Covering: Mutable Characteristics and Perceptions of Voice in the U.S. Supreme Court

Abstract

The emphasis on "fit" as a hiring criterion has raised the spectrum of a new form of subtle discrimination (Bertrand and Duflo 2016). Under complete markets, correlations between malleable characteristics and outcomes should not persist (Becker 1957). Yet using data on 1,901 U.S. Supreme Court oral arguments between 1998 and 2012, we document that voice-based snap judgments based on lawyers' identical introductory sentences, "Mr. Chief Justice, (and) may it please the Court?", predict court outcomes. The connection between vocal characteristics and court outcomes is specific only to perceptions of masculinity and not other characteristics, even when judgment is based on less than three seconds of exposure to a lawyer's speech sample. Consistent with employers mistakenly favoring lawyers with masculine voices, perceived masculinity is negatively correlated with winning and the negative correlation is larger in more masculine-sounding industries. The first lawyer to speak is the main driver. Among these petitioners, males below median in masculinity are 7 percentage points more likely to win in the Supreme Court. Female lawyers are also coached to be more masculine and women's perceived femininity predict court outcomes. Republicans, more than Democrats, vote for more feminine-sounding females, while Democrats, but not Republicans, vote for less masculine-sounding men. A de-biasing experiment finds that information reduces 40% of the correlation between perceived masculinity and perceived win, and incentives reduce another 20%. A model shows how the information treatment identifies statistical discrimination and the incentives treatment identifies taste. Perceived masculinity explains additional variance relative to and is orthogonal to the best random forest prediction model of Supreme Court votes.

Attorney Voice and the US Supreme Court

“…The data come from Chen, Halberstam, and Yu (2016a). Oral arguments at the Supreme Court have been recorded since the installation of a recording system in October 1955.…”

“…To control for the possibility of within-voice modeling by raters, instead of the basic design (in which the listener was presented with one voice sample and rates the sample on all scales), Chen, Halberstam, and Yu (2016b) also employed a design with only one question, randomly selected for each voice sample with only 60 clips and fewer subjects. Each voice clip was played aloud only once, in order to capture the respondents' first impressions and to avoid them over thinking their responses (Ballew and Todorov 2007).…”

“…In fact, in close cases affirming the lower court, the model predicts the outcome with only 25% accuracy. 1 Chen, Halberstam, and Yu (2016a) showed that vocal characteristics are predictive of outcomes with the swing voter, which is where the random forest model may do poorly. In brief, the vast majority of judge votes are in easy cases, where extra-legal and extra-ideological factors may play a smaller role.…”

“…The following analysis examines whether vocal characteristics have predictive power in a setting in which there are very powerful market forces that would seem well-poised to arbitrage away any low-cost competitive advantage: oral arguments before the U.S. Supreme Court. This work reports and builds on prior analyses (Chen, Halberstam, andYu 2016a, 2016b) and is based on the data of 1,901 U.S. Supreme Court oral arguments between 1998 and 2012. Specifically, we will examine whether voice-based snap judgments based on lawyers' identical introductory sentences, "Mr. Chief Justice, (and) may it please the Court?"…”

Machine Learning and the Rule of Law

“…• Politics and race also appear to influence judicial outcomes (Schanzenbach 2005;Bushway and Piehl 2001;Mustard 2001;Steffensmeier and Demuth 2000;Albonetti 1997; Thomson and Zingraff 1981;Abrams, Bertrand, and Mullainathan 2012;Boyd, Epstein, and Martin 2010;Shayo and Zussman 2011) as does masculinity (Chen, Halberstam, andYu 2016b, 2016a), birthdays (Chen and Philippe 2017), football game outcomes Eren and Mocan 2016), time of day (Chen and Eagel 2016;Danziger, Levav, and Avnaim-Pesso 2011), weather (Barry et al 2016), name , and shared biographies or dialects (Chen and Yu 2016).…”