PaperlayerView exact revision on arXiv

Source-linked AI summary

Politics of Feelings: Emotional Expression and Legislative Effectiveness in the U.S. Congress

Segun Aroyehun

arXiv:2609.10198v1cs.CL

TL;DR

Research has offered limited evidence on whether specific emotions in political speech relate to consequential legislative outcomes. This paper analyzes 1.7 million U.S. congressional speeches with computational emotion measures and finds systematic temporal, policy, and legislator-level variation, alongside emotion-specific and broader affective associations with legislative effectiveness.

  • Problem

    Existing research has primarily described emotional expression, leaving limited evidence on how specific emotions in U.S. congressional speech relate to legislative outcomes.

  • Method

    The study uses a transformer-based classifier to measure eight discrete emotions in more than 1.7 million congressional speeches from 1973 to 2024 and examines their correlates and associations with legislative effectiveness.

  • Results

    Enthusiasm and pride are positively associated with legislative effectiveness, anger is negatively associated, and emotional valence and diversity are positive while intensity is negative.

  • Takeaways & Limitations

    Computational measures of discrete emotions reveal systematic dimensions of congressional communication across time, policy domains, legislators, and legislative effectiveness.

  • Takeaways & Limitations

    Whether these relationships generalize to subnational, supranational, or differently organized legislatures remains an open question.

Abstract

from arXiv · show

Emotions are a pervasive feature of political communication, yet existing research has focused primarily on describing patterns of emotional expression rather than examining whether they are associated with consequential legislative outcomes. We address this gap by investigating the expression and correlates of discrete emotions in more than 1.7 million speeches delivered in the U.S. Congress between 1973 and 2024. Using a transformer-based emotion classifier, we measure eight discrete emotions: anger, fear, disgust, sadness, joy, enthusiasm, pride, and hope. We examine how these emotions vary over time, across policy topics, legislator characteristics, and their relationship with legislative effectiveness. We find that congressional speeches are becoming emotionally expressive over time. Emotional expression also varies systematically across policy domains and ideological positioning of legislators. Notably, the relationship between emotional expression and legislative effectiveness depends on the specific emotions expressed: enthusiasm and pride are positively associated with effectiveness, whereas anger exhibits a negative association. Emotional valence and emotional diversity are positively associated with legislative effectiveness, while emotional intensity is negatively associated with legislative effectiveness. These findings demonstrate that computationally derived measures of discrete emotions can provide insight into affective dimensions of legislative speeches and facilitate our understanding of how legislators communicate, interact, and perform within democratic institutions.

1 Introduction

The study addresses limited U.S. evidence on how specific emotions vary in congressional speech and whether emotional expression relates to legislative effectiveness. It analyzes eight discrete emotions across 1.7 million speeches and examines temporal, policy, legislator-level, and institutional patterns.

  • Existing U.S. congressional research has examined broad speech features, but much less is known about variation in specific emotions over time and across contexts.
  • The study analyzes 1.7 million congressional speeches from 1973 to 2024 using a transformer-based classifier for eight discrete emotions.The emotions are anger, fear, disgust, sadness, joy, enthusiasm, pride, and hope.
  • The paper also evaluates valence, intensity, and diversity as broader emotional characteristics that may provide explanatory power beyond individual emotions.
  • The research questions cover emotional change over time, differences between parties, policy-domain variation, legislator correlates, and links to legislative effectiveness.
  • The paper contributes a richer account of congressional speech and examines whether discrete emotional measures are associated with member-level legislative effectiveness.

2 Related Work

Prior research shows that emotional expression varies across parliamentary settings and speaker contexts, while U.S. congressional studies have focused mainly on broader language features. This motivates examining specific emotional patterns and their association with legislative effectiveness.

  • Research on U.K. and German parliamentary speeches links emotional expression to electoral incentives, institutional position, policy issues, and speaker characteristics.
  • U.S. congressional research has mainly examined emotion–reason balance and epistemic characteristics rather than detailed patterns of specific emotions.
  • Most prior emotion research treats emotional expression as an outcome, motivating analysis of whether speech emotion relates to member-level legislative effectiveness.
  • Emotion measurement ranges from lexicon-based methods to supervised and transformer-based classifiers, with lexicons often overlooking contextual word use.

