Pith. sign in

REVIEW 3 major objections 5 minor 29 references

Beyond Distance: Mobility Neural Embeddings Reveal Visible and Invisible Barriers in Urban Space

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Neural embeddings of 25 million trajectories show that amenity gaps, county borders, and income and racial divides shape urban movement more than highways, rivers, or parks do.

desk verdict Scales embedding-based barrier detection to 11 cities with strong robustness, but the barrier label needs external validation before the central ranking is trusted. read the letter →

arxiv 2506.24061 v1 pith:H6QUSK6S submitted 2025-06-30 cs.CY physics.soc-ph

Guangyuan Weng , Minsuk Kim , Yong-Yeol Ahn , Esteban Moro This is my paper · ORCID
classification cs.CYphysics.soc-ph
keywords UrbanmobilityBehavioralbarriersAmenityinequalityembeddingsWord2VecResidentialsegregation
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 tries to establish that the obstacles that actually structure urban movement are mostly invisible: the arrangement of amenities, county lines, and residential segregation by income and race matter more than familiar physical barriers, and physical barriers such as highways and parks matter mainly at short distances. The authors train word-embedding models on 25.4 million anonymized device trajectories across 11 U.S. metros, treating neighborhoods as words and trips as sentences, so each neighborhood receives a vector whose cosine distance reflects mobility context rather than geography. Declaring a barrier wherever nearby neighborhoods are far apart in this behavioral space relative to what raw distance predicts, they find these barriers cluster in urban cores, persist across cities and years, and are crossed mainly on discretionary trips by residents of diverse, transit-connected, employed areas. If the claim is right, urban accessibility is driven less by physical geography than by where amenities sit and where jurisdictional and social divides run, and the same neural-embedding method can monitor such exclusion at scale.

What carries the argument

The load-bearing object is the Word2Vec skip-gram negative sampling embedding trained on trajectories: each census block group is a word, each user's sequence of visited block groups is a sentence, and the learned vector $w_i$ defines an embedding distance $d^{(e)}_{ij}=1-\frac{w_i\cdot w_j}{\|w_i\|\|w_j\|}$ that reflects how often places appear in similar movement contexts. The equivalence between this model and the gravity law of mobility justifies treating this cosine distance as a behavioral distance. A barrier is then defined operationally as a block-group pair within 20 km whose embedding distance sits in the top 5% of positive residuals from the log-linear fit $\hat{d}^{(e)}_{ij}\sim\beta\log d^{(p)}_{ij}+\varepsilon$ within each 1 km distance bin; logistic regressions with standardized predictors compare the contribution of amenity, physical, demographic, and administrative variables to the odds that a pair is a barrier.

What would settle it

Re-run the barrier detection with road-network travel time added alongside geographic distance in the residual model for any one metro, for example Boston. If the positive residuals that define the top-5% barrier pairs shrink to near zero or no longer correlate with county boundaries and income and racial distance, then the 'invisible barriers' are unmeasured physical friction and the central claim fails.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that urban mobility barriers—pairs of census block groups less than 20 km apart whose embedding distance is in the top 5% of positive residuals from a log-linear fit on geographic distance—are best explained by amenity structure, administrative borders, and demographic distance. Differences in intervening opportunities, especially shopping, college, and sports venues, raise the odds of a barrier, while functional complementarity between areas lowers them; county boundaries and income and racial gaps are consistently strong predictors, and physical infrastructure is secondary and distance-dependent, with highways and parks acting as local obstacles that can facilitate movement at longer ranges. The detected barriers are concentrated in urban cores and persist across cities, spatial scales, and time periods, indicating stable behavioral structures rather than temporary artifacts.

Load-bearing premise

The whole analysis depends on the operational definition of a barrier as the top 5% of embedding residuals after a log-linear fit on geographic distance; if those residuals mostly reflect noise or unmeasured physical travel time rather than experienced disconnection, the ranking of factors that follows would be an artifact of the method.

