Self Promotion in US Congressional Tweets

Jun Wang, Kelly Cui, Bei YuNAACL 2021 · DOI

Introduction

Traditional social science research has long suggested that women in professional settings face a "double bind": they must demonstrate competence to succeed, yet self-promotion—the primary tool for asserting competence—often triggers social backlash for violating traditional gender stereotypes. However, new research utilizing Natural Language Processing (NLP) to analyze 2 million tweets from the US Congress reveals a significant reversal of this norm in the digital sphere. Contrary to expectations, congresswomen engage in significantly more self-promotion than their male colleagues, suggesting that social media serves as a vital tool for women to bypass traditional gatekeepers and assert their competence directly to constituents.

The Problem: The Political Double Bind

For politicians, self-advocacy is an essential part of the job. Yet, women in politics have historically been penalized for "aggressive" self-promotion, which can be perceived as a lack of likability. This study investigates the "equalization theory," which posits that the direct-to-consumer nature of social media allows marginalized groups to manage their own impressions more aggressively. The researchers sought to determine if the well-documented gender gap in self-promotion persists in the high-stakes environment of US Congressional Twitter.

Methodology

The study employed a multi-stage pipeline combining manual annotation, deep learning, and statistical modeling. The researchers analyzed a corpus of approximately 2 million tweets from 698 members of the 115th, 116th, and 117th Congresses (2017–2021).

  1. Classification: A BERT-based (Bidirectional Encoder Representations from Transformers) model was trained on a manually annotated subset of 4,003 tweets. To address the sparsity of self-promotional content, the team used an "I-tweet" sampling strategy, over-sampling tweets containing the word "I."
  2. Interpretability: Using LIME (Local Interpretable Model-agnostic Explanations), the researchers identified high-weight terms associated with self-promotion, such as "proud," "honored," "introduced," and "legislation."
  3. Statistical Analysis: To isolate the effect of gender, the study utilized a generalized linear mixed-effects regression model, controlling for political party, age, chamber (House vs. Senate), seniority, tweet frequency, and follower count.

Key Findings: Reversing the Gender Gap

The most striking result is the consistent reversal of traditional gender norms. Even after controlling for institutional and demographic factors, women were significantly more likely to tweet about their accomplishments and competence than men.

Expected mean of self-promotion over time, showing women (F) consistently outperforming men (M) in self-advocacy across the study period.

This trend suggests that social media provides a "leveling up" opportunity. Because women often face systemic obstacles in traditional media coverage, they appear to use Twitter strategically to build their personal brands and communicate their legislative successes directly to the public.

Statistical Significance and Institutional Factors

The regression analysis confirmed that gender is a significant predictor of self-promotional behavior. The model showed a significant positive coefficient for female members, with an odds ratio indicating they are more likely to self-promote than their male counterparts.

Extracted figure from page 7 of Self Promotion in US Congressional Tweets

Other notable findings include:

  • Institutional Differences: Senators self-promote more than House members, likely due to their broader statewide constituencies.
  • The "Focus" of Smaller Accounts: Members with fewer followers and lower tweet frequencies had a higher proportion of self-promotional content, suggesting they use Twitter primarily as a formal press gallery.
  • Age and Activity: Younger members and those with fewer followers showed higher rates of self-promotion, indicating a more aggressive digital strategy among newer or less established politicians.

Conclusion and Future Directions

While the BERT model achieved high performance (macro-F1 of 0.89), the study acknowledges limitations in capturing "implicit" self-promotion, such as tweets where a member attributes success to a constituent to indirectly praise themselves. Furthermore, while the study proves that women self-promote more, the causal mechanisms remain an area for future research. It remains to be seen whether this increased digital self-promotion leads to better electoral outcomes or if it continues to trigger a "likability" backlash in the interactive sections of social media platforms.

Summary updated

BibTeX
@inproceedings{wang2021self,
  author = {Jun Wang and Kelly Cui and Bei Yu},
  title = {{Self Promotion in US Congressional Tweets}},
  booktitle = {Conference of the North American Chapter of the Association for Computational Linguistics},
  year = {2021},
  doi = {10.18653/v1/2021.naacl-main.388},
}
Figure