The many (distinctive) faces of leadership: Inferring leadership domain from facial appearance

Abstract

Previous research has shown that people form impressions of potential leaders from their faces and that certain facial features predict success in reaching prestigious leadership positions. However, much less is known about the accuracy or meta-accuracy of face-based leadership inferences. Here we examine a simple, but important, question: Can leadership domain be inferred from faces? We find that human judges can identify business, military, and sports leaders (but not political leaders) from their faces with above-chance accuracy. However, people are surprisingly bad at evaluating their own performance on this judgment task: We find no relationship between how well judges think they performed and their actual accuracy levels. In a follow-up study, we identify several basic dimensions of evaluation that correlate with face-based judgments of leadership domain, as well as those that predict actual leadership domain. We discuss the implications of our results for leadership perception and selection.

Introduction

Understanding the factors that predict leader selection is clearly important: A leader influences the achievements of his/her organization and, by extension, the well-being of its members and all those who benefit (or suffer) from the organization's output. Therefore, organizations and their members should have strong incentives to identify and select effective leaders within their domain, namely by relying on objective indicators of leadership quality. Yet, the human mind often relies on superficial cues to form judgments or make decisions, and the choice of which leader to select is no exception: A large and growing literature shows that facial appearances predict success in reaching prestigious leadership positions (Antonakis and Jacquart, 2013, Olivola and Todorov, 2010a). In the domain of politics, numerous studies have found that more competent-looking political candidates garner larger vote shares (e.g., Antonakis and Dalgas, 2009, Ballew and Todorov, 2007, Poutvaara et al., 2009; for a review of this literature, see Olivola & Todorov, 2010a). Voters also seem to favor more attractive candidates (Berggren et al., 2010, Efran and Patterson, 1974) and those who look stereotypically like members of their preferred political party (Olivola, Sussman, Tsetsos, Kang, & Todorov, 2012). Similarly, in the domain of business, studies have found that CEOs who possess certain facial features command higher salaries and are hired by more successful companies (Graham et al., 2014, Harms et al., 2012, Livingston and Pearce, 2009, Pfann et al., 2000, Rule and Ambady, 2008, Wong et al., 2011). And in the military domain, facial dominance was found to predict military rank (Mazur et al., 1984, Mueller and Mazur, 1996, Mueller and Mazur, 1997; although see Loehr & O'Hara, 2013, for evidence that facial morphological correlates of dominance and aggression negatively predict military rank). In sum, there is ample research demonstrating associations, within several leadership domains (politics, business, military, etc.), between certain facial characteristics and success.³ Leaders in a particular domain (e.g., politics) who possess the “right” facial features (e.g., a competent-looking face) tend to be more successful within that domain (e.g., receive more votes) than other (potential) leaders in the same domain who do not possess those features, ceteris paribus.

While the relationship between facial appearance and success within leadership domains is now well established, much less is known about the relationship between facial appearance and selection into particular leadership domains. That is, are certain (visible) facial features associated with being a leader in one domain rather than another? Or, to put it differently, can people discriminate between leaders in one domain (e.g., military leaders) and those in another (e.g., business leaders), just by looking at their faces? This question is important: If leaders in a particular domain share facial features that distinguish them from leaders in other domains, this suggests that domain-specific facial stereotypes may also influence the leadership selection process, above-and-beyond facial cues that are broadly associated with leadership success across several domains (e.g., attractiveness and facial competence). Identifying such domain-specific facial stereotypes would therefore add a new “layer” to the role of face-based inferences in leadership selection.

This paper contributes to this important question in four ways. First, we determine whether people can accurately judge leadership domain from facial cues. To do so, we presented judges with the faces of leaders drawn from four different domains (business, military, politics, and sports) and asked them to infer which domain these leaders belong to. While there is an extensive literature on the (in)accuracy of appearance-based first impressions (e.g., Hassin and Trope, 2000, Olivola and Todorov, 2010b, Zebrowitz and Collins, 1997, Zebrowitz and Montepare, 2008), only a small fraction of these studies have specifically looked at judgments about leaders. Moreover, these studies have either examined the ability of judges to infer specific characteristics about leaders within a particular domain, such as their political orientation (e.g., Carpinella and Johnson, 2013, Jahoda, 1954, Olivola et al., 2012, Olivola and Todorov, 2010b, Wänke et al., 2012), or their ability to determine whether or not someone is a leader (Cherulnik, Turns, & Wilderman, 1990).⁴ We know of no studies that have asked participants to infer which domain a leader belongs to, solely from facial cues.

