Voice Pitch and Labor Market Success

Voice Pitch and Labor Market Success William J. Mayew* Duke University [email protected] Christopher A. Parsons University of California at San Diego ca...
Author: Shanon Parsons
2 downloads 0 Views 310KB Size
Voice Pitch and Labor Market Success William J. Mayew* Duke University [email protected] Christopher A. Parsons University of California at San Diego [email protected] Mohan Venkatachalam Duke University [email protected]

August 2011

Preliminary and Incomplete. Please do not cite without permission. Comments welcome. Abstract In this paper we show that a deep voice (i.e., low voice pitch) matters for labor market outcomes. We draw on research in anthropology and psychology that suggests males with lower vocal pitch are perceived to be more physically and socially dominant, and achieve actual success on these dimensions with respect to, for example, mating success and securing of popular votes during presidential elections. We examine whether the vocal pitch of male CEOs is associated with labor market success. We find lower pitched voice males are employed by larger firms and as a result attain both higher levels of compensation and more favorable perceptions by investors/analysts. These findings extend to male MBA students, where we observe lower pitched voice students securing employment at more coveted firms, that in turn, results in greater signing bonuses. Additional analysis does not support the notion that vocal pitch is simply capturing cognitive ability as vocal pitch is not associated with quality of school attended by CEOs or grade point averages of MBA students. Consistent with the social dominance viewpoint, we find CEOs with lower pitched voices have more social connections with other executives and board members. Our study is one of the first to apply evolutionary principles of voice to the business labor market.

_______________ *Address correspondence to Fuqua School of Business, Duke University, Durham, NC 27708. This research was supported by the Fuqua School of Business, Duke University, and the Rady School of Management, University of California at San Diego. We appreciate helpful comments from Dan Hamermesh, Alok Kumar and Sheridan Titman. We are also grateful for guidance and assistance from Sandy Throckmorton and Leslie Collins of the Pratt School of Engineering, Duke University, regarding pitch extraction.

1

Voice Pitch and Labor Market Success 1. Introduction A substantial body of economics research has documented that a person’s objective and subjective physical characteristics (such as height, beauty and obesity) matter for labor market outcomes. For example, Hamermesh and Biddle (1994) use US household data to document that wages of individuals with above-average looks enjoy a significant wage premium. 1 With respect to height, the economics literature is replete with evidence that taller individuals earn more and hold jobs of higher status (see for example, Case and Paxson (2008); Persico, Postlewaite and Silverman (2004)), and offer explanations, ranging from psychological forces such as self-esteem, social dominance and discrimination to physiological differences (e.g., differential cognitive abilities). There is some evidence, albeit weaker, of an obesity wage penalty (see Lindeboom, Lundborg, van der Klaauw (2010)). One common aspect of these – and in fact, nearly all such studies – is that they are based on visual characteristics. As the old adage goes, beauty is beheld in the “eye,” an organ that also detects height, skin tone, or hair color, or weight. In this paper, we take a different perspective, exploring the labor market consequences of an important and hitherto unexplored audio physical attribute, the human voice. Voice is an integral part of communication. In his book, The Third Chimpanzee, Physiologist Jared Diamond popularized the term “The Great Leap Forward” by suggesting that the rapid acceleration of the human evolution over 40,000 years ago was due to “the structure of the larynx, tongue, and the associated muscles that gave humans fine control over spoken

1

Physical characteristics have also been shown to affect other outcomes such as financial portfolio choice and political elections. For example, Todorov, Mandisodza, Goren and Hall (2005) show that a person’s face can influence perceptions about competence of candidates in congressional races that, in turn, influence election outcomes. Korniotis and Kumar (2011) provide evidence that taller and normal-weight individuals are more likely to invest in financial markets and when they do they hold more risky financial portfolios.

2

sounds.” That is, the ability to speak and communicate not only separated humans from the animal kingdom, but propelled the evolution of human civilization. Diamond’s theory is provocative in that he attributes a special, even monumental, historical significance to the human voice. From a practical standpoint, it is well recognized that people express emotions and convey complicated ideas through it. Moreover, voice can be an influential cue in determinations of dominance “because it is an innately used and recognized signal.” (Burgoon, Buller, and Woodall (1996)). Of the various acoustic features of the human voice, we focus on pitch as this feature is considered the most salient and perceptual correlate of the vocal fundamental frequency (Banse and Scherer 1996). Moreover, vocal pitch has been shown to influence perceptions of dominance and attractiveness (Feinberg et al. (2005), Zuckerman, Miyake and Elkin (1995)). That is, individuals with low pitch, i.e., with a deep voice, are perceived to be more persuasive and likely to achieve compliance from others (Burgoon et al. 1996; Gregory and Webster 1996). Indeed, Gregory and Gallagher (2002) show that vocal pitch of competing US presidential candidates obtained during debates from eight different elections predicts the popular vote outcomes in each of those elections. We posit that, if vocal pitch and, in particular low pitch, represents a salient signal of dominance and persuasion we should observe a negative relation between pitch and labor market outcomes. We test this hypothesis in the CEO labor market. The labor market for CEOs is important for two reasons. First, the appointment of CEOs and their compensation are hotly debated topics in the economics literature. Second, CEO compensation varies widely not only across industries but also across firms within an industry. However, the existing literature provides very little guidance on the CEO-specific attributes that are important for CEO selection and compensation. For example, Graham, Li and Qiu (2011) document that manager fixed effects explain a majority of the variation in executive pay suggesting the importance of CEO specific attributes. However, 3

they are silent on the specific attributes of the executive that are important. Recent work by Harvey, Graham and Puri (2010) focus on a physical feature, a CEO’s face, and show that CEOs whose faces are perceived to be more competent receive higher compensation. We extend this line of inquiry by examining the role of voice pitch on both CEO selection as well as their compensation. We use speech samples from public presentations of 788 CEOs in US corporations Specifically, we obtain the fundamental frequency (F0) of spoken words during the first 20 seconds of either the presentation section of quarterly earnings conference calls when available or other public audio presentations when conference calls are not available, to capture the vocal pitch of each CEO. For labor market outcomes, we consider several variables: size of the firm, executive compensation, and institutional investor/analyst recognition. We find that CEOs with a deep voice, i.e., low vocal pitch, are employed in bigger firms, receive relatively higher total compensation, have a larger network of social connections, and are more likely to be rated as a top CEO in their sector by analysts and institutional investors. Our evidence is quite striking in that we are able to demonstrate the influence of voice pitch in organizational setting from two different decision makers, i.e., board of directors and analysts, across three different decision outcomes. Yet, a reader may be skeptical to infer an economically meaningful relationship between voice pitch and labor market outcomes because we use a single cross-section of CEOs for our analysis. To probe whether our results are robust we conduct out-of-sample analyses using job placement and salary data for a sample of male MBA students at a major US university across three different graduating years. 2 In addition to testing for robustness, this sample helps us understand whether the labor market consequences of a low 2

We restrict our attention to male MBA students for two reasons. First, previous research evidence suggests that vocal pitch captures dominance only for males. Second, in our CEO sample all CEOs are males. In order to maintain comparability across the two samples we focus only on the male MBA students. For completeness, we plan to report analysis for the sample of female MBA students as well.

4

pitched voice occur early in a CEO’s career. We find that male MBA students with low voice pitch are more likely to secure employment at companies that are considered prestigious as measured by peer ranking of most sought out job placements by male MBA students. Results for the relation between voice pitch and starting male MBA salary is weak, although we find some evidence that students with lower pitched voices are more likely to receive bonus compensation. Overall, we conclude that a deep male voice has implications for the labor market. Our study identifies a new physical attribute that is related to labor market outcomes. To our knowledge, this paper is the first to document that a deep voice can have tangible benefits in the labor market. We are, however, unable to identify the precise explanation for why deep voice is related to labor market outcomes. While we conjecture that dominance is a plausible explanation for the observed relationship, a limitation of the study is that we do not provide direct tests that offer evidence in support of this conjecture. Could height explain our findings? We do not believe so as there is little robust evidence relating voice pitch and height (see Fitch (2000) and Evans et al. (2006)). Because vocal pitch is determined in part by vocal folds which in turn is influenced by body size and shape, it is conceivable that we are capturing to some extent the height premium in our data. Unfortunately, we do not have data on height for either of the two samples. Hence, we use cognitive ability as a proxy for height as recent research by Case and Paxson (2008) posit that the height premium is due to differential cognitive ability. We use the CEO’s school choice and MBA students’ grade point averages and prior education as proxies for differential cognitive ability and find that vocal pitch is not correlated with either the quality of school attended by the CEO or MBA students’ average GPA or prior education. Thus, we conclude that height is unlikely to be an explanation for our findings.

