant integration.

What carries the argument

The central object is the co-residence network. For each district d, the co-residence weight between nationalities i and j is the product of their resident counts in that district, $w^{d}_{ij} = \kappa^{d}_{i} \kappa^{d}_{j}$. These weights are compared with a multinomial null model that fixes each district's total population and each country's city-wide population, yielding per-district expected values and variances, and hence z-scores that are summed across all districts. Positive cumulative z-scores define significant co-residence links, negative ones avoidance links. Community detection is then carried out with the map equation (Infomap), which partitions the network of significant links by compressing the description of a random walk on it. Finally, the paper attaches each nationality to district-level average net income and Simpson-index diversity through Pearson correlations to show what separates the clusters.

What would settle it

Recompute the same co-residence network and community detection using the same nationality counts placed on a much finer spatial grid (for example individual registered addresses or 250-meter cells) and check whether the same two clusters emerge. If the clusters dissolve or change substantially, the district-level definition of co-residence is the cause; if they persist, the district resolution is sufficient to capture the segregation structure. A second check: randomly permute the 23 district boundaries (or use alternative administrative partitions) and see whether the two-cluster partition survives the perturbation.

Watch

Extended reading notes

Core claim

On the basis of a September 2023 snapshot of all registered foreign citizens in Vienna and official Austrian population counts, the paper reconstructs a statistically validated 'co-residence network' of nationalities and applies Infomap community detection to it. The network splits into two major clusters: a larger one containing Austria, Germany, Ukraine, Russia, Hungary, Iran, China, Italy, Slovenia, and the 'Others' category, and a smaller one containing Serbia, Turkey, Syria, Romania, Poland, Croatia, Bosnia and Herzegovina, North Macedonia, Bulgaria, Afghanistan, and Slovakia. The paper finds that the two clusters are cleanly separated by district-level wealth (the majority cluster concentrates in districts with higher average net income, the minority cluster in poorer districts) and by district-level diversity (the minority cluster lives in districts with higher Simpson-index diversity). It also reports that nationalities in the majority cluster tend to be geographically and culturally closer to each other, that the minority cluster contains some distant exceptions, and that national homophily, measured by a dissimilarity index, is highest for Turkey, Germany, and Italy.

Load-bearing premise

The analysis defines 'co-residence' as living in the same one of Vienna's 23 administrative districts, so any segregation that happens within a district (between blocks, streets, or buildings) is invisible to the network, and the two clusters could in principle be an artifact of how the city happens to be divided into districts.

Editorial extensions

If this is right

  • The two-cluster structure gives policymakers a concrete map for integration: mixing the minority cluster's districts with the majority cluster's districts would require crossing both an income divide and a diversity divide.
  • Stable co-residence networks can be built for any city with registration or census data at sub-city geographies, turning many pairwise segregation indices into a single interpretable map of sorting.
  • Because income and diversity correlations are large in opposite directions, policies that change a district's income mix or housing affordability are also likely to change its nationality diversity, and vice versa.
  • The dissimilarity-index ranking identifies which national communities are most spatially concentrated and therefore most likely to serve as anchor diasporas for later arrivals.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Inference: If the modifiable areal unit problem operates here, the two clusters are partly a product of Vienna's specific district boundaries; testing at finer scales or with synthetic districts is the natural next experiment.
  • Inference: The method transfers to a comparative urban typology: other European cities with similar registration data might yield three, four, or no clusters, allowing segregation structures to be compared across cities and over time.
  • Inference: The strong negative correlation between district migrant share and district income (r = -0.80) hints that the diversity-income feedback loop the paper describes may be generic rather than Vienna-specific; a multi-city replication would test that.
  • Inference: Vienna's large social-housing stock is not examined in this paper, but the framework makes a concrete question answerable: whether public-housing units are distributed across clusters, since their location would directly shape co-residence patterns.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper introduces a network approach to residential co-segregation in Vienna. Using a September 2023 snapshot of registered residents aggregated to the city's 23 administrative districts, it represents each of 21 groups (19 top migrant nationalities, 'Others', and Austrians) as a node and defines a weighted co-residence link between groups i and j from the product of their district populations (Eq. 3). Link weights are normalized against a multinomial null model that preserves district and nationality totals, yielding district-wise z-scores (Eqs. 5-6) summed over districts (Eq. 7). The 80 positive links form the network on which the map equation/Infomap detects two clusters: a 'majority' cluster (Austria, Germany, Ukraine, Russia, Hungary, etc.) and a 'minority' cluster (Serbia, Turkey, Syria, Romania, Poland, etc.). The paper then correlates each nationality's district population fraction with district average income (Fig. 3a) and district Simpson diversity (Fig. 3b), and computes a Dissimilarity homophily index (Eq. 10), arguing that wealth, diversity, and homophily jointly shape the two clusters.

