REVIEW 3 major objections 4 minor 18 references
Who Leads in the Shadows? ERGM and Centrality Analysis of Congressional Democrats on Bluesky
T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash
Pith's one-line read A backbench House Democrat, Marcy Kaptur, holds the most structurally central position in her party's Bluesky follows network, outranking formal leaders.
desk verdict Plausible and timely application of ERGM/centrality to Bluesky, but the missing tables and an inconsistent node count (219 vs. 182) currently make the central influence claims unverifiable. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The analysis is carried by three objects: (1) three directed networks among 182 verified Democratic members of Congress — follows, mentions, and reposts — built from Bluesky data collected via the AT Protocol; (2) standard social-network centrality measures (degree, closeness, betweenness, eigenvector) used to locate structurally influential actors; and (3) exponential random graph models (ERGMs), which estimate how endogenous effects like reciprocity and exogenous attributes like chamber, state, caucus membership, and leadership status shape tie formation. Topic analysis with BERTopic, a transformer-based model, supplements the network view by comparing the top eight topics per faction using cosine similarity. The machinery turns raw follows and repost counts into a ranking of who is best positioned to broker or diffuse information, and tests whether those positions are explained by formal attributes.
What would settle it
Recompute the follows-network centrality after adding all currently missing Democratic members, or after repeatedly deleting a random 10% of the 182 nodes; if Marcy Kaptur's top closeness and betweenness ranks do not survive the perturbation, or if the ERGM homophily effects change sign, the paper's central claim about latent structural influence would be undercut.
Extended reading notes
Core claim
The central discovery is that centrality in the Bluesky network of congressional Democrats identifies influential figures who are overlooked by mainstream political coverage. Concretely, Marcy Kaptur has the highest closeness centrality (0.9495) and betweenness centrality (0.0786) in the directed follows network despite having no formal caucus or leadership affiliation, while Dwight Evans and Donald Beyer emerge as central brokers in the sparser repost network. The authors interpret these positions as evidence that structural centrality on a decentralized platform can reveal latent influence within the party, complementing the visibility of formal leaders like Hakeem Jeffries and Elizabeth Warren. The accompanying ERGM shows significant homophily effects for chamber, state, verification status, and most caucus memberships, with Senate leadership negatively associated with tie formation, and the topic analysis finds both shared high-salience themes and faction-specific divergence, with The Squad least aligned to other groups.
Load-bearing premise
The load-bearing premise is that the 182 members with verified Bluesky accounts are representative of all congressional Democrats: if the omitted members (the paper reports 16.4%, though its own counts of 258 total and 182 mapped imply a larger share) occupy systematically different network positions, the centrality rankings and ERGM estimates could change, and the promised sensitivity analysis for random node removal is not reported.
Editorial extensions
If this is right
- If the centrality results hold, then no single visibility ranking — followers, media mentions, or formal posts — captures who is structurally influential inside the Democratic caucus on Bluesky.
- Because ERGM finds strong homophily on chamber and state, the follows network is largely organized by institution and geography, not just by ideology.
- Senate leadership's negative association with tie formation suggests a broadcast-oriented, less reciprocal communication style among top senators.
- The Squad's low topical overlap with other factions implies that discourse divergence and structural integration can coexist: a group can be well-connected yet semantically distinct.
Reading between the lines
- A testable extension is to treat the ERGM homophily coefficients as baseline expectations and check whether the same patterns appear for Republican members once enough verified accounts exist, or on other decentralized platforms such as Mastodon.
- The paper leaves open whether Kaptur's central position translates into real-world influence such as legislative effectiveness or fundraising; linking centrality to off-platform outcomes would sharpen the 'shadow leader' claim.