Editorial extensions

If this is right

  • Urban accessibility is primarily a product of amenity arrangement and jurisdictional and demographic divides, so maps of physical obstacles alone will mislead planning decisions.
  • Changing where amenities sit can either reinforce or reduce barriers, since intervening opportunities such as shopping and college venues suppress direct movement while complementary venues, groceries, services, and transport hubs encourage it.
  • Physical infrastructure has a dual role: highways, parks, railways, and waterways constrain movement at short distances but can connect places at longer distances, so barrier effects cannot be summarized by a single sign.
  • The barrier structure is stable across 2019–2021 and across the 11 metros, with only a temporary pandemic shift in 2020, so the detected invisible divides reflect persistent urban structure.
  • Barrier crossing is tied to discretionary and exploratory trips, and individuals from more racially diverse, transit-using, and employed neighborhoods cross barriers more often, implying that connectivity itself is unequally distributed.

Reading between the lines

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

  • A direct testable extension the authors do not run: replace geographic distance with road-network travel time in the residual fit; if the top-5% barrier pairs largely disappear, the 'invisible' effects would be visible travel-time friction in disguise.
  • The correlational design leaves causality open in both directions—diverse transit-rich neighborhoods may produce crossers, or people who cross anyway may sort into those neighborhoods—so panel or natural-experiment data would be needed to separate the two.
  • Because all distances and flows are symmetric here, an editorial next step would be to use directional embeddings and ask whether barriers are asymmetric, for instance blocking low-amenity to high-amenity trips more than the reverse.
  • The same embedding-residual machinery could be applied to other service geographies, such as schools or hospitals, to audit whether administrative catchment boundaries create invisible access barriers in domains beyond daily mobility.
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

3 major / 5 minor

Summary. The manuscript applies Word2Vec to CBG-level mobility trajectories from 25.4 million devices in 11 U.S. CBSAs, defining a 'mobility barrier' as a CBG pair within 20 km whose embedding cosine distance lies in the top 5% of positive residuals from a log-linear regression on geographic distance, within 1-km bins. It then fits logistic regressions of this barrier label on POI intervening opportunities, POI category complementarity, counts of crossed physical infrastructure, demographic distances, and county-boundary indicators, and reports a stable ranking across cities, years, distances, and robustness checks: amenity structure is the strongest predictor, administrative borders second, demographic segregation third, and physical infrastructure weakest. A second analysis characterizes barrier-crossing trips by time of day, venue category, and residential demographics. The paper claims that invisible, amenity- and segregation-driven barriers structure urban accessibility more than visible physical obstacles.

Significance. If the central claim survives validation, the paper is significant: it offers a scalable, data-driven way to identify behavioral disconnection between nearby neighborhoods, challenges infrastructure-centric barrier studies, and provides planning-relevant evidence on amenity layout and jurisdictional divides. Strengths include the scale (11 CBSAs, three years, 25.4 million devices), the systematic robustness checks across cities, years, spatial resolutions (H3), and pruning thresholds, the explicit link between window-1 skip-gram negative sampling and the gravity model, and public code for reproducing figures from aggregated data. The main caveat is that the barrier label is an internal residual of the embedding model and is not validated against independent measures; the stress-test concern about circularity is therefore central rather than peripheral.