Significance. The paper has genuine strengths: the code and aggregate data are public; the null model has closed-form moments (Eqs. 4-6); and the empirical claim that the two clusters differ in district income survives multiple independent operationalizations (income, rental prices, quartile decompositions, and a K-means socio-economic-status clustering, Supplementary Notes 3.2). If the load-bearing issues below are resolved, the paper would offer an interpretable, reproducible complement to classical segregation indices and a useful empirical map of Vienna's district-level sorting of nationalities. However, the headline claim that the two clusters are 'shaped by wealth disparities, district diversity, and nationality-based homophily' currently outruns the evidence: the diversity result is partly mechanical, the homophily result is descriptive rather than tested, and the cluster composition is sensitive to the significance threshold that the main analysis never actually applies.

major comments (4)
  1. [Results, 'Extracting significant co-living links'; Supplementary Note 2.2.3] The main analysis does not apply any statistical threshold: the text states that 'among the 21 groups... we identified 210 significant links, forming a fully connected' network, yet 210 = C(21,2) is the number of all possible pairs, and the actual filter for clustering is simply the sign of the cumulative z-score (80 positive links). Calling all pairs 'significant' misstates the method. This matters because the cluster partition is sensitive to the threshold: under the Bonferroni threshold t = 2.82√D introduced in Supplementary Note 2.2.3, four countries (China, Slovakia, Afghanistan, Poland) become disconnected and are no longer assigned to any cluster (Fig. S8), so the statement that the correction 'does not lead to any qualitative changes' is undercut for cluster membership. I recommend reporting the thresholded network as the primary object, and adding a sweep over edge-inclusion thresholds (e.g., z_ij exceeding 0, 5, 10, 13.5) with a consensus or stability measure for the Infomap partitions.
  2. [Methods, Eq. (8); Fig. 3b; Supplementary Table S4] The diversity determinant is partially circular. The Simpson index S_d = 1 − Σ_i (P^d_i)^2 (Eq. 8) is computed from the same nationality-district fractions that define the network and clusters, and for a given nationality k the term −(P^d_k)^2 enters S_d directly, so districts where k is over-represented are automatically less diverse (∂S_d/∂P^d_k = −2P^d_k < 0). For the dominant group this dependence is nearly exact: Austria's reported Diversity-Population correlation of r = −0.999 (Supplementary Table S4) is essentially a restatement of the index definition, and the abstract's claim that the clusters are shaped by 'district diversity' therefore rests partly on the mechanical component of the index. The rebuttal in Methods (the hypothetical district with a single migrant nationality) does not address this per-nationality dependence, since the focal nationality's own fraction is the quantity being correlated. I recommend recomputing Fig. 3b with a leave-one-out diversity index that excludes each focal nationality, and confirming that the cluster separation in Fig. 3b survives.
  3. [Abstract; Methods (Eqs. 3-7); Fig. 1] The in-paper abstract states that co-residence preferences are analyzed 'at the neighbourhood level', but every quantity in the analysis is defined on Vienna's 23 administrative districts: co-residence means sharing a district (Eqs. 3-7), and the income, diversity, and homophily measures are all district-level aggregates. This is a correctness-risk concern rather than an internal inconsistency, because a different partition of the city could produce different z-scores, a different network, and different clusters (the modifiable areal unit problem); the robustness checks in the supplement are all computed on the same 23-district aggregation and therefore cannot detect such an effect. A concrete test would be to repeat the pipeline on coarser groupings of the publicly available district data (e.g., merging districts into larger areas) and check whether the two-cluster structure and its income separation persist. The claims in the abstract and Discussion should be scoped to district-level co-residence patterns, and the manuscript's two abstracts should be harmonized (the arXiv abstract correctly says 'district level').
  4. [Abstract; Results, 'Cultural and national homophily'; Methods, Eq. (10)] The abstract lists 'nationality-based homophily' as one of the three factors shaping the two clusters, but no analysis connects the Dissimilarity index D_i (Eq. 10) to cluster membership. The text asserts that micro-level homophily 'can compound with the clustering effects observed across larger cultural groups', which is speculation, not evidence. Since D_i is computed from the same district-level data, a minimal supporting test is available: compare the D_i values of the majority-cluster and minority-cluster nationalities (e.g., a two-sample test or a regression of cluster assignment on D_i, the income correlation, and the diversity correlation jointly). Without such a test, homophily should be described as a measured descriptive statistic rather than a demonstrated determinant of the clusters.
minor comments (6)
  1. [Title] The arXiv title ('Mapping urban segregation through co-residence network reconstruction') differs from the in-paper title ('Quantifying urban socio-economic segregation through co-residence network reconstruction'); the two should be made consistent.
  2. [Abstract] The two versions of the abstract disagree on the spatial resolution of the analysis (the arXiv abstract says 'at the district level', the in-paper abstract says 'at the neighbourhood level'), and the latter overstates what the district-level analysis can support.
  3. [Fig. 2b caption] The caption states that edges with |z_ij| ≤ 20 are not shown 'for better readability', but this cutoff is arbitrary, is not mentioned in the main text, and is far above the Bonferroni threshold of t ≈ 13.5 used in the supplement; the choice should be justified or the figure should use a single, defined threshold.
  4. [Methods, Eq. (7)] The cumulative z-score z_ij is presented as a link weight, but as a sum of 23 district-wise z-scores it is not itself a standard normal statistic (the supplement's Bonferroni section correctly scales by √D); the main text should state this scaling explicitly when Eq. (7) is introduced.
  5. [Methods, 'Mapping co-residence network'] The null model is described ambiguously: the phrase 'each resident comes from a country randomly picked from the total population of Vienna' could be read as unconstrained sampling, whereas the implementation draws district populations from a multinomial distribution that preserves district totals; one sentence clarifying that both district totals and nationality totals are held fixed would remove the ambiguity.
  6. [References and typos] Minor presentation issues: the Bourdieu reference contains a formatting artifact ('F orms of Capital'), and the City of Vienna income-dataset URL contains a garbled fragment ('viewirtschaft'); both should be cleaned up before publication.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the co-residence network and clusters are derived from the raw nationality-district matrix with a fully specified null model, and the cited prior method is restated in the paper rather than being load-bearing self-citation.