- If the missing members without verified accounts were added, centrality rankings could shift; a published robustness check with the promised 5–10% random node removals would tell readers how confident to be.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies the Bluesky network of U.S. Congressional Democrats following the 2024 election. It constructs directed follows, mentions, and repost graphs from verified member accounts, fits ERGMs to test endogenous and exogenous tie-formation mechanisms, and applies BERTopic to compare factional discourse. The authors report that non-leader members (Marcy Kaptur, Dwight Evans, Donald Beyer) hold high structural centrality, that chamber/state/leadership homophily shapes ties, and that Senate leadership is negatively associated with tie formation. The main quantitative evidence appears in Tables 1–3 and Figures 1–5, none of which are included in the submitted manuscript.
Significance. If the quantitative results were available and reproducible, the paper could make a worthwhile contribution to computational political communication research, particularly on decentralized platforms. The mixed-methods design is standard and appropriate, and the paper is candid about not connecting centrality to off-platform popularity measures. The 'hidden influence' claim is interesting but, at present, is not backed by auditable numbers; the node-set inconsistency and missing sensitivity analysis are serious obstacles to assessing that claim.
major comments (3)
- [§4.1 / §3.1] The paper states that 182 of 258 Democrats had identifiable Bluesky accounts, yet §4.1 reports the follows network has 219 nodes and 15,015 edges with density 0.3145. These numbers are mutually consistent only for 219 nodes (219×218×0.3145 ≈ 15,015), implying 37 nodes beyond the verified set. Because every node-level centrality and every ERGM coefficient depends on the node set, please clarify whether these extra nodes are non-members, unverified accounts, or a reporting error, and rerun all analyses on the correct node set.
- [§5 / §3.1] The missing fraction is stated as 16.4% in §5, but the counts in §3.1 (211 House + 47 Senate = 258; 182 identified) imply 76 missing members, or 29.5%. The sensitivity analysis promised in §3.1 (random removal of 5–10% of nodes) is not reported anywhere. Since closeness centrality is especially sensitive to omitted intermediate nodes, the top closeness (0.9495) and betweenness (0.0786) attributed to Rep. Kaptur cannot yet be distinguished from an artifact of which members were included.
- [Tables 1–3; Figures 1–5] The manuscript repeatedly relies on Tables 1–3 (ERGM results) and Figures 1–5 (network visualizations and topic heatmap), but none of these items appears in the submitted text. Without the tables, the coefficients that support the homophily claims and the centrality values that support the 'hidden leaders' claim cannot be audited. Please include all tables and figures (with captions) and, if possible, the replication data or code.
minor comments (4)
- [§3.1] The data collection section does not state the date(s) on which the Bluesky networks were retrieved; this is essential for reproducibility and for interpreting the snapshot.
- [§5] The sentence 'Note that it was difficult to verify republican accounts on the Bluesky platform' reads as an abrupt aside in the conclusion; move the Republican-account verification discussion to the limitations subsection or the data-collection section.
- [Throughout] The terms 'follows', 'followers', and 'following' are used interchangeably (e.g., §4.1 heading vs. §4.1 text); standardize the terminology.
- [References] Several entries lack complete information (e.g., Kleppmann et al. is missing venue; House Democrats and Senate Democrats entries lack access dates).
Circularity Check
No significant circularity: the centrality and ERGM results are empirical measurements fitted to collected Bluesky data, not derivations from their own conclusions.
full rationale
The paper's central claims are empirical observations about the Bluesky follows, mentions, and repost networks of congressional Democrats. Marcy Kaptur's high closeness and betweenness centrality, and the ERGM homophily coefficients, are computed from collected network data using standard measures and statistical models; they are not derived from a definitional identity or from a fitted parameter renamed as a prediction. The authors' self-citations (McCulloh, Armstrong, & Johnson, 2013; Sadayappan, McCulloh, & Piorkowski, 2018) are used as background or methodological precedent, not as load-bearing justifications that force the results. No uniqueness theorem, ansatz adopted solely by citation, or renaming of a known result carries the argument. The paper does contain serious internal-consistency and completeness concerns: the follows network is reported as 219 nodes while only 182 member accounts are claimed, the stated missing-member fraction of 16.4% contradicts the arithmetic (76/258 is about 29.5%), and the sensitivity analysis promised in Section 3.1 is not reported. These are correctness and data-quality risks, not circular reasoning. Because the derivation chain does not reduce to its own inputs, the appropriate circularity score is 0.
