My research aims to uncover subtle acts of influence in strategic interactions across genders and any ensuing impacts on their performance evaluations and outcomes.
My work combines text and acoustic data mining techniques to derive persuasion tactics from novel data sets of inter-collegiate debate tournaments and negotiation exchanges. In my ongoing and upcoming projects, I integrate NLP & ML techniques into online, survey, and field experiments to understand: (1) heterogeneities and the effectiveness of persuasion strategies across social groups and networks; and (2) how group deliberation impacts individual beliefs and preferences.
Below you can find summary abstracts of my papers and projects.
Acoustic-Prosodic, Lexical and Demographic Cues to Persuasiveness in Competitive Debate Speeches (with Ralph Vente, Sarah Ita Levitan, David Lupea and Julia Hirschberg)
[Peer-reviewed Conference Proceeding@ INTERSPEECH 2021] [3-min intro video] [15-min presentation] [Interspeech Slides]
1st Prize STAR Presentation Award, NYAS NDS2020
Presented at: INTERSPEECH 2021 (Brno), STAR Talk @ Natural Language, Dialogue and Speech Symposium of the New York Academy of Sciences
We analyze the acoustic-prosodic and lexical correlates of persuasiveness, taking into account speaker, judge, and debate characteristics in a novel data set of 674 audio profiles, transcripts, evaluation scores, and demographic data from professional debate tournament speeches. By conducting 10-fold cross-validation experiments with linear, LASSO, and random forest regression, we predict how different feature combinations contribute toward speech scores (i.e. persuasiveness) between men and women. Overall, lexical features, i.e. word complexity, nouns, fillers, and hedges, are the most predictive features of speech evaluation scores; in addition to the gender composition of judge panels and opponents. In a combined lexical and demographic feature model, we achieve an R^2 of 0.40. Different lexical features predict speech evaluation scores for male vs. female speakers, and further investigation is necessary to understand whether differential evaluation standards are applied across genders. This work contributes a multi-modal analysis of a large-scale debate data set in a setting with high external relevance to persuasive speech education in other competitive contexts.
The (Great) Persuasion Divide? Gender Disparities in Debate Speeches and Evaluations
Presented at: ASSA/AEA Annual Meeting 2022, Women in Econ Leman – Rare Voices in Economics Conference (Geneva), Digital Methods in History and Economics Workshop, Young Economists’ Meeting, PaCSS 2021 Politics and Computational Social Science Conference, SIOE Society for Institutional & Organizational Economics Conference, SOLE Society of Labor Economists Annual Conference, SMYE Spring Meeting of Young Economists 2021, ESPE European Society of Population Economics Annual Conference 2021, GRAPE Gender Gaps Conference 2021, AYEW Applied Young Economist Webinar, PhD-EVS Seminar, RES Annual Conference, ETH Zurich Seminar in Economics & Data Science, AGEW Australian Gender Economic Workshop , Junior Americanist Workshop, Minerva Lab on Diversity & Gender Inequality, YSI – INET Plenary – “New Approaches to Old Questions : Current Debates in Feminist Economics” , YSI – INET Plenary – “Developing Countries: Challenges to Face Inequality “ , VfS Annual Conference – Gender Economics , Econometric Society World Congress , EALE SOLE AASLE World Conference 2020 , RGS Doctoral Conference in Economics (Dortmund), ESE Female Network Seminar, WUDC Gender Inclusion Panel Discussion (Bangkok), CBS Behavioral Working Group Seminar (New York), 2019 Econometric Society Winter Meeting (ESWM) (Rotterdam), Natural Language, Dialog and Speech Symposium of the New York Academy of Sciences (New York), CBS Chazen Institute Research Scholar Seminar (New York), Data Science Institute Poster Session (New York), Zurich Text As Data Workshop (Zurich), International Association of Applied Econometrics 2019 (Nicosia), TIBER Symposium on Psychology & Economics (Tilburg), Eastern European Machine Learning Summer School 2019 (Bucharest), Data Science Summer School 2018 (Paris), EUR Brownbag Seminar 2019, EUR Diversity Research Seminar 2018
Do men and women persuade differently? Are they evaluated differently? Using a novel data set of 1517 speech transcripts, evaluation scores, and demographic data from highest-profile inter-varsity debate tournaments, this chapter investigates spoken verbal tactics across genders and any ensuing impacts on their performance evaluations. I find significant variation in speech patterns across genders. Female speakers use a more personal and disclosing speaking style, with more hedging phrases and non-fluencies in their speeches. In their answers to questions from opponents, women negate less, while having longer and more vague answers. On average, women receive lower evaluation scores than men. Across debates, having a less analytical speaking style and more positive sentiment is associated with higher scores for speeches by women, but not by men. Within debates, except for non-fluencies, there is no robust evidence of gender-specific evaluation standards. These findings suggest that the gender score gap arises because speeches of female speakers contain more score-reducing and fewer score-enhancing features, rather than discrimination.
Gender Composition of Committees and Performance Evaluation: Evidence from Debate Tournaments
Presented at: EALE European Society of Labor Economists Conference, Max Planck Institute 7th Annual Conference on “Contest: Theory and Evidence” (Berlin), 7th Potsdam Ph.D. Workshop in Empirical Economics, IAAEU Workshop on Gender Inequality in Labor Markets
Does the gender composition of committees matter in evaluating speeches? This research exploits the random composition of 4896 evaluation panels in the European and World Universities Debate Championships to understand the causal impact of women on committees in performance evaluation across genders. On average, female speakers receive significantly lower scores than male speakers. Committees with a female chair judge are harsher to both male and female speakers, particularly in higher-ranked debates. The gender of other committee members does not affect evaluations. While accomplished male chair judges are more generous in scoring, they are notably less so towards female speakers. These results suggest that gender quotas on evaluation committees do not necessarily eliminate the glass ceiling for women in high-stake, repeated competition contexts.
Choking upon Facing (Fe)male Opponents? Evidence from Debate Tournaments
Does the gender composition of opponents affect the performance of men and women in real-world contests? This research exploits the random assignment of 3153 participants to multiple rounds of debate matches in high-profile student debate tournaments to study the causal impact of the gender composition of opponents on speech performance. I find that, on average, the performance of neither men nor women is affected by the gender composition of opponents. In higher-ranked debates, female speakers perform comparatively worse in rooms with more female opponents. These results indicate that more inflow of women into competitions for high-profile careers does not necessarily reduce the thickness of the glass ceiling.
Two of a Kind Try Together? The Impact of Homophily on Innovation Decisions in Teams (with Josse Delfgaauw)
Accessit Best Paper Award at 16th Institutional & Organizational Academy 2017 (Corsica)
Presented at: 33rd European Economic Association Annual Meeting (Cologne), 3rd Lectures on Organizational Economics & Human Resources (Cologne), TI Ph.D. Seminar, EUR Brownbag Seminar
The tendency to disproportionately interact with similar others (i.e. homophily) is a ubiquitous social phenomenon. While it is commonly hypothesized that homophily hampers group creativity and innovation, empirical findings are mixed. This research examines the impact of homophily on innovation in teams, in a simple setting where agents decide whether or not to implement a project that embodies strategic complementarity. Agents receive conditionally correlated private signals about the innovation quality, where homophily is the degree of correlation between the signals. Our key result is as follows: When agents share information truthfully, homophily reduces the probability of implementation; whereas the opposite occurs without information sharing. Given these effects, we discuss an alternative interpretation of homophily as correlated benefits and several extensions in correlation neglect, information collection behavior, and strategic communication within the organizational context.