full rationale

The paper's central derivation chain is self-contained and not circular. Co-residence weights are defined directly from the nationality-district counts (Eq. 3), z-scores are computed against a multinomial null model whose mean and variance are explicitly stated (Eqs. 4-6), and clusters are obtained by applying Infomap to the resulting cumulative z-score network (Eq. 7). No parameter is fitted to the cluster labels, and no outcome is predicted from a fitted input. The citation to Karimi et al. (2015) for the null-model formulas is a self-citation (one co-author), but it is not load-bearing because the paper reproduces the analytical expressions and the analysis is reproducible from public aggregate data and code. The homophily and diversity indices (Eqs. 8-10) are computed from the same nationality-district data as the network, but they are used as post-hoc correlates, not as inputs to the clustering. The diversity-population fraction correlation in Fig. 3b has a compositional component—since the Simpson index includes each nationality's own share—but this is a validity caveat rather than a definitional identity equating the cluster output to the diversity input. The paper even partially acknowledges this by discussing the feedback between diversity and clustering. No uniqueness theorem is imported, no ansatz is smuggled via citation, and no known result is merely renamed. The mismatch between 'neighbourhood-level' language and district-level aggregation is a spatial-resolution concern, not a circularity. Overall, the derivation does not reduce to its own inputs by construction.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The analysis is largely non-parametric: no regression coefficients are fitted, and the null model preserves margins. The load-bearing choices are the selection of nationalities, the spatial resolution (districts), and the interpretive leap from co-residence to preference. The diversity correlation contains a mechanical component because the explanatory variable is a function of the same nationality shares that define the outcome.

free parameters (2)
  • Top-19 nationality cutoff = 19 nationalities plus 'Others'
    Only the 19 most populous migrant nationalities are retained; the remainder are pooled. This hand-chosen cutoff determines the network's node set and could affect the cluster structure.
  • Minimum residents per nationality-district cell = 10
    Privacy threshold: cells with fewer than 10 residents are suppressed. This censors small groups and may bias z-scores for smaller nationalities.
assumptions (4)
  • domain assumption Above-chance co-residence in a district reflects residential preference or affinity, rather than housing constraints, historical path dependence, or administrative boundaries.
    This interpretation is stated in the Results ('common interest in co-living') and is the basis for building a network of preferences. It is not directly tested.
  • domain assumption Residents are assigned to districts independently within the multinomial null model, conditional on district size and city-wide nationality proportions.
    Used in Methods Eq. (4)-(6) to compute expected co-residence and variance. If household clustering or spatial autocorrelation exists, z-scores may be inflated and the number of positive links overestimated.
  • domain assumption District-level average net income approximates the income of every resident of that district.
    Stated in Supplementary Note 1.1.3; used for the income correlations in Fig. 3a and box plots.
  • ad hoc to paper The Simpson diversity index, computed from the same nationality shares used in the network, can be correlated with those shares to reveal an explanatory relation.
    The paper uses this correlation (Fig. 3b) as evidence that diversity separates clusters, but the mechanical dependence between a share and the diversity index built from all shares (including that share) creates a built-in association. This is a circularity issue, not an independent test.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Mapping urban segregation through co-residence network reconstruction." pith.science (2026). https://pith.science/paper/E5LLAXFC