3 Methods

The study combines a large Congressional Record corpus with transformer-based emotion and topic classification, standardized measures, validation checks, aggregation, and mixed-effects models. These analyses connect emotional expression to temporal, policy, individual, and legislative-effectiveness outcomes.

  • The corpus contains 1,766,113 speech transcripts from the 93rd through 118th Congresses, covering 1973–2024 after procedural interventions were excluded.Every session contains at least 20,000 speeches.
  • A multilabel transformer classifier assigns continuous scores for eight emotions, allowing multiple emotions to co-occur within a speech.
  • Emotion scores are standardized across the full corpus, while valence, intensity, and diversity summarize broader affective characteristics.Diversity is operationalized as entropy of the normalized emotion-score distribution.
  • Classifier face validity is assessed through high-scoring speech inspection, emotion correlations, and manual annotation of high- and low-scoring speeches.
  • Speeches are assigned to 21 CAP policy topics using a transformer topic classifier with macro-averaged F1 of 0.857 on held-out data.
  • Temporal and policy analyses aggregate standardized scores by Congress, party, and topic, using visualizations and Kruskal–Wallis tests for topic differences.
  • Separate mixed-effects models estimate emotion correlates and effectiveness associations while accounting for repeated observations, policy topic, and congressional time.Legislative effectiveness is measured with the Leg-islative Effectiveness Score, which captures progress of sponsored bills from introduction to final passage.

4 Results

Emotional expression generally increased across congressional sessions and varied systematically by policy domain and legislator characteristics. Associations with legislative effectiveness differed by emotion: enthusiasm and pride were positive, anger was negative, and broader valence and diversity were positive while intensity was negative.

  • Temporal trends: Most emotions increased over time for both parties, with notable rises in enthusiasm and pride and significant trends for nearly all emotions except Democratic joy.
  • Temporal trends: Anger increased but fluctuated substantially, with party-specific peaks and higher Republican anger in the 2021–2024 sessions.
  • Policy domains: All eight emotion distributions differed significantly across policy topics, with International Affairs and Law and Crime relatively high in several negative emotions.
  • Policy domains: Education showed the highest average pride alongside elevated joy and enthusiasm, while Health had relatively high enthusiasm and Government Operations relatively high joy.
  • Legislator correlates: Ideology, gender, chamber, party status, and majority status were associated with several emotions, including lower sadness and pride with higher ideology scores.
  • Legislative effectiveness: Enthusiasm (b=0.016, 95% CI [0.007, 0.025], p<0.001) and pride (b=0.011, 95% CI [0.003, 0.019], p < 0.01) were positively associated with legislative effectiveness, whereas anger was negatively associated.Anger had b=-0.037, 95% CI [-0.046, -0.029], p<0.001.
  • Legislative effectiveness: Emotional diversity and valence were positively associated with legislative effectiveness, whereas emotional intensity was negatively associated.

5 Discussion and conclusion

Computationally measured emotions show systematic temporal, substantive, and relational structure in congressional speech, with distinct emotions associated differently with legislative effectiveness.

  • Emotional expressions vary systematically across time, policy domains, legislators, and legislative effectiveness.
  • Temporal patterns: Anger, fear, disgust, sadness, enthusiasm, and pride increase over time, indicating greater emotional expressiveness in congressional speeches.Democrats and Republicans show similar temporal trajectories across most emotions.
  • Policy domains: Policy domains exhibit distinct emotional profiles that remain broadly stable over time, reflecting recurring communicative demands in legislative debate.
  • Communicative functions: High pride scores often involve recognizing colleagues, constituents, communities, or legislative achievements, while high anger scores focus on criticism, responsibility, or policy failures.
  • Communicative functions: Discrete emotions can function as communicative resources through which legislators evaluate issues, position themselves, and pursue legislative objectives.
  • Legislative effectiveness: Enthusiasm and pride are positively associated with legislative effectiveness, whereas anger is negatively associated with it.Emotional diversity and valence are positively associated with effectiveness, while emotional intensity is negatively associated.