5

The rest of the paper is organized as follows. In section 2, we develop the reasoning for why vocal pitch may be related to labor market outcomes. Sections three describe the CEO sample, and sections four, five and six investigate the association between CEO voice pitch and firm size, compensation and perceptions of superiority. In section seven we explore the association between male MBA student vocal pitch and labor market success and in section eight we conclude. 2. Hypothesis Development Human voice is an important conduit for communicating ideas, express emotions and even influence people. There is extensive literature on the wealth of information that human voice can convey, ranging from gender (Coleman (1976)), age (Mulac and Giles (1996)), personality (Aronovitch (1976), Zuckerman and Driver (1989)), social status (Harms (1963)) to moods and affective states (Scherer (2003)). There are several acoustic features of voice and are broadly categorized into 1) fundamental frequency (F0), 2) voice intensity, 3) voice quality and 4) temporal aspects of speech (see Harrigan, Rosenthal and Scherer (2008). Of these, the most frequently used acoustic cue is F0, i.e., the fundamental frequency, which closely determines the voice ‘pitch’. F0 is essentially the rate at which the vocal folds vibrate or oscillate and is a function of the length, density and stress of the vocal folds (Titze 2000). 3 Research in anthropology and social psychology suggests that voice pitch is a robust signal of dominance for males. In particular, low pitch, an important component of a person’s deep voice, is associated with perceived physical dominance and perceived attractiveness by women (Feinberg et al. (2005). Low pitch is associated with higher mating success (HodgesSimon et al. (2010)) as well as actual reproductive success (Apicella et al. (2007)). Voice pitch is

3

The nature of human voice depends on the air flow through the vocal tract powered by the larynx. When a person speaks, air is forced through the vocal folds in the larynx, causing them to vibrate and creating the sound.

6

also associated with perceived physical and social dominance in intrasexual interactions in that male human subjects rated other males with a masculine, low pitch voice as more dominant (Puts et al. (2006)). Moreover, in interpersonal interactions they find that in competitive environments males who perceive they are more (less) dominant tend to lower (increase) their voice pitch when addressing others (see also Gregory and Webster 1996). Using real word data from presidential debates and presidential election outcomes, Gregory and Gallagher (2002) find that candidates with lower voice pitch are perceived to be socially dominant and that in turn, helps achieve desirable election outcomes, i.e., more popular votes. In addition to providing a signal of dominance, voice pitch also impacts assessment of other personality traits such as competence and credibility. For example, Apple, Streeter and Krauss (1979) document that men with high-pitched voices are judged as less persuasive, weaker and more nervous. Zuckerman and Miyake (1993) document that males with low pitch are perceived to have an attractive voice that in turn, lead to more favorable personality impressions. In organizational contexts, DeGroot and Motowidlo (1999) use videotaped interviews of managers designed to assess management potential and examine whether vocal cues such as voice pitch can influence interviewer’s judgments about and performance ratings. They find that vocal cues correlate with both performance ratings and interviewer judgments. Moreover, pitch appears to play a part in interviewer decision making as low pitched individuals are perceived to be better managers through a dominant sounding voice. 4 Overall, the evidence in both the psychology and anthropology literature suggests that vocal pitch contain information about personality traits that are useful for decision making. In this paper we draw on this literature to posit that males, with a deep voice, i.e., low pitch, are 4

Real world voice training facilities tout the benefits of vocal training for success in the business world, for both men and women (See http://www.usatoday.com/money/companies/management/2007-07-17-ceo-dominantbehavior_N.htm ).

7

more likely to experience better labor market outcomes. 5 While there is some experimental evidence in the psychology literature on the labor market success for deep voiced individuals (e.g., DeGroot and Motowidlo (1999)) there is very little direct evidence using ecologically valid speech samples. Our paper is an attempt to fill this void. 3. CEO Sample Construction Analysis of vocal pitch of corporate executives requires a speech corpus. We obtain audio files for analysis from ThomsonReuters StreetEvents, who provides audio restreaming of earnings conference calls and other conference presentations. For firms not covererd by ThomsonReuters StreetEvents, we search YouTube and company websites for audio files. Firms commonly do not allow direct download of audio files nor provide an unlimited historical archive of presentation audio. Rather, restreaming of audio is allowed for a limited period typically ranging from 90 days to one year following the event. Thus, audio must be captured while it is available, where audio is restreamed and recorded down for acoustic analysis. Because the process of locating audio for restreaming, isolating the portion of a broadcast that pertains specifically to a given CEO, and executing the restreaming is labor intensive, we sought audio files for the cross section of CEOs analyzed in Engelberg et al. (2011). In particular, we first isolate the latest fiscal year in the Engelberg et al. (2011) sample for fiscal years beginning in 2006 or later. This yields a cross section of 1,409 unique male CEOs. We only consider observations beginning in 2006 because our audio collection efforts began in fiscal year 2006. We only consider males because vocal pitch varies considerably between males and females and there are very few female CEOs. We are able to identify speech samples for 788

5

We restrict our analysis to male voices because we focus on the CEO labor market where over 95% of the CEOs are males.

8

of the CEOs, with 214, 549 and 25 of the cross sectional observations having fiscal year ends in 2006, 2007 and 2008. To procedurally create audio files for analysis, we select the most recent audio file available relative to the fiscal year end of the observation. 6 We then restream the audio to isolate time of the initial 20 seconds of CEO speech. The restreamed audio is encoded in mono directly onto computer hard disk, using Total Recorder 7.1 Professional Edition software, at 11.025 kHz sampling rate and 16 bit quantization, and saved as uncompressed .wav files. Each .wav file is digitally analyzed using Praat acoustics software version 5.2.05 (Boersma and Weenink 2010). We use the GSU Praat “quantifySource” add-on tool with system default settings to extract fundamental frequency, denoted MeanF0 (Owren 2008). 7 Table 1 provides descriptive statistics, where we observe the average voice pitch to be 140.82 Hz (140.82 = exp(LogMeanF0) = exp(4.948)), which falls within the typical adult male range of 85-180 Hz, and exhibits substantial variation. 4. CEO voice pitch and firm size We begin with an important caveat: because our sample is relativelysmall compared to most contemporary studies of CEO labor market outcomes (788 observations), we will be very limited in our ability to perform robustness checks, investigate cross-sectional attenuating factors, or a number of other interesting extensions. On the other hand, there are two advantages. First, the relatively low statistical power of our tests implies that any effects we do find must be large.

6

When an audio file is not available for restream in ThomsonReuters StreetEvents, we search YouTube and Google to identify speeches made by CEOs on television shows or other venues. The majority of our audio files come from ThomsonReuters StreetEvents, with only 83 of the 788 CEO observations coming from non-ThomsonReuters sources. 7 Default parameters are: % of file interval selection = 100%; window in milliseconds = 50; Pitch extraction minimum 75Hz, Pitch extraction maximum 600hz, jitter type = rap, shimmer type = local_dB, periods per window = 4.5). These settings and 20 second file sizes result in roughly 2,085 frames analyzed per file.

9

Second, our results can be represented using simple scatter plots to provide the reader a better appreciation of the economic effects. Figure 1 shows several such two dimensional graphs of voice pitch versus firm size. In the CEO labor market, firms size represents the magnitude of the “empire” over which a CEO can reside. Firm size is measured various ways: Total Assets, Total Sales, market capitalization (Equity), and Number of Employees, all expressed in natural logarithms. Each of these is plotted against the fundamental frequency of the CEO's voice Pitch (Hz), also in natural logarithms. Given that size measures are highly correlated, these plots should not be viewed as independent pieces of evidence; instead, we show them merely to illustrate the robustness of the underlying relationship. The graph in the top left (Figure 1A.1) plots the log-log relationship between Total Assets and Pitch, along with the best-fit OLS line. The magnitude of the pitch-size relation is large. As the related regression output in the first column of Table 2 shows, the elasticity between Pitch and Total Assets is -1.782 (t=-3.83), so that on average, a one standard deviation drop in log pitch (0.139) is associated with an increase in firm size of about 25 percent, or in the neighborhood of $700 million dollars. Another way to judge the magnitude of the effect is to split the sample into equal groups, ranked by vocal pitch, and then sum the Total assets within each group. In untabulated analysis, when we split into quintiles (20% of the observations in each group), we find that the quintile of CEOs with the deepest voices preside over 25% of the total assets. The relationship is even stronger with coarser cuts: roughly 44% of the total assets are controlled by the one third of CEOs with the deepest voices, implying a type of “amplification factor” equal to approximately one third (133%*133% = 1.44).