@misc{pith2026250115920,
  author       = {Pith},
  title        = {Pith review of: Mapping urban segregation through co-residence network reconstruction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/E5LLAXFC}},
  note         = {Machine review of arXiv:2501.15920}
}
read the original abstract

Urban segregation poses a critical challenge for cities, exacerbating inequalities, social tensions, fears, and polarisation. It emerges from the interplay of socio-economic disparities, housing constraints, and residential preferences, and can disproportionately affect migrant communities. Here, we study residential segregation in Vienna using a city-wide administrative snapshot of registered residents, covering the full foreign population and Austrian nationals at the district level. We introduce a network-based approach by constructing a statistically validated co-residence network in which nodes represent nationalities and links capture whether pairs of groups live in the same districts more or less often than expected under a population-size-preserving null model. Applying community detection to this network reveals two major clusters of nationalities with distinct co-residence patterns. These clusters are systematically associated with district-level income disparities and diversity, while also reflecting the geographical proximity of countries of origin, with nationalities from nearby regions tending to share similar residential patterns within Vienna. Our results show how network methods can provide an intuitive and interpretable map of urban residential sorting, complementing traditional segregation indices and highlighting the multiple dimensions underlying migrant integration in diverse cities.

Figures

Figures reproduced from arXiv: 2501.15920 by the authors.

Figure 1
Figure 1. Melting Pot in Vienna. a Population of migrants in the districts of Vienna on 22nd September 2023 (in %). Districts 15 and 13 have the highest and lowest percentage of migrants, respectively (see Supplementary Table S1 for a detailed list of values). b Population distributions of the top 19 most populous migrant nationalities in Vienna represented by at least ten individuals per district, the label “Others” is used … view at source ↗
Figure 2
Figure 2. Clustering countries sharing similar residence interests. a Diagram of the method used to build the network of co-residence. b z-score network. Edges with |zi j| ≤ 20 are not shown for better readability. The width and colour transparency of the links is proportional to the z-score values, while the size of the nodes reflects the total population of each country, except for Austria. The network has been produced usi… view at source ↗
Figure 3
Figure 3. Socio-economic and diversity factors underlying nationality clusters. a Correlation between district wealth and population fraction for each nationality. b Correlation between diversity and population fraction for each nationality. Nationalities are ranked by their correlation values, from highest to lowest for wealth-population fraction correlation in a, and from lowest to highest for diversity-population fraction … view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: District diversity and national homophily in Vienna. a District diversity. Districts are ranked from highest to lowest based on the estimated diversity of Vienna districts. b Vienna diversity map. c National homophily. Nationalities are ranked from highest to lowest ba…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 22 canonical work pages

  1. [1]

    Vienna’s planning history: periodizing stable phases of regulating urban development, 1820–2020

    Suitner, J. Vienna’s planning history: periodizing stable phases of regulating urban development, 1820–2020. Planning Perspectives 36, 881–902 (2021)

    work page 2021

  2. [2]

    & Hengl, I

    Kral, U., Reimer, F., Tuz, H. & Hengl, I. Building schematic of Vienna in the late 1920s. Scientific Data 8, 44 (2021)

    work page 2021

  3. [3]

    & Reeger, U

    Fassmann, H. & Reeger, U. Austria: From guest worker migration to a country of immigration. IDEA Working Papers1, 1–39 (2008)

    work page 2008

  4. [4]

    & Rupnow, D

    Bischof, G. & Rupnow, D. Migration in Austria (innsbruck university press, 2017)

    work page 2017

  5. [5]

    Vienna population 2023: data and facts on migration and integra- tion

    City of Vienna. Vienna population 2023: data and facts on migration and integra- tion. https://www.wien.gv.at/menschen/integration/daten-fakten/ bevoelkerung-migration.html (2023). Accessed: 5 Apr. 2024

    work page 2023

  6. [6]

