Text as Causal Mediators: Research Design for Causal Estimates of Differential Treatment of Social Groups via Language Aspects

Keith, Katherine A.; Rice, Douglas; O’Connor, Brendan

doi:10.18653/v1/2021.cinlp-1.2

Cited by 5 publications

(3 citation statements)

References 52 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Text counterfactuals, drawing from and informing current developments for causal inference in NLP (Keith et al, 2021;Feder et al, 2021;Jin et al, 2021) can be used for training, testing, and explaining models (Wu et al, 2021). We build on research that explores the former -using counterfactually augmented data (CAD) as training examples, to in-vestigate conditions of model robustness and unintended false positive bias.…”

Section: Discussionmentioning

confidence: 99%

Counterfactually Augmented Data and Unintended Bias: The Case of Sexism and Hate Speech Detection

Sen¹,

Samory²,

Wagner³

et al. 2022

Preprint

View full text Add to dashboard Cite

Counterfactually Augmented Data (CAD) aims to improve out-of-domain generalizability, an indicator of model robustness. The improvement is credited to promoting core features of the construct over spurious artifacts that happen to correlate with it. Yet, over-relying on core features may lead to unintended model bias. Especially, constructdriven CAD-perturbations of core featuresmay induce models to ignore the context in which core features are used. Here, we test models for sexism and hate speech detection on challenging data: non-hateful and nonsexist usage of identity and gendered terms. On these hard cases, models trained on CAD, especially construct-driven CAD, show higher false positive rates than models trained on the original, unperturbed data. Using a diverse set of CAD-construct-driven and constructagnostic-reduces such unintended bias.

show abstract

Section: Discussionmentioning

confidence: 99%

Counterfactually Augmented Data and Unintended Bias: The Case of Sexism and Hate Speech Detection

Sen¹,

Samory²,

Wagner³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Quantitative research means that the data from this study are in the form of numbers, and the analysis method uses statistics to test hypotheses (Disman, Ali, & Barliana, 2017). Causal research describe a particular problem through the process of proving cause-and-effect relationships between variables (Keith et al, 2021) . The unit of analysis in this study is the company.…”

Section: Research Methods Research Designmentioning

confidence: 99%

The Impact of Firm Performance and Corporate Governance on Merger and Acquisition Performance With Earning Quality as a Moderating

Sutrahitu

Herawaty

2023

j. ekon. n.a. econo.

View full text Add to dashboard Cite

Penelitian ini bertujuan untuk mengkaji apakah terdapat pengaruh Firm Performance dan Corporate Governance terhadap Merger and Acquisition Performance dengan Earning Quality sebagai moderasi. Penelitian ini merupakan penelitian kuantitatif yang menggunakan perusahan yang melakukan Merger and Acquisition tahun 2017-2020 sebagai sampel, dan menggunakan data perusahan selama 3 tahun, sebanyak 111 perusahaan non-keuangan yang terdaftar di Bursa Efek Indonesia yang digunakan didalam penelitian ini. data penelitian tersebut dikumpulkan dari laporan tahunan, website perusahaan, website bursa efek Indonesia, dan website yahoo finance. Hasil penelitian ini menunjukkan bahwa kinerja perusahaan yang diindikasikan dengan profitability, Liquidity dan Solvability perusahaan tidak memiliki pengaruh positif terhadap Merger and Acquisition Performance. Corporate Governance yang diindikasikan dengan Institutional Ownership dan Board of Independence memiliki pengaruh positif terhadap Merger and Acquisition.

show abstract

“…With the updated data, we undertake a causal mediation analysis, including speech disfluencies, experience, and ideological alignment as mediators. We calculate the natural direct effects (NDE) and natural indirect effects (NIE) via the assumptions and formulas presented in Keith, Rice and O'Connor (2021). We report the overall results in Table 4.…”

Section: Corroboration: Causal Mediation Analysismentioning

confidence: 99%

"Let Me Just Interrupt You": Estimating Gender Effects in Supreme Court Oral Arguments

Cai¹,

Gupta²,

Keith³

et al. 2023

Preprint

View full text Add to dashboard Cite

Oral argument is the most public and visible part of the U.S. Supreme Court’s decision-making process. Yet, what if some advocates are treated differently when they argue their case before the Court, solely because of aspects of their identity unrelated to the case? In this work, we use a causal inference framework to quantify the effect of an advocate’s gender on interruptions of advocates. Leveraging nearly four decades of U.S. Supreme Court oral argument transcript data, we identify a clear and consistent gender effect, with female advocates interrupted more frequently than male advocates. We demonstrate that, for most justices, the gender effect dwarfs other influences on justice interruption behavior including ideological alignment between justices and advocates. Through a series of corroborative analyses, we further demonstrate the consistency and strength of the gendered interruption behavior of the justices.

show abstract

Text as Causal Mediators: Research Design for Causal Estimates of Differential Treatment of Social Groups via Language Aspects

Cited by 5 publications

References 52 publications

Counterfactually Augmented Data and Unintended Bias: The Case of Sexism and Hate Speech Detection

Counterfactually Augmented Data and Unintended Bias: The Case of Sexism and Hate Speech Detection

The Impact of Firm Performance and Corporate Governance on Merger and Acquisition Performance With Earning Quality as a Moderating

"Let Me Just Interrupt You": Estimating Gender Effects in Supreme Court Oral Arguments

Contact Info

Product

Resources

About