10

The second column of Table 2 replicates the first column but includes age indicator variables, since voice pitch tends to change with age (Kent 1994) and one might argue that older CEOs are more likely to run larger firms. The voice pitch remains negatively and significantly associated with firm size. The final two columns of Table 2 repeat the regressions in the first two columns, but employ a discrete specification to focus on CEOs with particularly deep voices. Here, we see that a dummy variable for being in the deepest voice pitch quintile (Deep), is associated with a 46.0% larger firm, 41.5% when age controls are included. Each is significant at the one percent level or better. For robustness, we present graphs and regression analysis for the additional measures of firm size, all of which paint similar qualitative and quantitative pictures. As revealed in Figure 1A2 and Table 2 Panel B, the elasticity for Sales is slightly less than for Assets (coefficient = 1.685), but still very statistically significant (t=-4.08) and robust to a discrete specification. Both the continuous and discrete specifications associating vocal pitch and Market Value of Equity (Figure 1A3 and Table 2 Panel C) render inferences very similar to both Sales and Market Value of Equity. Only when we measure size using the Number of Employees do we observe no differences at all, i.e., Deep is not statistically significant at conventional levels. As in the previous regressions, however, the continuous measure, LogMeanF0, is significantly negative, indicating an elasticity of slightly less (in an algebraic sense) than negative one. 5. CEO voice pitch and compensation The previous evidence indicates that voice pitch and firm size are negatively related, with deep-voiced CEOs controlling a disproportionate share of assets. While managing a larger firm can be viewed in and of itself as evidence of labor market success, compensation reflects another dimension by which labor market success can be measured. The bottom two scatterplots in Figure 1 reveal, in the left and right panels respectively, the natural logarithm of each CEO's 11

Salary and Total Compensation, against log Pitch. Perhaps unsurprisingly, the patterns in both graphs closely resemble the size-pitch graphs shown above. The raw salary-pitch elasticity without controls is -0.35, with a heteroskedasticity-robust t-statistic of -3.30 (see first Column of Table 3 Panel A). In the second column, we add a number of traditional controls for CEO pay such as CEO age, tenure, recent stock returns, and idiosyncratic volatility. At first, we do not include firm size because in our analysis firm size itself represents a labor market outcome. We also include a proxy for each CEO's number of external connections (Rolodex), as Engelberg et al. (2011) show that CEOs with more connects receive higher compensation. Following Engelberg et al. (2011), we use the BoardEx database to form each CEO’s Rolodex by summing the number of other executives or directors with whom he: 1) attended school (i.e., “educational connections), 2) worked in the past (i.e., “professional connections”), or 3) is currently affiliated through a common board, charitable organization, or other similar venue (i.e., “social connections). Although the inclusion of these controls somewhat reduces the magnitude of the pitch coefficient, it remains statistically significant, with an elasticity of -0.23. When we control for firm size in the third column, the economic magnitude of voice pitch coefficient plummets and becomes statistically insignificant, indicating that the relation between pitch and salary is almost completely driven through the size channel. Likewise, the discrete specification for pitch and salary, shown in the final three columns of Table 3, do not perform as well, with only the specification without control variables (column four) indicating a statistically significant effect. The relations are considerably stronger for total compensation, as shown in both scatterplot IA6 in Figure 1 and the regression estimation in Table 3, Panel B. Without controls, the first column of Table 3, Panel B indicates that a one percent drop in the CEO's pitch 12

corresponds to a roughly two-third percent increase in total compensation, significant at the one percent level. When controls other than firm size are added in the second column, this elasticity drops only slightly to -.54. The inclusion of firm size (log assets), as with salary, wipes out most of the pitch-pay relation, but retains a negative point estimate. Unlike the regressions with salary alone, the discrete specification performs much better, as can perhaps be inferred from the scatter plot. A CEO with a voice deep enough to place him in the bottom quintile can expect to earn a premium of 10-25%, depending on the specification. Surprisingly, the relation between pitch and total compensation marginally survives the inclusion of Total Assets (column 6), with a t-statistic of -1.67. Thus, although the relation between Vocal Pitch and compensation is mostly driven by its simultaneous relation with size, there is some, albeit weak, evidence that vocal pitch may nevertheless matter on the margin. 6. CEO Voice pitch and the perceptions of others An obvious question is whether having a deep voice is correlated with a CEO’s productivity. If so, then assortive models like Gabaix and Landier (2008) provide theoretical guidance on both the positive vocal pitch-firm size and pitch-pay relations we observe empirically. 8 On the other hand, one might also imagine a deep voice proxying for a CEO’s dominance, and in so doing, provide a measure of the CEO’s power or ability to capture the paysetting process (e.g., Bebchuk and Fried, 2004). Making this distinction is notoriously difficult, particularly for CEOs, where power and productivity may share a causal or complimentary relation. Nonetheless, we try to present at least some suggestive evidence that deep voices may, in fact, reflect at least the perception of

8

In Gabaix and Landier (2008), CEOs with higher marginal product are matched with larger firms. Because productivity is a function of both, pay increases both with CEO “talent” and growth in the firm’s asset base.

13

dominance. Whether these perceptions are accurate we cannot tell, given that we do not observe the CEO’s direct productivity (we observe only firm-level performance data). CEO Network. The first measure of a CEO’s perceived dominance is shown in Table 4, where instead of using the CEO’s voice to explain his pay, we relate it to his external network of other executives and directors, or Rolodex. As shown in Engelberg, et al. (2011), CEO pay is robustly related to the size of his Rolodex, consistent with firms and executives sharing the surplus associated with the network’s value. Consequently, any relation between pitch and compensation might be obscured, if a CEO’s pitch and network size are correlated. The univariate regression specification in the first column indicates a large, negative point estimate relating the logarithm of Pitch to Rolodex, but the t-statistic is only -1.86. However, when we break the Rolodex into its three components, social connections in the fourth column school connections in the seventh column, and professional connections in the tenth column, we see that this overall relationship is driven exclusively by social connections. In particular, the coefficient on the logarithm of Pitch in column four is negative and significant (t-statistic = 2.99), while in columns seven and ten the coefficient is positive and not significant (t-statistic = 0.38 and 0.54, respectively). Keeping in mind that connections from each of these three settings reflect relationships formed only with other public directors and executives, this result suggests that deep-voiced CEOs are disproportionately invited to participate in social organizations involving other CEOs or board members (e.g., sitting on the board of the Bronx Zoo or the United Way). If these venues provide opportunities for the exchange of information that ultimately benefits the firm (e.g., Engelberg et al. (2011)), then one interpretation is that a CEO’s deep voice commands a wage premium via a social networking channel. Of course, the demand for a CEO to be part of a social network may also be influenced by other individual or firm characteristics. As such, we include the same control variables as Table 3. Column five reveals that younger CEOs 14

who have had less time to develop a social network and CEOs working for firms with higher stock return volatility fewer social connections, but vocal pitch remains negatively associated with social connections (t-statistic = -1.99). Only when we also control for firm size in column six do the effects of vocal pitch disappear, consistent with our findings with respect to CEO compensation. Award Winning CEOs. Institutional Investor magazine annually surveys more than 1,500 portfolio managers and analysts to solicit their views on who the top performing CEOs are across over sixty sectors of the economy. Once a year the magazine publishes the “Best CEOs in America” ranking, reflecting the top three vote getters in each economic sector. We acquire an electronic listing of each CEO identified as a top three vote getter via www.instiutitonalinvestor.com from 2005 through 2010. We code any CEO in our sample who was an Institutional Investor award winner during this period as one (AwardII = 1) and zero otherwise. As shown in Table 1, 11.8% of sample CEOs attained award winning status. We then estimate probit regressions to assess whether CEOs with lower pitched voices are more likely to be award winners and report the results in Table 5. The first column of Table 5 reveals the univariate relation, where we observe a significantly negative coefficient on vocal pitch (coefficient = -0.911, p-value = 0.034). In economic terms, an interquartile decrease in vocal pitch increases the predicted probability of being ranked as a top CEO from 10.14% to 13.34%. In the second column, we control for various firm and CEO characteristics as in the compensation analysis earlier. We observe that CEOs with longer tenure, with less volatile stock returns and with more growth options are more likely to be named a top CEO. These associations may not be surprising as analysts would likely value CEOs who they know well, and who operate firms that are somewhat predictable but have untapped growth options. With respect to our variable of interest, the association between award winning and voice pitch remains negative but 15

the statistical significance decreases (p-value = 0.074). In terms of economic significance, in the second column, an interquartile decrease in voice pitch increases the predicted probability of begin named a top CEO 8.57% to 11.14% when other variables are held at their sample means. An interquartile increase in tenure, in comparison, increases the predicted probably from 3.76% to 13.43%. Thus, the effects of tenure are of an order of magnitude larger than the effects of voice pitch. In the third column we also include firm size as measured by logAssets as a control variable. As in earlier analysis, including firm size wipes out the effects of vocal pitch, and CEOs of larger firms exhibit a higher probability of achieving award winning status. The remaining four columns in Table 5 re-estimate the first four columns using a discrete specification for voice pitch, where we observe positive coefficients in each case as expected, but no significant associations. As a collection, the results in Table 5 are consistent with analysts and institutional investors viewing CEOs with lower pitched voices as superior although the statistical associations are not particularly strong. 7.

Analysis of male MBA students Since our CEO sample is only one cross section, the ability to generalize our findings is

limited. Moreover observing that lower pitched male CEOs control larger firms reflects the ultimate matching process in the labor market but provides no insights on how the matching process evolves. To glean further insights, we obtain a sample of male daytime MBA graduate speech samples from a top US MBA program. In particular, incoming daytime MBA students from graduating classes in 2008, 2009 and 2010 create audio files where they at a minimum state their names and in some cases also state where they are from. 9 In our sample, the average 9

In year 2008, creating an audio file was mandatory as part of orientation. In subsequent years it was voluntary. One might conjecture that foreign students may disproportionately select into providing an audio files since at a US business school, foreign students are more likely to have names that other students and faculty find difficult to

16

duration of these files is 5.06 seconds. Links to the audio files are placed on the MBA school intranet so that other students and faculty can learn how to appropriately state the name of the student. The audio file format and the pitch extraction procedure for these audio files are identical to the CEO sample. For each of the students for which an audio file exists, we obtain data from the MBA placement office regarding the name of the student’s employer, the dollar amount of the student’s starting salary, and the amount of any signing bonus. Placement data is self reported and not available for each student. From the admissions office, we obtain student demographic and academic performance data, including gender, age, MBA and undergraduate grade point average, whether the student held an advanced degree prior to joining the MBA program, and whether the student undertook a finance concentration as part of the MBA program. We restrict our attention to male MBA students for comparability to our CEO sample, which is entirely male. We are able to obtain data for 282 male MBA students, where 194, 58 and 30 observations represent graduation years 2008, 2009 and 2010. Descriptive statistics are provided in Panel A of Table 6 for this sample. Average voice pitch is 121.15 Hz (121.1 = exp(logMeanF0) = exp(4.797)). The average voice pitch is lower than the 140.82 Hz average pitch observed for the CEO sample. Given that the median age in the MBA sample is 28 years compared with 56 years of age the CEO sample, difference in average pitch is likely attributable to the increase in pitch in male voices as aging occurs (Kent 1994). To ensure subject age is not a confounding factor we include controls for age in our regression specifications as in the CEO level analysis.

pronounce. To test this conjecture, we utilize the proportion of students by nationality in 2008 as the expected proportion subsequent to 2008. A chi-square goodness of fit test cannot reject that the nationality composition differs from expectations at the 10% significance level.