    (see Supplementary Fig. S1 b). To estimate the Austrian population in each district (see Supplementary Fig. S1 c), we calculated the difference between the total population figures (see Supplementary Fig. S1 b) and the number of registered migrants recorded in our dataset (see Supplementary Fig. S1a). This approach maintains alignment with official statis...

    work page 2023

  7. [7]

    Population by gender since 1869 in Vienna districts.https://www.data.gv.at/ katalog/de/dataset/vie-bez-pop-sex-stk-1869f (2023)

    City of Vienna. Population by gender since 1869 in Vienna districts.https://www.data.gv.at/ katalog/de/dataset/vie-bez-pop-sex-stk-1869f (2023). Accessed: 30 Oct. 2023

    work page 2023

  8. [8]

    Median income by country

    World Population Review. Median income by country. https://worldpopulationreview. com/country-rankings/median-income-by-country (2024). Accessed: 5 Apr. 2024

    work page 2024

Show all 23 references
  1. [9]

    Online analysis tool for global poverty monitoring

    World Bank Group. Online analysis tool for global poverty monitoring. https://www. worldbank.org/en/home (2024). Accessed: 5 Apr. 2024

    2024

  2. [10]

    Mean and median income by household type

    Eurostat. Mean and median income by household type. https://appsso.eurostat.ec. europa.eu/nui/show.do?dataset=ilc_di04 (2024). Accessed: 5 Apr. 2024

    2024

  3. [11]

    Measuring global inequality: Median income, gdp per capita, and the gini index

    Giving What We Can. Measuring global inequality: Median income, gdp per capita, and the gini index. https://www.givingwhatwecan.org/blog/ measuring-global-inequality-median-income-gdp-per-capita-and-the-gini-index (2021). Accessed: 5 Apr. 2024

    2021

  4. [12]

    Worldwide, median household income about $10,000

    Gallup. Worldwide, median household income about $10,000. https://news.gallup.com/ poll/166211/worldwide-median-household-income-000.aspx (2013). Accessed: 5 Apr. 2024

    2013

  5. [13]

    US dollar to euro spot exchange rates for 2020

    Exchange Rates UK. US dollar to euro spot exchange rates for 2020. https://www. exchangerates.org.uk/USD-EUR-spot-exchange-rates-history-2020.html (2024). Accessed: 5 Apr. 2024

    2024

  6. [14]

    Rental prices by Vienna areas

    Derstandard. Rental prices by Vienna areas. https://www.virtualvienna.net/ moving-to-vienna/accommodation-in-vienna/real-estate-rental-prices/ (2023). Accessed: 30 Oct. 2023

    2023

  7. [15]

    & Pentland, A

    Moro, E., Calacci, D., Dong, X. & Pentland, A. Mobility patterns are associated with experienced income segregation in large US cities. Nature Communications 12, 4633 (2021)

    2021

  8. [16]

    Income from dependent employment by gender since

    City of Vienna. Income from dependent employment by gender since

  9. [17]

    Abdi, H. et al. Bonferroni and ˇSid´ak corrections for multiple comparisons. Encyclopedia of measure- ment and statistics 3, 2007 (2007)

    2007

  10. [18]

    Parera, M. S. ViennaMosaic. https://github.com/MarcSadurniParera/ ViennaMosaic (2024). Accessed: 24 Dec. 2024

    2024

  11. [19]

    MacQueen, J. et al. Some methods for classification and analysis of multivariate observations. In Proceedings of the fifth Berkeley symposium on mathematical statistics and probability , vol. 1, 281– 297 (Oakland, CA, USA, 1967)

    1967

  12. [20]

    & Rosvall, M

    Edler, D., Holmgren, A. & Rosvall, M. The MapEquation software package. https:// mapequation.org (2023). Accessed: 11 Nov. 2023

    2023

  13. [21]

    White, M. J. Segregation and diversity measures in population distribution. Population Index 198–221 (1986). Supp. Inf. Quantifying urban socio-economic segregation through co-residence network reconstruction

    1986

  14. [22]

    Shannon, C. E. A mathematical theory of communication. The Bell system technical journal 27, 379–423 (1948)

    1948

  15. [2002]

    Accessed: 5 Apr

    https://www.data.gv.at/katalog/de/dataset/stadt-wien_ viewirtschaftwienergemeindebezirke1 (2023). Accessed: 5 Apr. 2024. Supp. Inf. Quantifying urban socio-economic segregation through co-residence network reconstruction Marc Sadurn´ı et al. 26

    2023

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.