Assumptions & free parameters
free parameters (1)
- ERGM homophily coefficients =
not reported (tables missing)
assumptions (5)
- domain assumption The Bluesky accounts mapped to the 182 members are their official accounts.
- domain assumption Missing members do not systematically bias centrality rankings or ERGM estimates.
- domain assumption Follow, mention, and repost ties are a meaningful signal of political communication and influence.
- domain assumption Ideological affiliation is adequately captured by a categorical variable built from caucus memberships.
- domain assumption Bluesky lacks algorithmic amplification, so observed networks are organic.
Cite this review
Pith. "Pith review of Who Leads in the Shadows? ERGM and Centrality Analysis of Congressional Democrats on Bluesky." pith.science (2026). https://pith.science/paper/Y72CYGJB
@misc{pith2026250716858,
author = {Pith},
title = {Pith review of: Who Leads in the Shadows? ERGM and Centrality Analysis of Congressional Democrats on Bluesky},
year = {2026},
howpublished = {\url{https://pith.science/paper/Y72CYGJB}},
note = {Machine review of arXiv:2507.16858}
}
read the original abstract
Following the 2024 U.S. presidential election, Democratic lawmakers and their supporters increasingly migrated from mainstream social media plat-forms like X (formerly Twitter) to decentralized alternatives such as Bluesky. This study investigates how Congressional Democrats use Bluesky to form networks of influence and disseminate political messaging in a platform environment that lacks algorithmic amplification. We employ a mixed-methods approach that combines social network analysis, expo-nential random graph modeling (ERGM), and transformer-based topic mod-eling (BERTopic) to analyze follows, mentions, reposts, and discourse pat-terns among 182 verified Democratic members of Congress. Our findings show that while party leaders such as Hakeem Jeffries and Elizabeth War-ren dominate visibility metrics, overlooked figures like Marcy Kaptur, Donald Beyer, and Dwight Evans occupy structurally central positions, suggesting latent influence within the digital party ecosystem. ERGM re-sults reveal significant homophily along ideological, state, and leadership lines, with Senate leadership exhibiting lower connectivity. Topic analysis identifies both shared themes (e.g., reproductive rights, foreign conflicts) and subgroup-specific issues, with The Squad showing the most distinct discourse profile. These results demonstrate the potential of decentralized platforms to reshape intra-party communication dynamics and highlight the need for continued computational research on elite political behavior in emerging digital environments.
Reference graph
Works this paper leans on
-
[1]
Who Leads in the Shadows? ERGM and Centrality Analysis of Congressional Democrats on Bluesky Gordon Hew1 and Ian McCulloh1[0000-0003-2916-3914] 1 Johns Hopkins University, Baltimore MD 21218, USA imccull4@jhu.edu Abstract. Following the 2024 U.S. presidential election, Democratic lawmakers and their supporters increasingly migrated from mainstream social ...
work page 2025
-
[2]
Bonacich, P. (1987). Power and centrality: A family of measures. American Journal of So-ciology, 92(5), 1170–1182. https://doi.org/10.1086/228631
doi:10.1086/228631 1987
-
[3]
shows strong homophily effects: chamber membership, state, verified status, and most group affiliations significantly increase the likelihood of a tie. However, membership in the Democratic Senate Leadership has a negative coef-ficient, indicating these members are less likely to form reciprocal or follower ties—perhaps reflective of more hierarchical or ...
work page 2021
-
[4]
Congressional Democrats Follower Network ERGM Results. This network uncovers that legislative seniority or media profile may not correlate with online embeddedness. Marcy Kaptur emerges as an unheralded yet structurally in-fluential figure, exemplifying how social network metrics can spotlight overlooked ac-tors (McCulloh, Armstrong, & Johnson, 2013). 4.2...