17

To mimic the CEO analysis, we would ideally like to examine whether the vocal pitch of graduating male MBA students is associated with the size of the firm that hired the student. However, a large majority of students work for non-publicly traded firms where measures of size are not available. As such, we instead use the annual Fortune Magazine “100 Top MBA Employers” rankings. Each year Fortune Magazine surveys graduating MBA students to solicit their views on where MBA students would most like to work. The survey is conducted by gender, and so for the graduating year of each student in our sample, we obtain the ranking for the student’s employer in for the survey year prior to graduation (i.e. the ranking of the employer when the student was interviewing). Fortune Magazine ranks the most desired firm 1 and the least desired firm 100. For empirical analysis, we construct a reverse ranking variable, RF_Rank, which instead scores the top (lowest) ranked firm per Fortune Magazine as 100 (1), all unranked employers are scored as zero. Thus higher values indicate more prestigious employers. We also measure a discrete specification, FRanked, which simply identifies whether the employer was ranked by Fortune Magazine or not. 57.4% of students were employed by a ranked employer in our sample. In Panel B, we provide a Pearson correlation matrix to provide some insights into the MBA sample. We find that both the continuous and discrete Fortune Ranking specifications are positively associated with size of the employer when we could obtain measures of size, suggesting perceptions of employer prestige is correlated with the size of the firm. Fortune Ranked employers appear to offer higher salaries and bonuses. Interestingly, we do not observe a statistically positive correlation between Fortune ranking and MBA grade point average, age of the student, or whether the student possessed an advanced degree beyond the MBA degree. To gain further insights beyond these bivariate relationships, in Table 7 we examine multiple regression determinants of employer prestige. For the discrete Fortune ranking variable, 18

FRanked, we use a Probit estimation whereas for the continue Fortune ranking variable, RF_Rank, we use a Tobit estimation. In the first column we observe a negative relation between pitch and employer rank (coefficient of LogMeanF0 = -0.923; p-value = 0.03). In economic terms, this result implies that a decrease in vocal pitch from the third quintile to the first quintile increases the predicted probability of being employed by Fortune ranked employer from 54.2% to 61.6%. In the second column, we include a host of control variables including, student age, ability as proxied by grade point average and prior schooling, whether the student opted for a finance concentration, nationality, and graduation year fixed effects. In the presence of these control variables, the association between vocal pitch and being hired by a Fortune ranked employer is virtually identical to the first column. Columns three and four replicate the first two columns, but use a discrete specification for voice pitch. The coefficients are positive but not significant. In the final four columns, we estimate tobit regressions where the dependent variable (RF_Rank) is no longer dichotomous, but rather continuous and captures the actual ranking of the employer. Each tobit specification is estimated with an upper (lower) limit of 100 (0). In all specifications, we find statistical support for MBA students with lower pitched voices securing employment at firms that are more prestigious. The final two columns of Table 7 suggest that MBA students in the deepest voice pitch quintile secure employment with firms that are approximately 21 ranks better, on average. In Panel B of Table 7, we investigate whether voice pitch is associated with the starting salary of the student. As before, we stepwise report i) a univariate specification, ii) a specification that includes control variables but excluding the Fortune Ranking and iii) include Fortune Ranking. In the first three columns the coefficient on vocal pitch is negative, but not statistically significant at convention levels. We do observe, however, that younger students, students with 19

higher MBA grade point averages, students who have an advanced degree in addition to the MBA, and students employed by firms with higher Fortune Rankings all command higher salaries. In Panel C, we examine whether vocal pitch influences whether, and to what extent, a student receives a signing bonus. In the first (third) column, we estimate probit (tobit) regression models that include control variables except for the employer’s Fortune Ranking. We observe that students are more likely to receive a bonus, and receive higher bonus levels, when voice pitch is lower. In particular, the coefficient in column one (three) is -0.992 (=3.018) and significant at the 10% level (p-value = 0.065 (0.085)). In column two (four) we replicate column one (three), but include a control for employer prestige. Prestigious employers are more likely to provide signing bonuses and larger bonuses, and after controlling for employer prestige the statistical association with vocal pitch disappears. The disappearance of statistical association for pitch in the MBA sample when employer rank is included mirrors the findings for CEOs when firm size is included. In the final four columns we re-estimate the first four columns using a discrete pitch specification, and all coefficients statistically insignificant. Collectively, the evidence in Table 7 suggests that MBA students with lower pitched voices secure employment by more coveted firms. Vocal pitch plays no role in the level of base salary an MBA student receives, and a slight role bonus compensation before controlling for employer prestige. These results are broadly consistent with the results observed in the CEO sample – males with deeper voices are employed by larger more coveted firms, and receive more compensation as a result of it. 8. Do deep voices proxy for other productivity determinants? A considerable challenge in the growing literature linking physical characteristics to labor market outcomes – particularly wages – is distinguishing between taste-based discrimination a la 20

Becker (1957) and underlying productivity differences. For example, it is now well established that more attractive people earn a wage premium (Hamermesh and Biddle (1994)), as do taller workers. Often however, the physical characteristic of interest is shown to be correlated with some other potential driver of productivity, and this correlation at least partly accounts for the observed wage premium. In these specific cases for example, beautiful people seem to be more confident (Mobius and Rosenblat (2006)), and tall adults are likely to have had positive adolescent experiences (Persico et al. (2004), or may simply be smarter (Case and Paxson (2008)). Similar arguments undoubtedly apply to vocal pitch, and consequently, we will generally not be able to “prove” that it is the voice per se that matters, versus some unobserved (to us) other characteristic with which it is correlated. On the other hand, there are at least two reasons to think that pitch might be special. First, although voice pitch is correlated with testosterone (Evans et al. 2008), it is not systematically related to other observable features such as height or physical size (Fitch 2000; Evans et al. 2006). Thus, at least insofar as we are interested in ruling out one physical characteristic (i.e., voice pitch) simply proxying for another (e.g., height), existing studies seem to indicate that this is possible. More problematic is that voice pitch might be correlated with “softer” productivity determinants like confidence, intelligence, positive life experiences, and the like. Although we will not be able to definitely distinguish between these, we gain some insight into cognitive ability by examining educational attainment and performance, for both the CEO and MBA sample. Returning to our CEO sample, the BoardEx database lists the names of each CEO’s educational institutions, e.g., “University of Texas Law School,” or “Harvard Business School.” While we do not observe a CEO’s academic performance, we can tell whether a CEO was trained 21

at an “elite” institution, as judged by the total number of CEOs it trains. For example, Figure 1 in Engelberg et al. (2011) indicates that Harvard Business School trains more than twice the number of CEOs than its next competitor (Stanford), followed by Harvard University (undergraduate), Wharton, and MIT. Perhaps unsurprisingly, such a list corresponds closely to any number of alternative rankings, such as entrance scores or research funding. We are interested in whether having attended an elite institution is correlated with voice pitch, with the idea that any such correlation might capture cognitive ability. Regardless of exactly how we designate “elite,” we find absolutely no relation between pitch and school quality (results not tabulated). For example, 178 of our 788 CEOs attended at least one of the top five schools listed above. The mean (median) Pitch for those having attended an elite school is 141.5 Hz (140.7 Hz), compared to 142.4 Hz (139.3 Hz) for ones that did not. A nearly identical (non-) result is found if we compare the top 10, 20, or even 50 to the compliment set of schools. A more clinical analysis is feasible using the MBA sample. For this smaller set, we have data not only on their academic performance while in graduate business school, but also for prior educational histories and undergraduate academic performance. In Table 8, we estimate determinant models of MBA grade point average, whether the student held an advanced degree prior to MBA school and undergraduate grade point average. In each specification, we observe no statistically significant associations with vocal pitch. Further, the coefficient signs in the grade point average specifications are more consistent with higher pitched voices, not lower pitched voices, attaining higher grade point averages. As a collection, while we can not completely rule out pitch as a cognitive ability proxy, we find no evidence in our data that would support such a conjecture. 9. Conclusions

22

Our paper is the first to provide evidence that vocal pitch plays a role in CEOs’ labor market outcomes. In particular, we find that deeper voiced CEOs lead larger firms, receive higher pay, and are perceived more favorably by investors and analysts. These findings extend to a sample of MBA students. Students with lower pitched voices secure employment in highly ranked firms and receive higher signing bonuses. This suggests that deep voice shapes the labor market outcome of a CEO much early in their careers. The study is subject to the following limitations. First, although we document that voice is an important physical attribute associated with labor market outcomes, it is possible that some unobserved (or unobservable) characteristic (e.g., testosterone) correlated with voice pitch may account for our results. Even if that were the case, our evidence is still noteworthy because we are able to document the tangible benefits of an otherwise unobservable physical attribute via an observable characteristic, the deepness of an individual’s voice. Second, our evidence is based on a single cross-section of CEOs and MBA students, and hence, we view our paper as a first attempt at uncovering the importance of voice in the labor market. However, the fact that we obtain consistent findings across two different samples gives us comfort that the relationship between voice pitch and labor market outcomes is not spurious. Third, we restrict our analysis on male vocal pitch and an examination of female vocal pitch is a worthy future endeavor. Finally, lack of data precludes us from comparing and contrasting various physical attributes (height, weight, voice and beauty), so as to determine the relative and incremental importance of each of the attributes. We view such comparisons as an important question for future research.