major comments (3)
  1. [Methods, Defining Mobility Barriers; SI Note 3] The outcome variable in Eq. (1) is constructed from the top 5% residuals of an embedding distance model trained on the same trajectories whose barriers the paper aims to explain, and SI Note 3 states that most of these pairs have zero observed flow. The R^2 validation in Fig. 1c-d demonstrates that embeddings track average flows, but it does not validate the residual tails used as labels. Because Word2Vec with window=1 and min_count=50 (SI Note 4.1) gives poorly constrained vectors for low-frequency CBGs, and low-frequency CBGs may be systematically different in income, transit use, or county membership, the demographic and administrative coefficients in Eq. (1) could be artifacts of embedding frequency rather than behavioral isolation. Please add a baseline that labels barriers from residuals of observed flow, or from an independent source such as surveys or travel diaries, and rerun the full predictor ranking; also test explicitly how the ranking changes when low-frequency CBGs are excluded.
  2. [SI Note 3; Methods, Defining Mobility Barriers] The individual-level analysis changes the definition of the outcome: because main barrier pairs are mostly zero-flow dyads, the cross-barrier ratio is computed on 'soft barriers' defined as the top quartile of residuals among non-zero-flow pairs. The paper does not show that these soft barriers correspond to the same places or the same socioeconomic patterns as the hard barriers in Eq. (1), so Fig. 4 may characterize a different phenomenon. Please report overlap statistics between the two barrier sets and repeat the temporal and demographic analyses on the hard-barrier definition where data permit.
  3. [Methods, Defining Mobility Barriers; SI Section 5] The barrier definition depends on several free parameters—the 20 km cutoff, the 1 km bin size, and the top 5% residual threshold—but the robustness section varies only the flow-pruning quantiles and the spatial unit. Since the paper's headline is a ranking of predictor groups, a sensitivity analysis that varies the residual quantile (for example, top 1% and top 10%) and the distance cutoff is necessary to show that the ranking is not an artifact of one threshold choice.
minor comments (5)
  1. [Introduction] The sentence beginning 'To address this gap by combining large-scale mobility data...' lacks a main verb; it should read 'We address this gap by combining...'.
  2. [SI Note 5.1, Fig. 14] The caption of Fig. 14(b) labels the second pruning range as [0.94,0.997], while the text specifies [0.95,0.997]; these should be reconciled.
  3. [Abstract] The abstract contains a line-break artifact 'Y et' that should read 'Yet'.
  4. [SI Note 1.4; Fig. 3c] The paper would benefit from a table defining the 20 POI categories in the main text, since the taxonomy is only described in the supplementary information and is used for the headline POI effect.
  5. [Fig. 3b] The right-hand axis of Fig. 3b shows '50%' and '100%' ticks without explanation; please state what the normalized likelihood-ratio values represent.

Circularity Check

1 steps flagged · score 4.0 of 10

The headline barrier-importance ranking largely restates Stouffer's intervening-opportunities law, but the analysis contains independent external predictors and the embedding is validated against flows.

  1. renaming known result [Results, 'Identification and Nature of Mobility Barriers' (Fig. 3c); Methods, 'Defining Mobility Barriers']
    "we identify CBG pairs where observed mobility is significantly lower than expected given their geographic distance... the number of intervening opportunities between CBGs is positively associated with barrier formation, as nearby alternative destinations reduce the likelihood of direct movement"

    The barrier label is a high positive residual of embedding cosine distance regressed on geographic distance. Because the embeddings are trained on the same origin-destination trajectories, this residual is effectively a nonlinear proxy for abnormally low inter-CBG flow. The top-ranked predictor, POI Intervening Opportunities, is exactly Stouffer's classic mechanism for low mobility, which the paper itself states in SI Note 2.1: 'the number of persons going a given distance is ...

full rationale

No direct self-definitional circularity is present: the barrier residual is not algebraically equal to any predictor, and the main regressors (county boundaries, income/race distances, physical crossings) are measured externally from trajectories. The embedding step is independently supported within the paper by the gravity-model comparison (R2 = 0.61 vs. 0.33 in Boston, Fig. 1c-d), so the self-citation to Murray et al. for the SGNS-gravity equivalence is not load-bearing. The main circularity concern is narrower: the outcome is a flow-deficit residual, and the dominant predictor, POI intervening opportunities, is the classic Stouffer explanation for flow deficits. Thus the central 'amenity structure is primary' result is partly a re-expression of a known law in embedding coordinates. Other findings, such as the secondary role of physical infrastructure and the cross-barrier temporal/demographic patterns, are not forced by this construction and give the paper genuine independent content. Overall score reflects partial, not total, circularity.

Assumptions & free parameters 5 free parameters · 4 assumptions · 1 invented entities

The central claim rests on the assumption that embedding residuals faithfully encode behavioral disconnection, plus a set of threshold choices for the outcome and preprocessing. The Word2Vec-to-gravity equivalence is imported from a self-cited paper. No invented physical entities are introduced; the main constructed entity is the 'mobility barrier' latent variable.