work page 2013
-
[5]
This sparse network reinforces findings from Himelboim et al
Congressional Democrats Repost Network ERGM Results. This sparse network reinforces findings from Himelboim et al. (2013) that information diffusion in political networks is not solely driven by central elites, and that influential amplifiers often occupy intermediate or bridging roles. Beyer and Evans exemplify such figures: connected enough to receive r...
work page 2013
-
[6]
5 Conclusion This study provides one of the first comprehensive examinations of U.S
Cosine Similarity Heatmap of Top Eight Topics for Democrat Coalitions. 5 Conclusion This study provides one of the first comprehensive examinations of U.S. Congressional Democrats' activity on the decentralized social media platform Bluesky. By integrating social network analysis, exponential random graph models (ERGM), and transformer-based topic modelin...
work page 2024
-
[7]
Ambrose, T. (2024, November 16). What is Bluesky and why are so many people suddenly leaving X for the platform? The Guardian. https://www.theguardian.com/technol-ogy/2024/nov/16/what-is-bluesky-and-why-are-so-many-people-suddenly-leaving-x-for-the-platform-elon-musk
work page 2024
-
[8]
Freeman, L. C. (1979). Centrality in social networks: Conceptual clarification. Social Net-works, 1(3), 215–239. https://doi.org/10.1016/0378-8733(78)90021-7
Show all 18 references
-
[9]
R., Locatelli, M
Buzelin, A., Dutenhefner, P. R., Locatelli, M. S., Malaquias, S., Bento, P., Aquino, Y., Dayrell, L., Estanislau, V., Santana, C., Alzamora, P., Vasconcelos, M., Meira Jr., W., & Almeida, V. (2025). Analyzing Political Discourse on Discord during the 2024 U.S. Presi-dential El...
2025 arXiv
-
[10]
M., Spezzano, F., Kettler, J
Chamberlain, J. M., Spezzano, F., Kettler, J. J., & Dit, B. (2021). A Network Analysis of Twitter Interactions by Members of the U.S. Congress. ACM Transactions on Social Com-puting, 4(1), 1–22. https://doi.org/10.1145/3439827
2021 doi
-
[12]
Grootendorst, M. (2022). BERTopic: Neural topic modeling with class-based TF-IDF. arXiv preprint arXiv:2203.05794. https://arxiv.org/abs/2203.05794
2022 arXiv
-
[13]
Himelboim, I., Smith, M., Rainie, L., Shneiderman, B., & Espina, C. (2013). Classifying Twitter topic networks using social network analysis. Social Media and Society, 1(1). https://doi.org/10.1177/2056305113479598
2013 doi
-
[14]
Kleppmann, M., Frazee, P., Gold, J., Graber, J., Holmgren, D., Ivy, D., Johnson, J., New-bold, B., & Volpert, J. (2024). Bluesky and the AT Protocol: Usable Decentralized Social Media. arXiv
2024
-
[15]
McCulloh, I., Armstrong, H., & Johnson, A. (2013). Social Network Analysis with Applica-tions. Hoboken, NJ: Wiley
2013
-
[16]
(2024, November 19)
NPR Staff. (2024, November 19). Traffic on Bluesky, an X competitor, is up 500% since the election. NPR. https://www.npr.org/2024/11/19/g-s1-34898/bluesky-traffic-surge-after-election:contentReference[oaicite:22]{index=22}
2024
-
[17]
(2016, August)
Ozer, M., Kim, N., & Davulcu, H. (2016, August). Community detection in political twitter networks using nonnegative matrix factorization methods. In 2016 International conference on advances in social networks analysis and mining (ASONAM) (pp. 81-88). IEEE
2016
-
[19]
Sadayappan, S., McCulloh, I., & Piorkowski, J. (2018). Evaluation of Political Party Cohe-sion Using Exponential Random Graph Modeling. In Proceedings of the 2018 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM) (pp. 1–8). IEEE
2018
-
[2024]
Starter Pack
using the BeautifulSoup Python library. The scraped data included names, chamber affiliation, and states, which were stored in a structured CSV file for reproducibility and used to assign node-level attributes in sub-sequent network modeling. 4 To map each congressional Democr...
1979
Reviewed August 6, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.