23

References Apicella, C. L., D.R. Feinberg, and F.W. Marlowe, 2007, Voice pitch predicts reproductive success in male hunter–gatherers. Biology Letters: 682–684. Aronovitch, C. D., 1976, The voice of personality: Stereotyped judgments and their relation to voice quality and sex of speaker, Journal of Social Psychology: 207-220. Averett, S., and S. Korenman, 1996, The economic reality of the beauty myth, Journal of Human Resources: 304–330. Banse, R., and K.R. Scherer, 1996, Acoustic profiles in vocal emotion expression, Journal of Personality and Social Psychology: 614-636. Bebchuk, L., and J. Fried, 2004, Pay without performance, Harvard University Press. Becker, G.S., 1957, The Economics of Discrimination, 2nd ed. Chicago: University of Chicago Press. Becker, G.S., 1973, A theory of marriage: part I, Journal of Political Economy: 813–846. Boersma, P., and D. Weenink, 2010, Praat: Doing phonetics by computer (version 5.2.05) [Computer program], Retrieved from http://www.praat.org/. Burgoon, J.K., D.B. Buller, and W.G. Woodall. 1996. Nonverbal Communication: The Unspoken Dialogue. New York: McGraw-Hill. Case, A., and C. Paxson, 2008, Stature and status: height, ability, and labor market outcomes, Journal of Political Economy: 499–532. Cawley, J., 2004, The impact of obesity on wages, Journal of Human Resources: 452–474. Coleman, R.O., 1976, A comparison of the contributions of two voice quality characteristics to the perception of maleness and femaleness in the voice, Journal of Speech and Hearing Research: 168-180. DeGroot T., and S.J. Motowidlo, 1999. Why visual and vocal interview cues can affect interviewers’ judgments and predict job performance, Journal ofApplied Psychology: 986-993. J. Diamond, 1992, The third chimpanzee: the evolution and future of the human animal, HarperCollins, New York. Engelberg, J., P. Gao, and C. Parsons, 2011, The Price of CEO's Rolodex, Working paper, University of North Carolina. Evans, S., N. Neave, and D. Wakelin, 2006, Relationships between vocal characteristics and body size and shape in human males: An evolutionary explanation for a deep male voice, Biological Psychology: 160-163.

24

Evans, S., N. Neave, D. Wakelin, and C. Hamilton, 2008, The relationship between testosterone and vocal frequencies in human males, Physiology and Behavior: 783-788. Feinberg, D.R., B.C. Jones, A.C. Little, D.M. Burt, and D.I. Perrett, 2005, Manipulations of fundamental and formant frequencies affect the attractiveness of human male voices, Animal Behaviour: 561-568. Fitch, W.T., 2000, The evolution of speech: a comparative review, Trends in Cognitive Sciences: 258-267. Gabaix, X. and A. Landier, 2008, Why Has CEO Pay Increased So Much? Quarterly Journal of Economics: 49-100. Graham, J., C.R. Harvey, and M. Puri, 2011, A corporate beauty contest, Working Paper, Duke University. Graham, J., S. Li, and J. Qiu, 2011, Managerial attributes and executive compensation, Working Paper, NBER. Gregory, S.W., and T.J. Gallagher, Spectral Analysis of Candidates' Nonverbal Vocal Communication: Predicting U.S. Presidential Election Outcomes, Social Psychology Quarterly: 298-308. Gregory, S.W., and S. Webster, 1996, A nonverbal signal in voices of interview partners effectively predicts communication accommodation and social status perceptions, Journal of Personality and Social Psychology: 1231-1240. Hamermesh, D., and J. Biddle, 1994, Beauty and the labor market, American Economic Review: 1174–1194. Harms, L. S., 1963, Listener comprehension of speakers of three status groups, Language Speech: 109–112. Harper, B., 2000, Beauty, structure and the labour market: a British cohort study, Oxford Bull. Economic Statistics: 771–800. Hodges-Simeon, C.R., S.J.C. Gaulin, and D.A. Puts, 2010, Voice correlates of mating success in men: Examining “Contests” versus “Mate Choice” modes of sexual selection, Archives of Sexual Behavior, doi:10.1007/s10508-010-9625-0. Kent, R. D., 1994, Reference Manual for Communicative Sciences and Disorders: Speech and Language, First Edition, Pro-Ed, Inc. Korniotis, G.M., and A. Kumar, 2011, Physical attributes and portfolio choice, Working Paper, University of Miami. Kyle, D., and H. Mahler, 1996, The effects of hair color and cosmetic use on perceptions of a female's ability, Psychology of Women Quarterly: 447–455. 25

Lindeboom, M., P. Lundborg, and B. Van der Klaauw, 2010, Asssessing the impact of obesity on labor market outcomes, Economics and Human Biology: 309-319. Mobius, M., and T. Rosenblatt, 2006, Why beauty matters, American Economic Review: 222– 235. Mulac, A., and H. Giles, 1996, You’re only as old as you sound': perceived vocal age and social meanings, Health Communication: 199–215. Owren, M. J., 2008, GSU Praat tools: Scripts for modifying and analyzing sounds using Praat acoustics software, Behavior Research Methods: 822-829. Persico, N., A. Postlewaite, and D. Silverman, 2004, The effect of adolescent experience on labor market outcomes: the case of height, Journal of Political Economy: 1019–1053. Price, M., 2008, Fund-raising success and a solicitor's beauty capital: do blondes, raise more funds?, Economic Letters: 351–354. Rich, M., and T. Cash, 1993, The American image of beauty: media representations of hair color for four decades, Sex Roles: 113–124. Scherer, K.R., 2003, Vocal communication of emotion: A review of research paradigms, Speech Communication: 227-256. Titze, I.R., 2000, Principles of voice production, National Center for Voice and Speech, Iowa City, IA. . Todorov, A., A.N. Mandisodza, A. Goren, and C.C.Hall, 2005, Inferences of competence from faces predict election outcomes, Science: 1623-1626. Zuckerman, M., and R. Driver, 1989, What sounds beautiful is good: the vocal attractiveness stereotype, Journal of Nonverbal Behavior: 67–82. Zuckerman, M., and K. Miyake, 1993, The attractive voice: What makes it so?, Journal of Nonverbal Behavior: 119-135. Zuckerman, M., K. Miyake, and C.C. Elkin, 1995, Effects of attractiveness and maturity of face and voice on interpersonal impressions, Journal of Research in Personality: 253-272

26

Figure 1: Scatterplots of Size and Compensation by Vocal Pitch with Regression Line Figure 1A2: Scatterplot of Natural Log of Total Sales by Log MeanF0