free parameters (5)
  • Barrier residual threshold = top 5% per 1 km bin
    Defines which CBG pairs are labeled as mobility barriers; this choice sets the outcome variable for all regressions.
  • Soft barrier threshold (cross-barrier ratio) = top 25% among nonzero-flow pairs
    Used in SI Note 3 to define cross-barrier activities; different threshold from the main barrier definition.
  • Flow pruning quantile range = [0.95, 0.995]
    Trajectories outside this flow frequency range are removed before Word2Vec training; sensitivity checks only vary the range slightly.
  • Distance cutoff and bin size = 20 km, 1 km bins
    Defines the spatial scope and stratification for barrier detection and regression.
  • POI attribution distance = 100 m
    Threshold for assigning a stay to a venue; affects POI category composition for intervening opportunity and activity analysis.
assumptions (4)
  • domain assumption Word2Vec skip-gram with negative sampling and window size 1 is formally equivalent to the mobility gravity model.
    Invoked in Results and Methods 4.1 via Murray et al. (2023); it justifies interpreting cosine distances between learned vectors as behavioral mobility distances. The equivalence is a cited derivation by a coauthor group, not re-derived here.
  • ad hoc to paper Large positive residuals from a log-linear fit of embedding distance on geographic distance denote real mobility barriers.
    This is the central operationalization (Methods, Defining Mobility Barriers). The paper assumes the residual is not dominated by noise or nonlinearity in the distance-decay curve, and provides no independent validation.
  • domain assumption The Spectus opt-in smartphone panel adequately represents urban population mobility.
    SI 1.2 reports Spearman correlation of about 0.64 between device density and Census population, indicating imperfect representativeness; the paper proceeds assuming this is sufficient.
  • domain assumption CBG-level aggregation preserves barrier-relevant movement structure.
    Trajectories are aggregated to Census Block Groups; robustness is checked with H3 tiles, but CBGs vary in size and may themselves reflect administrative divisions.
invented entities (1)
  • Mobility barrier
    purpose: A latent construct labeling CBG pairs that are geographically close but behaviorally disconnected, used as the outcome in all regression analyses.
    The barrier is defined entirely by embedding residuals from the same data used to fit the model. No survey, travel diary, or social network measure confirms that these pairs are experienced as barriers by residents.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Beyond Distance: Mobility Neural Embeddings Reveal Visible and Invisible Barriers in Urban Space." pith.science (2026). https://pith.science/paper/H6QUSK6S

@misc{pith2026250624061,
  author       = {Pith},
  title        = {Pith review of: Beyond Distance: Mobility Neural Embeddings Reveal Visible and Invisible Barriers in Urban Space},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/H6QUSK6S}},
  note         = {Machine review of arXiv:2506.24061}
}
read the original abstract

Human mobility in cities is shaped not only by visible structures such as highways, rivers, and parks but also by invisible barriers rooted in socioeconomic segregation, uneven access to amenities, and administrative divisions. Yet identifying and quantifying these barriers at scale and their relative importance on people's movements remains a major challenge. Neural embedding models, originally developed for language, offer a powerful way to capture the complexity of human mobility from large-scale data. Here, we apply this approach to 25.4 million observed trajectories across 11 major U.S. cities, learning mobility embeddings that reveal how people move through urban space. These mobility embeddings define a functional distance between places, one that reflects behavioral rather than physical proximity, and allow us to detect barriers between neighborhoods that are geographically close but behaviorally disconnected. We find that the strongest predictors of these barriers are differences in access to amenities, administrative borders, and residential segregation by income and race. These invisible borders are concentrated in urban cores and persist across cities, spatial scales, and time periods. Physical infrastructure, such as highways and parks, plays a secondary but still significant role, especially at short distances. We also find that individuals who cross barriers tend to do so outside of traditional commuting hours and are more likely to live in areas with greater racial diversity, and higher transit use or income. Together, these findings reveal how spatial, social, and behavioral forces structure urban accessibility and provide a scalable framework to detect and monitor barriers in cities, with applications in planning, policy evaluation, and equity analysis.

Figures

Figures reproduced from arXiv: 2506.24061 by the authors.

