I checked 7 public opinion journals on Thursday, September 17, 2026 using the Crossref API. For the period September 10 to September 16, I found 5 new paper(s) in 3 journal(s).

Journal of Elections, Public Opinion and Parties

Election delay: public support for postponing elections during public health crises
Hyoungrohk Chu, Haruka Nagao, Rigao Liu, William Hatungimana, Leeann Youn
Full text

Public Opinion Quarterly

Voting for Democracy? Testing Political-Cultural Constraints on Democratic Backsliding
Christopher Claassen
Full text
Democratic backsliding continues to pose a clear danger to contemporary democracies. While a growing literature shows that voters often tolerate antidemocratic candidates when partisan or ideological incentives pull in that direction, we know less about whether support for democracy, institutional trust, political tolerance, and other democratic orientations ameliorate such temptations. We address this question with a systematic comparison of six measures of political culture as predictors of electoral resistance to antidemocratic candidates. Using candidate-choice conjoint experiments fielded in three democracies facing backsliding pressures—the United States, Poland, and South Korea—we find that several democratic orientations are associated with greater resistance to antidemocratic candidates. Among them, support for the core institutions and processes of liberal democracy stands out as the most consistent predictor. Its relationship with prodemocratic choice remains meaningful even in “cross-pressured” choices where partisan or ideological incentives pull voters away from the prodemocratic option. These findings suggest that political culture matters for democratic resilience, though some components of it matter much more consistently than others.

Social Science Computer Review

Leveraging Structural Information to Reconstruct Hidden Nodes in Propagation Graphs
Damian FrÄ…szczak
Full text
While widely used for information exchange, social media networks are increasingly exploited to spread rumors and malicious content. It is essential to reconstruct propagation graphs to hold nodes accountable for such propagation. However, the dynamic and vast nature of social networks makes this task challenging. This study introduces Structural Hidden Node Identification (SHNI), a heuristic method for reconstructing partially observed propagation snapshots by identifying likely hidden participants. SHNI relies on the observed active nodes and structural neighborhood information, without requiring temporal traces, message content, user attributes, or network-specific supervised retraining; its structural parameters may be selected by expert knowledge or calibrated once and then reused across networks. Instead of materializing the entire graph in advance, the method supports on-demand neighborhood expansion from observed active nodes toward structurally relevant candidate nodes. It combines local activation pressure with a structural bridging component inspired by sociological principles of opinion formation, including local social influence, information bubbles, and the role of bridging nodes between groups. SHNI explores candidate hidden nodes through a priority-driven neighborhood expansion strategy, in which the most promising candidates are evaluated first, and further exploration stops when the heuristic evidence becomes insufficient. Experiments across different network structures show that SHNI achieves competitive reconstruction quality while maintaining low data requirements and transparent decision logic. These results indicate that SHNI is especially useful as an interpretable structural baseline when temporal traces, message content, or network-specific supervised training data are unavailable. The method may also be adapted to other propagation phenomena, such as epidemic spread or malware diffusion in computer networks.
Parsing Causal Relationships in Social Science Publications
Rasoul Norouzi, Bennett Kleinberg, Jeroen Vermunt, Caspar Van Lissa
Full text
Understanding the causes of phenomena is a key goal of scientific enquiry. For historic reasons, social scientists are reluctant to explicitly discuss causal assumptions. Nevertheless, causal claims do appear in the literature. Systematically cataloging these causal claims and parsing them into structured cause-effect relationship would allow us to synthesize them and, ultimately, construct theories based on oft-repeated causal claims. However, manually extracting these claims is infeasible due to the scale of the literature and linguistic ambiguity, and existing automated methods often lack domain specificity or fail to extract specific cause-effect pairs. To overcome these limitations, we introduce a BERT-based multitask deep learning model trained on a novel, domain-specific benchmark dataset of 3,014 annotated sentences from social science papers. The model simultaneously (1) identifies sentences containing causal claims, (2) extracts cause and effect spans, and (3) links the associated causal pairs. In our study, the multi-head model outperformed a model with sequential architecture, as well as two generative LLMs (Llama 3 8B and Qwen3 8B), which were prompted to code the sentences with a few examples of the coding schema. Specifically, the multi-head model had the highest macro F1-score across all three subtasks and lower false-positive error propagation compared to the sequential model. Its agreement with human coders on the held-out test set (Krippendorff’s α = 0.81) was of similar magnitude to the interrater agreement among the human coders, computed on an interrater reliability subsample (α = 0.80).
Rethinking Scale: The Efficacy of Fine-Tuned Open-Source LLMs in Large-Scale Reproducible Social Science Research
Marcello Carammia, Stefano Maria Iacus, Giuseppe Porro
Full text
Large language models (LLMs) are distinguished by their architecture, parameter size, and performance capabilities. Social scientists have increasingly adopted LLMs to automate high-dimensional text classification tasks which are difficult to scale with human coders. Very large, closed-source models often deliver superior performance. However, reliance on large proprietary models raises concerns about transparency, reproducibility, and costs that make them impractical for large-scale research projects. In contrast, open-source models offer distinct advantages: they can be run locally (ensuring data privacy), fine-tuned for specific tasks, shared within the research community, and integrated into reproducible workflows. On the other hand, open-source models often underperform compared to commercial alternatives. In this study, we assess whether small, open-source LLMs – when fine-tuned appropriately – can match or exceed the performance of commercial alternatives such as GPT-4, while remaining feasible for large-scale academic research. To produce the labeled data that fine-tuning requires, we develop a hybrid annotation workflow in which several LLMs propose labels that human reviewers verify; this infrastructure is what makes fine-tuning feasible at scale. We apply this pipeline to three substantive social science classification tasks: coding tweets for human flourishing indicators, classifying European Parliament questions by policy area, and tagging datasets from the Harvard Dataverse. Our results show that fine-tuned models as small as Llama2-7B achieve performance on par with much larger closed models, offering a reproducible, low-cost alternative for computational social science research. We also explore the limits of fine-tuning effectiveness across model sizes and training set volumes. This work demonstrates a scalable and replicable approach to integrating open-source LLMs into empirical social science workflows.