6 Limitations

The study’s conclusions are bounded by its U.S. congressional setting, incomplete identification of emotional causes and targets, and selective coverage of discrete emotions.

  • Generalization to other contexts: Whether the observed emotional patterns and effectiveness relationships generalize beyond the U.S. Congress remains an open question.Relevant comparison settings include subnational and supranational legislatures with different institutional and party systems.
  • Causes and targets of emotions: The analysis accounts for policy domains but does not identify the causes and targets of emotional expressions.Identifying them could clarify emotions’ strategic and communicative functions in legislative discourse.
  • A broader range of discrete emotions: The eight analyzed emotions are selective rather than exhaustive.Future research could examine emotions such as curiosity, guilt, shame, and nostalgia.

A Dataset

The dataset contains at least 20,000 speeches from every Congressional session, although corpus size varies across sessions.

  • At least 20,000 speeches are included for each Congressional session in the dataset.Session-level corpus size varies, but each period contains enough observations for reliable analyses of emotional expressions.

B Classifier evaluation

The classifier is evaluated through human-annotation agreement, emotion co-occurrence patterns, and corpus coverage across Congressional sessions.

  • Validation: AUC compares classifier ranking quality by testing whether higher predicted probabilities correspond to human-labeled positive rather than negative cases.The validation sample contains 25 highest- and 25 lowest-scoring responses for each emotion.
  • Corpus coverage: Each Congressional session contains at least 20,000 speeches, although corpus size varies across sessions.The session-level coverage is described as sufficient for reliable analysis of emotional expression.
  • Emotion structure: Negative emotions show modest positive associations with one another, while positive emotions are more differentiated.Cross-valence correlations are generally weak, indicating related but distinct emotion measures.

C Illustrative examples of text with high scores based on the emotion classifier

High-scoring speech examples provide qualitative evidence that the emotion classifiers capture the intended discrete emotional constructs.

  • Illustrative examples: Examples of texts with high classifier scores suggest that the model captures the intended discrete emotions.The examples are presented for each emotion category.

D Regression models

The paper estimates mixed-effects models separately for emotional-expression correlates and for the relationship between emotions and legislative effectiveness.

  • Model design: Two mixed-effects regression sets model emotional-expression correlates and legislative effectiveness, respectively.The second set predicts member-level LES using emotion measures and established legislative-effectiveness controls.
  • Effectiveness model: The legislative-effectiveness models assess how discrete emotions and broader emotional characteristics relate to member-level LES.The reported model outputs cover emotion-specific and aggregate emotional-expression specifications.
  • Emotion model: The emotion model uses standardized average emotion scores as outcomes and includes legislator characteristics, policy-topic controls, Congress fixed effects, and legislator random effects.Legislators are indexed by i and congressional sessions by c.
  • Controls: The analyses use congressional-session and policy-topic structure to account for temporal and substantive differences in legislative speech.The model descriptions identify Congress fixed effects and CAP policy-topic controls as part of the specification.

F Temporal patterns in discrete emotions across CAP topics

The paper examines emotional patterns across CAP policy topics using topic comparisons and temporal analyses, with separate regression tables reporting emotion and effectiveness models.

  • Topic patterns: Figure 7 shows temporal patterns for each discrete emotion across CAP policy topics.The topic-level distributions are compared across 21 CAP policy topics using Kruskal–Wallis tests.
  • Temporal trends: Table 3 reports Mann–Kendall tests and Sen’s slope estimates for monotonic emotion trends by party across U.S. Congresses.P-values are adjusted using the Benjamini–Hochberg false discovery rate procedure.
  • Emotion models: Tables 5 and 6 report mixed-effects estimates for emotions with positive and negative valence.These models include congressional-session and policy-topic fixed effects.
  • Effectiveness models: Tables 7 and 8 report mixed-effects estimates predicting legislative effectiveness from negative- and positive-valence emotions.The tables use LES as the dependent measure and include congressional-session and policy-topic fixed effects.
  • Aggregate characteristics: Table 9 reports mixed-effects estimates linking legislative effectiveness to broader emotional characteristics.The broader characteristics are presented separately from the discrete-emotion models.
Loading 2609.10198v1…