Figure 1
Figure 1. Understanding mobility borders through human movement patterns and spatial embeddings. (a) Example trajectory in Downtown Boston showing how physical obstacles, income differences, and POI distribution influence destination choice. At time t5, the individual selects a restaurant, bypassing closer options at t (1) 5 and t (2) 5 probably due to income differences and a large detour caused by the river. Maps created in… view at source ↗
Figure 2
Figure 2. Unsupervised embeddings uncover hidden mobility barriers and multi-scale urban structure. (a) Left: eographic map of the Boston metro area CBGs colored by county. Maps created in Python using TIGER shapefiles10. Center: UMAP projection of the embedding space, where each point is a CBG and colors indicate county membership. The projection shows different hierarchies of mobility at diverse spatial scales. The separati… view at source ↗
Figure 3
Figure 3. Visible and Invisible Factors Shape Urban Mobility Barriers (a) Sample of detected mobility barriers in Boston. Lines connect CBG pairs with barriers; CBGs involved in over 100 barriers are highlighted. Some align with visible infrastructure, while others occur between adjacent areas without physical separation. (b) Normalized likelihood ratio test showing the explanatory power of four factor groups across distances… view at source ↗
Figures from the paper (16 more)
Figure 4
Figure 4. Figure 4: Characteristics of cross-barrier mobility patterns. (a) Conceptual framework for classifying activities as cross-barrier (red) or within-barrier (blue) movements. The cross-barrier ratio measures the share of barrier-crossing trips per individual. (b) Temporal distribu…
Figure 5
Figure 5. Figure 5: Distribution of raw flows for each CBSA in 2019. Flow values are displayed on a logarithmic scale along the x-axis. 0.000 0.005 0.010 0.015 0.020 0.025 Frequency Boston Seattle Detroit Dallas Los Angeles New York 0 500 1000 1500 Flow 0.000 0.005 0.010 0.015 0.020 0.025…
Figure 6
Figure 6. Figure 6: Distribution of pruned flows for each CBSA in 2019. 15/40 [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Correlation between the smartphone population detected in our data and Census population. The Spearman correlation between them is shown at the top. 1.2 Representativeness of the Locational data Our location data is derived from smartphone devices across major U.S. met…
Figure 8
Figure 8. Figure 8: (a) Distribution of points of interest from 11 CBSAs across categories in the venue dataset. (b) Distribution of spatial attribution distances across venue categories for 2019. Error bars denote standard errors of attribution distances. 1.4.1 Spatial Attribution of Sta…
Figure 9
Figure 9. Figure 9: Illustration of POI intervening opportunity quantification for a given CBG pair. The convex hull encompasses all POIs that could serve as intervening opportunities between the two CBGs. Map was produced in kepler.gl using the TIGER shapefiles from the U.S. Census Burea…
Figure 10
Figure 10. Figure 10: Illustration of physical barrier quantification for a given CBG pair. The straight line connecting CBG centroids intersects with one river, one highway, and four parks. Map is produced in kepler.gl using the TIGER shapefiles from the U.S. Census Bureau10 . 2.3 Measuri…
Figure 11
Figure 11. Figure 11: Distribution of distance measures used to characterize barriers across all 11 CBSAs. 3 Cross Barrier Ratio Fig. 4a summarizes the workflow used to compute the cross-barrier ratio (CBR). Because the origin–destination matrix is highly sparse—only 3% of CBG pairs regist…
Figure 12
Figure 12. Figure 12: Distribution of the user-level cross-barrier ratio (CBR) across the 11 CBSAs analysed. The pronounced spike at zero indicates that the majority of users do not cross a detected barrier during the observation period. exhibit non-zero flow and (ii) recompute residuals v…
Figure 13
Figure 13. Figure 13: Pairwise correlation matrices for all predictors in six representative distance groups (11 CBSA, 2019). With the exception of a handful of correlations above 0.30 at the shortest ranges, interdependencies are weak, indicating negligible multicollinearity. 22/40 [PITH…
Figure 14
Figure 14. Figure 14: Heatmap of logistic regression coefficients for barrier predictors across distances for trajectory pruning with thresholds (a) [0.94,0.995] and (b) [0.94,0.997]. Results are presented across 20 distance bins in 11 CBSAs, including POI features, physical infrastructure…
Figure 15
Figure 15. Figure 15: Heatmap of logistic regression coefficients for barrier predictors across distances in H3 (resolution 8) tiles. Results are presented across 20 distance bins in 11 CBSAs, including POI features, physical infrastructure, demographic differences, and county boundaries. …
Figure 16
Figure 16. Figure 16: (a), (b) and (c) Heatmap of logistic regression coefficients for barrier predictors across distances for years 2019 to 2021. Results are presented across 20 distance bins in 11 CBSAs, including POI features, physical infrastructure, demographic differences, and county…
Figure 17
Figure 17. Figure 17: Explanatory power of distance metrics in the normalized gravity model. For each CBSA–year pair, the darker bar shows the R 2 obtained with geographical distance, whereas the lighter bar shows the R 2 obtained with embedding cosine distance. Positive gaps indicate that…
Figure 18
Figure 18. Figure 18: Heatmap of logistic regression coefficients for barrier predictors in each CBSA. Results are presented across 20 distance bins for each CBSAs, including POI features, physical infrastructure, demographic differences, and county boundaries. Empty symbols correspond to …
Figure 19
Figure 19. Figure 19: Estimated coefficients of model (4) for different CBSA in 2019 study period. Error bars denote standard errors of coefficient estimates. 31/40 [PITH_FULL_IMAGE:figures/full_fig_p031_19.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 25 canonical work pages