0

2

4

5

logSales 6 8

logAssets

10

10

12

15

Figure 1A1: Scatterplot of Natural Log of Total Assets by Log MeanF0

4.6

4.8

5

5.2 logMeanF0

Fitted values

5.4

5.6

4.6

5

4.8

Fitted values

logAssets

5.4

5.2 logMeanF0

5.6

logSales

Figure 1A4: Scatterplot of Natural Log of Total Employees by Log MeanF0

4

10

6

12

logMVE 14 16

logEmp 8 10

18

12

20

14

Figure 1A3: Scatterplot of Natural Log of Market Capitalization by Log MeanF0

4.6

4.8

5

5.2 logMeanF0

Fitted values

5.4

5.6

4.6

4.8

logMVE

5

5.2 logMeanF0

Fitted values

5.4

5.6

logEmp

Figure 1A6: Scatterplot of Natural Log of Total Compensation by Log MeanF0

4

4

5

6

logSalary 6 7

logTotComp 8 10

8

9

12

Figure 1A5: Scatterplot of Natural Log of Salary by Log MeanF0

4.6

4.8

5

5.2 logMeanF0

Fitted values

5.4

5.6

4.6

logSalary

4.8

5

5.2 logMeanF0

Fitted values

27

5.4

logTotComp

5.6

Table 1: Descriptive Statistics of CEO Sample LogMeanF0 is the natural logarithm of vocal fundamental frequency in Hz, measured with Praat from 20 second CEO speech samples. logAssets, logSales, logMVE and logEmp are the natural logarithm of Total Assets, Total Sales, Market Value of Equity and Total Number of Employees, respectively, each in millions, as of fiscal year end per the Compustat North America Database. LogSalary and LogTotComp are Total Salary and Total Compensation, respectively, in thousands, for each fiscal year from the ExecuComp Database. Rolodex is the sum of other external executives or directors related to the CEO through schools attended, past or current business relationships, and affiliations with charitable or volunteer organizations is identified in the BoardEx database following Engelberg et al. (2011). Rolodex equals the sum of School Connections (Conn_Sch), Social Connections and Past Professional Connections (Conn_Prof), where Past Professional Connections are between executives who no longer work for the same firm, School Connections are between two people that attend the same university and have graduation years that are less than 2 years apart, and Social Connections are between two people who are members of the same social organization. AwardII is an indicator that equals one if the CEO was ranked in Institutional Investor magazine as one of the “Best CEOs in America” in any year from 2005-2010, and zero otherwise. PY_Ret (PY2_Ret) is the one-year (two-year) raw cumulative return ending on the fiscal year end date. Idio_Vol is the idiosyncratic stock return volatility measured as the average squared error taken from a CAPM regression of monthly returns over the past 5 years. MTB is the market value of equity divided by the book value of equity at fiscal year end. Tenure is the time in years since the executive became CEO of the firm. TenureSq equals Tenure*Tenure. AudioAge is the age of the CEO in years as of the date of the audio recording utilized to extract vocal pitch.

logMeanF0 logAssets logSales logMVE logEmp logSalary logTotComp Rolodex Conn_Soc Conn_Sch Conn_Prof AwardII PY_Ret PY2_Ret Idio_Vol MTB Tenure TenureSq AudioAge

N 788 788 788 788 788 788 788 788 788 788 788 788 788 788 788 788 788 788 788

Mean 4.948 7.995 7.712 14.881 8.819 6.603 8.246 111.748 55.476 13.376 42.897 0.118 0.130 0.316 0.002 2.858 6.431 81.305 56.490

Std. Dev. 1st Quartile 0.139 4.856 1.659 6.776 1.502 6.691 1.483 13.866 1.521 7.759 0.409 6.330 0.914 7.594 119.193 20.000 85.511 0.000 20.177 0.000 58.013 5.000 0.323 0.000 0.374 -0.100 0.603 -0.026 0.003 0.001 2.564 1.429 6.324 2.000 168.405 4.000 6.440 52.000

28

Median 3rd Quartile 4.941 5.034 7.801 9.106 7.608 8.764 14.787 15.792 8.784 9.981 6.626 6.904 8.233 8.884 75.500 161.000 18.000 74.500 5.000 17.000 19.000 60.000 0.000 0.000 0.099 0.295 0.219 0.552 0.001 0.003 2.225 3.345 5.000 9.000 25.000 81.000 56.000 61.000

Minimum 4.536 3.014 2.910 10.935 4.543 4.538 5.089 0.000 0.000 0.000 0.000 0.000 -0.776 -0.856 0.000 0.000 0.000 0.000 37.000

Maximum 5.654 14.117 12.790 20.054 13.541 8.580 11.122 729.000 503.000 109.000 481.000 1.000 3.005 5.286 0.031 18.647 39.000 1521.000 82.000

Table 2: Association between Vocal Pitch and Firm Size Proxies This table presents OLS regressions of various size proxies on vocal pitch. LogMeanF0 is the natural logarithm of vocal fundamental frequency in Hz, measured with Praat from 20 second CEO speech samples. Deep is an indicator variable that equals one of logMeanF0 is in the lowest quintile and zero otherwise. logAssets, logSales, logMVE and logEmp are the natural logarithm of Total Assets, Total Sales, Market Value of Equity and Total Number of Employees, respectively, each in millions, as of fiscal year end per the Compustat North America Database. AgeQ1, AgeQ2, AgeQ3, and AgeQ4 indicate age quintiles of the CEO at the time of the audio recording. CEOs in the highest age quintile are captured in the intercept. Robust standard errors are presented in parenthesis below the coefficients. *, **, *** represent statistical significance at better than the 0.10, 0.05, and 0.01 levels, respectively. Pane l A: Association be twe e n Vocal Pitch and Total Asse ts Dependent Variable: logAssets logAssets logAssets logAssets logMeanF0 -1.782 *** -1.669 *** (0.465) (0.467) Deep 0.460 *** 0.415 *** (0.150) (0.151) AgeQ1 -0.117 -0.109 (0.186) (0.186) AgeQ2 0.045 0.062 (0.193) (0.195) AgeQ3 0.262 0.261 (0.196) (0.197) AgeQ4 0.485 ** 0.505 ** (0.196) (0.198) Intercept 16.808 *** 16.115 *** 7.900 *** 7.762 *** (2.301) (2.298) (0.065) (0.151) N 2

Adjusted R

788

788

788

788

0.213

0.0336

0.0111

0.0238

logSales

logSales

Pane l B: Association be twe e n Vocal Pitch and Total Sale s Dependent Variable: logSales logSales logMeanF0 -1.685 *** -1.568 *** (0.413) (0.414) Deep AgeQ1

16.047 *** (2.043)

7.638 *** (0.060)

788

788

788

788

0.0233

0.0453

0.0078

0.031

AgeQ3 AgeQ4

N 2

Adjusted R

0.311 (0.132) -0.062 (0.169) 0.231 (0.173) 0.443 (0.173) 0.595 (0.175) 7.401 (0.133)

-0.058 (0.167) 0.221 (0.172) 0.444 *** (0.171) 0.579 *** (0.173) 15.227 *** (2.034)

AgeQ2

Intercept

0.357 *** (0.131)

29

**

** *** ***

Panel C: Association between Vocal Pitch and Market Value of Equity Dependent Variable: logMVE logMVE logMVE logMeanF0 -1.342 *** -1.269 *** (0.398) (0.398) Deep 0.398 *** (0.135) AgeQ1 -0.033 (0.167) AgeQ2 0.185 (0.175) AgeQ3 0.238 (0.180) AgeQ4 0.382 ** (0.179) Intercept 21.523 *** 21.001 *** 14.802 *** (1.972) (1.979) (0.058) N 2

Adjusted R

2

Adjusted R

0.374 *** (0.135) -0.019 (0.168) 0.202 (0.175) 0.239 (0.181) 0.398 ** (0.179) 14.637 *** (0.136)

788

788

788

788

0.0147

0.0268

0.0103

0.0165

Panel D: Association between Vocal Pitch and Number of Employees Dependent Variable: logEmp logEmp logEmp logMeanF0 -1.164 *** -1.049 *** (0.381) (0.381) Deep 0.170 (0.129) AgeQ1 -0.076 (0.169) AgeQ2 0.155 (0.176) AgeQ3 0.459 *** (0.171) AgeQ4 0.527 *** (0.174) Intercept 14.577 *** 13.794 *** 8.784 *** (1.884) (1.887) (0.061) N

logMVE

logEmp

0.121 (0.131) -0.091 (0.172) 0.155 (0.177) 0.457 *** (0.173) 0.537 *** (0.175) 8.580 *** (0.132)

788

788

788

788

0.0101

0.0304

0.0007

0.0221

30

Table 3: Association between Vocal Pitch and CEO Compensation This table presents OLS regressions of Salary (Panel A) and Total Compensation (Panel B) on vocal pitch. LogMeanF0 is the natural logarithm of vocal fundamental frequency in Hz, measured with Praat from 20 second CEO speech samples. Deep is an indicator variable that equals one of logMeanF0 is in the lowest quintile and zero otherwise. LogSalary and LogTotComp are Total Salary and Total Compensation, respectively, in thousands, for each fiscal year from the ExecuComp Database. AgeQ1, AgeQ2, AgeQ3, and AgeQ4 indicate age quintiles of the CEO at the time of the audio recording. CEOs in the highest age quintile are captured in the intercept. Remaining control variables are defined in Table 1. Industry fixed effects represent Fama-French 30 industry classifications. Robust standard errors are presented in parenthesis below the coefficients. *, **, *** represent statistical significance at better than the 0.10, 0.05, and 0.01 levels, respectively. Panel A: Association between Vocal Pitch and Total Salary Dependent Variable: logSalary logSalary logMeanF0 -0.347 *** -0.227 ** (0.105) (0.100) Deep AgeQ1

-0.119 ** (0.048) -0.091 * (0.046) -0.055 (0.052) -0.041 (0.046) 0.001 *** (0.000) 0.023 (0.054) -0.006 (0.032) -13.413 *** (3.664) -0.014 *** (0.005) 0.005 (0.006) -0.000 * (0.000)

AgeQ2 AgeQ3 AgeQ4 Rolodex PY_Ret PY2_Ret Idio_Vol MTB Tenure TenureSq logAssets Intercept

Year Fixed Effects Industry Fixed Effects N Adjusted R2

8.318 *** (0.521)

8.131 *** (0.504)

logSalary -0.033 (0.084)

-0.116 *** (0.038) -0.098 *** (0.036) -0.089 ** (0.045) -0.077 ** (0.036) 0.000 (0.000) 0.003 (0.040) -0.025 (0.024) -1.682 (2.940) -0.002 (0.004) 0.015 *** (0.004) -0.001 *** (0.000) 0.184 *** (0.010) 5.611 *** (0.449)

logSalary

logSalary

0.072 ** (0.036)