Second, we examine whether some leadership categories (military leaders, business leaders, etc.) are more easily identified (from facial cues) than others. In particular, we compared the accuracy of face-based leadership inferences across different leadership domains. Studies comparing face-based inferences across domains (e.g., Hassin and Trope, 2000, Olivola and Todorov, 2010b) have found that these judgments vary considerably in their accuracy levels. We might therefore expect that some leadership category inferences will be more accurate than others. In particular, it would be interesting to see whether leaders who are elected by the general population (e.g., U.S. state Governors) have more or less distinct faces than those who are selected by a smaller group of expert members within their domain (e.g., U.S. Army Generals). On the one hand, we might predict that experts, being more knowledgeable (about their respective domains), would be less influenced by superficial appearance cues than most voters (Lenz & Lawson, 2011). On the other hand, since elite members of the same organization tend to be more like-minded than the general voting population, they may be more likely to share common (but possibly erroneous) stereotypes about what good leaders in their domain look like, and therefore to select leaders who possess certain, distinctive facial features. We return to this question, below, after we present the four leadership categories in our study.

Third, we assess the meta-accuracy of face-based leadership judgments – how well people can evaluate their own ability to draw (correct) inferences from facial stimuli. Specifically, we asked our participant–judges to report their confidence in each judgment and to estimate their overall accuracy. We then compared these estimates with their actual likelihoods of correctly inferring leadership category. Research on the validity of face-based inferences has focused, almost exclusively, on the narrow question of accuracy (see Olivola & Todorov, 2010b for a critical discussion of this issue). In contrast, much less attention has been paid to the correspondence (if any) between the confidence that people hold in their face-based judgments (subjective accuracy) and their actual likelihood of being correct (objective accuracy). Yet meta-accuracy is an essential component of judgment validity since it determines whether (and when) one relies on appearances to form impressions: Regardless of their actual (i.e., objective) accuracy-levels, individuals who doubt their ability to draw useful inferences from faces are unlikely to deliberately rely on these judgments (and they risk ignoring a potentially useful social cue), whereas those who trust their first impressions are more likely to do so (and they risk giving these inferences too much weight).⁵ Consequently, the relative weight that individuals place on their first impressions of leaders can impact organizational dynamics, including a leader's ability to exert influence (we return to this point in Section 10). Therefore, an important goal for researchers should be to understand, not just whether human judges can (on average) draw accurate inferences from facial cues, but also the extent to which people recognize whether (as a general rule) and when (depending on the situation) they should rely on these inferences or refrain from doing so. Those few studies that did compare the accuracy and confidence associated with first impressions tended to find that judges were poorly calibrated in their self-evaluations (Ames et al., 2010, Hassin and Trope, 2000). We might therefore predict low levels of meta-accuracy in leadership category inferences. On the other hand, given the sizeable stakes involved in selecting or interacting with leaders – in particular, the high costs of relying on invalid cues and/or failing to rely on valid cues when evaluating a (potential) leader – we might expect judges to be cognizant of their ability (or lack thereof) to infer leadership characteristics from appearances.

Finally, we attempt to identify the basic dimensions of evaluation (e.g., how competent a leader looks) that correlate with face-based inferences of leadership domain. That is, we examine which dimensions of evaluation might potentially underlie inferences concerning leadership domain, as well as those that may actually “give away” a leader's domain. In particular, we had the leaders' faces rated on a variety of personality traits and physical characteristics, and we correlated these ratings with guesses about leadership domain (subjective categorizations), as well as the actual domains of the leaders in our sample (objective categories). To the extent that a particular dimension of evaluation (e.g., facial competence) simultaneously distinguishes (i) leaders who look (stereotypically) like they belong to their domain from those (in the same domain) who do not and (ii) leaders in one domain from those in another, this judgment variable might underlie and contribute to the perception of leadership domain. For example, if business leaders who are easily identified (by their facial features) have very competent-looking faces compared to their less identifiable peers, and also compared to other types of leaders (e.g., sports leaders), this would suggest that people associate facial competence with business leadership (and vice-versa). By contrast, to the extent that a particular evaluation dimension objectively and reliably distinguishes types of leaders, regardless of how stereotypic they look, this variable might be an accurate predictor of leadership domain.

Study 1

Evaluating the accuracy and meta-accuracy of face-based leadership domain inferences

We recruited a large (non-student) sample of (mainly) British participants and presented them with photos of leaders in the U.S. In particular, they were shown pairs of faces drawn from two (of the four) leadership categories (e.g., business leaders and military leaders). On each trial, they were asked to judge which face belongs to a target leadership category (e.g., to identify the business leader in each pair). They also reported their confidence in each judgment. Finally, after each block of trials, they estimated how well they performed. Every participant completed two blocks of trials: one with faces drawn from two of the four leadership categories (e.g., business leaders and military leaders) and a second block with faces drawn from the remaining two domains (e.g., political leaders and sports leaders). We counterbalanced leadership pairings, target categories, and block orderings across participants.