  1. [1]

    Mobility patterns are associated with experienced income segregation in large us cities

    Esteban Moro, Dan Calacci, Xiaowen Dong, and Alex Pentland. Mobility patterns are associated with experienced income segregation in large us cities. Nature communications, 12(1):4633, 2021

    work page 2021

  2. [2]

    Distributed representations of words and phrases and their compositionality

    Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. Distributed representations of words and phrases and their compositionality. Advances in neural information processing systems, 26, 2013

    work page 2013

  3. [3]

    Efficient estimation of word representations in vector space

    Tomas Mikolov. Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781, 2013

    arXiv 2013

  4. [4]

    Constraining weighted word co-occurrence frequencies in word embeddings

    Paula Lauren. Constraining weighted word co-occurrence frequencies in word embeddings. In 2021 IEEE International Conference on Big Data (Big Data), pages 5193–5198. IEEE, 2021

    work page 2021

  5. [5]

    American Community Survey 5-Year Data

    United States Census Bureau. American Community Survey 5-Year Data. https://www.census.gov/ programs-surveys/acs, 2022. Accessed: 2024-08-22

    work page 2022

  6. [6]

    https://docs

    Foursquare open source places: A new foundational dataset for the geospatial community. https://docs. foursquare.com/data-products/docs/fsq-os-places-release-notes , 2024. Accessed: 2024-11- 01

    work page 2024

  7. [7]

    Inferring human mobility from sparse low accuracy mobile sensing data

    Andrea Cuttone, Sune Lehmann, and Jakob Eg Larsen. Inferring human mobility from sparse low accuracy mobile sensing data. In Proceedings of the 2014 ACM International Joint Conference on Pervasive and Ubiquitous Computing: Adjunct Publication, pages 995–1004, 2014

    work page 2014

  8. [8]

    Planet dump retrieved from https://planet.osm.org

    OpenStreetMap contributors. Planet dump retrieved from https://planet.osm.org . https://www.openstreetmap. org, 2017

    work page 2017

Show all 29 references
  1. [9]

    Modeling and analyzing urban networks and amenities with osmnx

    Geoff Boeing. Modeling and analyzing urban networks and amenities with osmnx. Geographical Analysis, 2025

    2025

  2. [10]

    Census Bureau

    U.S. Census Bureau. TIGER Data Products Guide. https://www.census.gov/programs-surveys/ geography/guidance/tiger-data-products-guide.html , 2019. Accessed: 2024-8-22

    2019

  3. [11]

    Intervening opportunities: a theory relating mobility and distance

    Samuel A Stouffer. Intervening opportunities: a theory relating mobility and distance. American sociological review, 5(6):845–867, 1940

    1940

  4. [12]

    Quantifying barriers of urban mobility

    Gerg˝o Pintér and Balázs Lengyel. Quantifying barriers of urban mobility. arXiv preprint arXiv:2312.11343, 2023

    2023 arXiv

  5. [13]

    Urban highways are barriers to social ties

    Luca Maria Aiello, Anastassia Vybornova, Sándor Juhász, Michael Szell, and Eszter Bokányi. Urban highways are barriers to social ties. Proceedings of the National Academy of Sciences, 122(10):e2408937122, 2025

    2025

  6. [14]

    The death and life of great American cities

    Jane Jacobs. The death and life of great American cities. Modern Library, New York, 1993

    1993

  7. [15]