0.025 (0.034) -0.121 ** (0.048) -0.091 * (0.046) -0.057 (0.052) -0.040 (0.046) 0.001 *** (0.000) 0.016 (0.054) -0.006 (0.032) -13.354 *** (3.687) -0.014 *** (0.005) 0.005 (0.006) -0.000 (0.000)

6.587 *** (0.016)

7.012 *** (0.097)

logSalary

-0.009 (0.026) -0.118 *** (0.038) -0.099 *** (0.035) -0.090 ** (0.045) -0.078 ** (0.036) 0.000 (0.000) 0.002 (0.040) -0.025 (0.024) -1.660 (2.943) -0.002 (0.004) 0.015 *** (0.004) -0.001 *** (0.000) 0.185 *** (0.010) 5.444 *** (0.120)

YES YES 788

YES YES 788

YES YES 788

YES YES 788

YES YES 788

YES YES 788

0.0127

0.2376

0.5377

0.0037

0.2323

0.5377

31

Panel B: Association between Vocal Pitch and Total Compensation Dependent Variable: logTotComp logTotComp logMeanF0 -0.662 *** -0.538 ** (0.252) (0.230) Deep AgeQ1

-0.153 (0.107) -0.107 (0.104) -0.114 (0.107) -0.054 (0.100) 0.003 *** (0.000) 0.135 (0.133) 0.196 ** (0.088) -14.037 * (8.522) -0.010 (0.011) -0.022 * (0.013) -0.000 (0.001)

AgeQ2 AgeQ3 AgeQ4 Rolodex PY_Ret PY2_Ret Idio_Vol MTB Tenure TenureSq logAssets Intercept

Year Fixed Effects Industry Fixed Effects N Adjusted R2

11.520 *** (1.247)

11.699 *** (1.172)

logTotComp -0.107 (0.174)

-0.147 * (0.082) -0.123 (0.080) -0.190 ** (0.086) -0.133 * (0.079) 0.001 *** (0.000) 0.092 (0.092) 0.154 ** (0.064) 12.02 * (6.571) 0.017 (0.011) -0.001 (0.009) -0.001 (0.000) 0.409 *** (0.023) 6.101 *** (0.949)

logTotComp

logTotComp

0.257 *** (0.089)

0.176 ** (0.080) -0.143 (0.108) -0.098 (0.104) -0.117 (0.107) -0.046 (0.100) 0.003 *** (0.000) 0.121 (0.132) 0.194 ** (0.088) -13.66 (8.515) -0.011 (0.011) -0.023 * (0.013) -0.000 (0.001)

8.193 *** (0.035)

9.037 *** (0.223)

logTotComp

0.100 * (0.060) -0.136 * (0.081) -0.115 (0.080) -0.190 ** (0.086) -0.130 * (0.078) 0.001 *** (0.000) 0.090 (0.092) 0.152 ** (0.064) 12.195 * (6.523) 0.016 (0.011) -0.001 (0.009) -0.001 (0.000) 0.409 *** (0.023) 5.570 *** (0.268)

YES YES 788

YES YES 788

YES YES 788

YES YES 788

YES YES 788

YES YES 788

0.0089

0.2507

0.5480

0.0114

0.2499

0.5497

32

Table 4: Association between Vocal Pitch and CEO Connections This table presents OLS regressions of CEO connections on vocal pitch. LogMeanF0 is the natural logarithm of vocal fundamental frequency in Hz, measured with Praat from 20 second CEO speech samples. Rolodex is the sum of other external executives or directors related to the CEO through schools attended, past or current business relationships, and affiliations with charitable or volunteer organizations is identified in the BoardEx database following Engelberg et al. (2011). Rolodex equals the sum of School Connections (Conn_Sch), Social Connections and Past Professional Connections (Conn_Prof), where Past Professional Connections are between executives who no longer work for the same firm, School Connections are between two people that attend the same university and have graduation years that are less than 2 years apart, and Social Connections are between two people who are members of the same social organization. AgeQ1, AgeQ2, AgeQ3, and AgeQ4 indicate age quintiles of the CEO at the time of the audio recording. CEOs in the highest age quintile are captured in the intercept. Remaining control variables are defined in Table 1. Industry fixed effects represent Fama-French 30 industry classifications. Robust standard errors are presented in parenthesis below the coefficients. *, **, *** represent statistical significance at better than the 0.10, 0.05, and 0.01 levels, respectively. Dependent Variable: logMeanF0

Rolodex -57.944 * (31.222)

AgeQ1 AgeQ2 AgeQ3 AgeQ4 PY_Ret PY2_Ret Idio_Vol MTB Tenure TenureSq logAssets Intercept

Year Fixed Effects Industry Fixed Effects N Adjusted R2

398.428 ** (155.053)

Rolodex Rolodex Conn_Soc Conn_Soc Conn_Soc Conn_Sch -29.899 13.044 -67.995 *** -43.873 ** -13.400 2.049 (30.438) (27.401) (22.744) (22.046) (19.743) (5.325) -43.183 *** -34.873 ** -20.639 * -14.743 (15.582) (14.170) (11.101) (10.181) -38.597 ** -32.991 ** -21.206 * -17.228 * (15.845) (14.349) (11.265) (10.303) -49.466 *** -47.115 *** -19.429 * -17.761 * (15.627) (13.820) (11.241) (10.060) -7.989 -13.454 -8.117 -11.995 (16.319) (14.653) (11.569) (10.582) -7.555 -10.003 -9.213 -10.950 (14.329) (13.905) (9.856) (9.525) -3.769 -6.770 2.319 0.189 (8.931) (8.916) (6.241) (5.978) -1243.183 1252.570 -1470.390 ** 300.643 (1043.526) (973.603) (603.618) (623.658) 4.548 ** 6.077 *** 0.982 2.067 * (1.783) (1.699) (1.149) (1.057) 1.805 3.293 ** 1.709 2.765 ** (1.643) (1.574) (1.164) (1.098) -0.080 -0.098 -0.060 -0.073 * (0.060) (0.060) (0.041) (0.040) 35.679 *** 25.319 *** (3.338) (2.449) 292.551 * -248.230 * 391.882 *** 283.104 *** -100.644 3.237 (150.207) (142.194) (113.249) (108.769) (98.666) (26.330)

Conn_Sch Conn_Sch 5.805 9.013 (5.574) (5.517) -8.635 *** -8.014 *** (2.663) (2.658) -1.939 -1.520 (2.842) (2.824) -6.983 ** -6.808 ** (2.780) (2.746) -1.417 -1.825 (2.898) (2.890) 0.375 0.192 (3.066) (3.023) -2.036 -2.261 (1.736) (1.705) -91.551 94.908 (228.289) (224.925) 0.458 0.572 (0.388) (0.381) 0.765 *** 0.876 *** (0.293) (0.294) -0.018 -0.019 * (0.011) (0.011) 2.666 *** (0.624) -28.414 -68.816 ** (27.060) (27.996)

Conn_Prof 8.002 (14.095)

3.308 (69.667)

Conn_Prof Conn_Prof 8.169 17.430 (13.992) (13.657) -13.909 * -12.117 * (7.342) (7.167) -15.452 ** -14.243 ** (6.888) (6.665) -23.053 *** -22.546 *** (6.764) (6.529) 1.544 0.366 (8.116) (7.786) 1.283 0.756 (7.661) (7.664) -4.051 -4.698 (5.058) (5.204) 318.759 857.02 (886.266) (882.055) 3.109 *** 3.439 *** (1.090) (1.100) -0.669 -0.348 (0.859) (0.885) -0.002 -0.006 (0.034) (0.035) 7.695 *** (1.939) 37.861 -78.769 (69.061) (72.634)

NO NO 788

YES YES 788

YES YES 788

NO NO 788

YES YES 788

YES YES 788

NO NO 788

YES YES 788

YES YES 788

NO NO 788

YES YES 788

YES YES 788

0.0033

0.1016

0.2632

0.0110

0.1114

0.2696

-0.0011

0.0415

0.0719

-0.0009

0.0651

0.0958

33

Table 5: Association between Vocal Pitch and CEO Institutional Investor Award This table presents probit regression estimations of the probability of a CEO being denoted one of the Best CEOs in America by Institutional Investor magazine as a function of vocal pitch. LogMeanF0 is the natural logarithm of vocal fundamental frequency in Hz, measured with Praat from 20 second CEO speech samples. Rolodex is the sum of other external executives or directors related to the CEO through schools attended, past or current business relationships, and affiliations with charitable or volunteer organizations is identified in the BoardEx database following Engelberg et al. (2011). Rolodex equals the sum of School Connections (Conn_Sch), Social Connections and Past Professional Connections (Conn_Prof), where Past Professional Connections are between executives who no longer work for the same firm, School Connections are between two people that attend the same university and have graduation years that are less than 2 years apart, and Social Connections are between two people who are members of the same social organization. AgeQ1, AgeQ2, AgeQ3, and AgeQ4 indicate age quintiles of the CEO at the time of the audio recording. CEOs in the highest age quintile are captured in the intercept. Remaining control variables are defined in Table 1. Industry fixed effects represent Fama-French 30 industry classifications. Robust standard errors are presented in parenthesis below the coefficients. *, **, *** represent statistical significance at better than the 0.10, 0.05, and 0.01 levels, respectively.