    Software Framework for Topic Modelling with Large Corpora

    Radim ˇReh˚ uˇrek and Petr Sojka. Software Framework for Topic Modelling with Large Corpora. In Proceedings of the LREC 2010 Workshop on New Challenges for NLP Frameworks, pages 45–50, Valletta, Malta, May 2010. ELRA. http://is.muni.cz/publication/884893/en

    2010

  8. [16]

    Unsupervised embedding of trajectories captures the latent structure of scientific migration

    Dakota Murray, Jisung Yoon, Sadamori Kojaku, Rodrigo Costas, Woo-Sung Jung, Staša Milojevi´c, and Yong-Yeol Ahn. Unsupervised embedding of trajectories captures the latent structure of scientific migration. Proceedings of the National Academy of Sciences, 120(52):e2305414120, 2023

    2023

  9. [17]

    Principled approach to the selection of the embedding dimension of networks

    Weiwei Gu, Aditya Tandon, Yong-Yeol Ahn, and Filippo Radicchi. Principled approach to the selection of the embedding dimension of networks. Nature Communications, 12(1):3772, 2021

    2021

  10. [18]

    H3: Uber’s Hexagonal Hierarchical Spatial Index

    Uber Technologies, Inc. H3: Uber’s Hexagonal Hierarchical Spatial Index. Uber Engineering Blog, June 2018

    2018

  11. [19]

    UMAP: Uniform Manifold Approximation and Projection

    Leland McInnes, John Healy, Nathaniel Saul, and Lukas Großberger. UMAP: Uniform Manifold Approximation and Projection. Journal of Open Source Software, 3(29):861, 2018

    2018

  12. [20]

    Data Structures for Statistical Computing in Python

    Wes McKinney. Data Structures for Statistical Computing in Python. In Stéfan van der Walt and Jarrod Millman, editors, Proceedings of the 9th Python in Science Conference, pages 56 – 61, 2010

    2010

  13. [21]

    Geopandas: Python tools for geographic data, 2014

    K Jordahl. Geopandas: Python tools for geographic data, 2014

    2014

  14. [22]

    statsmodels: Econometric and statistical modeling with python

    Skipper Seabold and Josef Perktold. statsmodels: Econometric and statistical modeling with python. In 9th Python in Science Conference, 2010

    2010

  15. [23]

    Oliphant, Matt Haberland, Tyler Reddy, David Cournapeau, Evgeni Burovski, Pearu Peterson, Warren Weckesser, Jonathan Bright, Stéfan J

    Pauli Virtanen, Ralf Gommers, Travis E. Oliphant, Matt Haberland, Tyler Reddy, David Cournapeau, Evgeni Burovski, Pearu Peterson, Warren Weckesser, Jonathan Bright, Stéfan J. van der Walt, Matthew Brett, Joshua Wilson, K. Jarrod Millman, Nikolay Mayorov, Andrew R. J. Nelson, E...

    2020

  16. [24]

    John D. Hunter. Matplotlib: A 2D Graphics Environment. Computing in Science & Engineering, 9(3):90–95, 2007

    2007

  17. [25]

    Michael L. Waskom. seaborn: statistical data visualization. Journal of Open Source Software, 6(60):3021, 2021

    2021

  18. [26]

    ggplot2: Elegant Graphics for Data Analysis

    Hadley Wickham. ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag New York, 2016

    2016

  19. [27]

    Efficient estimation of maximum likelihood models with multiple fixed-effects: the R package FENmlm

    Laurent Bergé. Efficient estimation of maximum likelihood models with multiple fixed-effects: the R package FENmlm. Technical Report 2013-13, CREA Discussion Papers, 2018

    2013

  20. [28]

    tidycensus: Load US Census Boundary and Attribute Data as ’tidyverse’ and ’sf’-Ready Data Frames, 2025

    Kyle Walker and Matt Herman. tidycensus: Load US Census Boundary and Attribute Data as ’tidyverse’ and ’sf’-Ready Data Frames, 2025. R package version 1.7.1

    2025

  21. [29]

    tigris: Load Census TIGER/Line Shapefiles, 2025

    Kyle Walker. tigris: Load Census TIGER/Line Shapefiles, 2025. R package version 2.2.1. 33/40 Table 6. Estimated coefficients of model given by Eq. (1) for the 2019 study period with respective standard errors in parentheses. P-values correspond to two-sided tests for the hypot...

    2025

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.