Dependent Variable: logMeanF0

AwardII -0.911 ** (0.431)

AwardII -0.835 * (0.465)

AwardII -0.387 (0.508)

Deep

AwardII

0.201 (0.138)

AgeQ1

0.169 (0.222) 0.018 (0.212) 0.018 (0.216) 0.297 (0.199) 0.135 *** (0.027) -0.004 *** (0.001) -0.206 (0.323) -0.186 (0.211) -86.365 * (30.122) 0.036 * (0.024)

AgeQ2 AgeQ3 AgeQ4 Tenure TenureSq PY_Ret PY2_Ret Idio_Vol MTB logAssets Intercept

N 2

Pseudo R

AwardII

AwardII

0.174 (0.147) 0.156 (0.222) 0.022 (0.212) 0.018 (0.216) 0.304 (0.199) 0.134 *** (0.027) -0.003 *** (0.001) -0.229 (0.320) -0.182 (0.210) -88.843 * (30.263) 0.036 * (0.024)

0.299 (0.251) 0.150 (0.235) 0.053 (0.241) 0.199 (0.223) 0.200 *** (0.032) -0.005 *** (0.001) -0.283 (0.402) -0.390 (0.270) -20.372 (32.013) 0.092 *** (0.027) 0.454 *** (0.051) -4.471 * (2.586)

-1.229 *** (0.066)

-1.863 *** (0.233)

0.048 (0.163) 0.292 (0.252) 0.151 (0.236) 0.055 (0.242) 0.201 (0.223) 0.199 *** (0.032) -0.005 *** (0.001) -0.294 (0.400) -0.388 (0.269) -21.401 (32.110) 0.093 *** (0.027) 0.455 *** (0.051) -6.394 *** (0.599)

3.312 (2.126)

2.283 (2.288)

788

788

788

788

788

788

0.008

0.108

0.282

0.004

0.104

0.281

34

Table 6: Descriptive Statistics of Male MBA Sample LogMeanF0 is the natural logarithm of vocal fundamental frequency in Hz, measured with Praat from MBA student speech samples. RF_Rank is the employer perception ranking by male MBA graduates published by Fortune Magazine in the year of the student’s graduation year, where the top 100 employers are listed annually. The top employer (bottom) ranked employer is scored as 100 (1), and unranked employers are scored as zero. FRanked is an indicator variable that equals one RF_Rank > 0, and zero otherwise. LogAssets, logSales, logMVE and logEmp are the natural logarithm of Total Assets, Total Sales, Market Value of Equity and Total Number of Employees, respectively, each in millions, as of fiscal year end per the Compustat North America Database (these variables are not always available for sample employers who are not publicly traded companies). LogSalary is the natural logarithm of the student’s self reported base salary, in dollars. LogBonus is the natural logarithm of the student’s self reported signing bonus in dollars, where we set logBonus to zero in the event the student does not receive a signing bonus. Bonus is an indicator variable that equals one if the student self reported receiving a signing bonus and zero otherwise. AudioAge is the age of the MBA student in years as of the date of the audio recording utilized to extract vocal pitch. Grad_School is an indicator variable that equals one if the student entered MBA school with a degree beyond a bachelor’s degree, and zero otherwise. LogMBA_gpa is the natural logarithm of the student’s final MBA grade point average, on a four point scale. LogUNDER_gpa is the natural logarithm of the student’s final undergraduate grade point average, on a four point scale. Fin_Conc is an indicator variable that equals one if the student elected to undertake a finance concentration as part of the MBA degree and zero otherwise. Panel A provides summary statistics for each of the aforementioned variables, while Panel B provides a Pearson correlation matrix with p-values presented in parenthesis below the correlation coefficients.

Panel A: Summary Statistics for Male MBA Student Sample N Mean Std. Dev. 1st Quartile Median 3rd Quartile logMeanF0 282 4.797 0.177 4.681 4.767 4.887 RF_Rank 282 40.684 40.338 0.000 39.000 85.000 FRanked 282 0.574 0.495 0.000 1.000 1.000 logAssets 171 10.983 2.314 9.513 10.586 13.446 logSales 171 10.196 1.485 9.717 10.469 11.229 logMVE 168 10.512 1.655 9.800 10.850 11.898 logEmp 170 4.039 1.468 3.352 4.369 5.136 logSalary 282 11.511 0.208 11.430 11.513 11.653 logBonus 282 8.381 3.695 9.210 9.903 10.309 Bonus 282 0.840 0.367 1.000 1.000 1.000 AudioAge 282 28.957 2.967 27.000 28.000 31.000 Grad_School 282 0.227 0.420 0.000 0.000 0.000 logMBA_gpa 282 1.274 0.059 1.229 1.276 1.315 logUNDER_gpa 188 1.196 0.133 1.128 1.205 1.273 Fin_Conc 282 0.301 0.460 0.000 0.000 1.000

35

Minimum 4.446 0.000 0.000 4.526 5.281 5.163 -0.468 10.692 0.000 0.000 20.000 0.000 1.123 0.122 0.000

Maximum 5.626 100.000 1.000 14.915 12.790 13.131 5.990 12.206 10.968 1.000 40.000 1.000 1.386 1.386 1.000

Pane l B: Pe arson Corre lation Matrix for Male MBA Stude nt Sample Variable 1 2 3 4 5 1 logMeanF0 1.000

6

7

8

9

10

11

12

13

14

2 RF_Rank

-0.124 (0.037)

1.000

3 FRanked

-0.128 (0.031)

0.870 (0.000)

1.000

4 logAssets

-0.058 (0.448)

0.562 (0.000)

0.498 (0.000)

1.000

5 logSales

0.045 (0.559)

0.576 (0.000)

0.592 (0.000)

0.866 (0.000)

1.000

6 logMVE

-0.140 (0.070)

0.693 (0.000)

0.671 (0.000)

0.720 (0.000)

0.824 (0.000)

1.000

7 logEmp

0.088 (0.253)

0.383 (0.000)

0.478 (0.000)

0.674 (0.000)

0.870 (0.000)

0.653 (0.000)

1.000

8 logSalary

-0.014 (0.822)

0.269 (0.000)

0.240 (0.000)

-0.115 (0.134)

0.044 (0.564)

0.063 (0.421)

0.111 (0.149)

1.000

9 logBonus

-0.109 (0.069)

0.389 (0.000)

0.455 (0.000)

0.207 (0.007)

0.227 (0.003)

0.129 (0.096)

0.251 (0.001)

0.266 (0.000)

1.000

10 Bonus

-0.110 (0.066)

0.344 (0.000)

0.408 (0.000)

0.145 (0.058)

0.185 (0.015)

0.090 (0.246)

0.226 (0.003)

0.248 (0.000)

0.990 (0.000)

1.000

11 AudioAge

0.071 (0.235)

0.001 (0.988)

0.009 (0.875)

-0.077 (0.316)

0.001 (0.991)

0.005 (0.951)

0.033 (0.672)

0.023 (0.701)

-0.014 (0.818)

-0.013 (0.831)

1.000

12 Grad_School

0.019 (0.755)

0.020 (0.743)

0.004 (0.947)

-0.001 (0.993)

-0.036 (0.644)

-0.021 (0.785)

0.006 (0.938)

0.160 (0.007)

0.043 (0.475)

0.051 (0.392)

0.322 (0.000)

1.000

13 logMBA_gpa

0.056 (0.349)

0.015 (0.803)

-0.105 (0.077)

-0.029 (0.707)

-0.001 (0.986)

0.001 (0.986)

0.040 (0.601)

0.182 (0.002)

-0.026 (0.666)

-0.016 (0.784)

-0.046 (0.440)

0.100 (0.094)

1.000

14 logUNDER_gpa

0.078 (0.288)

-0.049 (0.501)

-0.072 (0.329)

-0.217 (0.017)

-0.172 (0.060)

-0.196 (0.033)

-0.130 (0.156)

0.214 (0.003)

-0.136 (0.063)

-0.148 (0.043)

-0.100 (0.171)

0.079 (0.284)

0.341 (0.000)

1.000

15 Fin_Conc

0.095 (0.112)

0.021 (0.725)

0.018 (0.760)

0.264 (0.001)

0.154 (0.044)

0.078 (0.317)

0.091 (0.236)

-0.035 (0.564)

-0.134 (0.024)

-0.157 (0.008)

0.109 (0.069)

0.068 (0.252)

0.132 (0.027)

0.104 (0.154)

36

Table 7: Association between Vocal Pitch and Employer Prestige for the Male MBA Student Sample In Panel A, this table estimates probit or tobit regressions associating voice pitch with employer prestige as indicated by Fortune Magazine. In Panel B, this table estimates OLS regressions associating voice pitch with base salary. In Panel C, this table estimates probit or tobit regressions associating voice pitch with whether the student received a signing bonus and the amount of the bonus. LogMeanF0 is the natural logarithm of vocal fundamental frequency in Hz, measured with Praat from MBA student speech samples. Deep is an indicator variable that equals one of logMeanF0 is in the lowest quintile and zero otherwise. AgeQ1, AgeQ2, AgeQ3, and AgeQ4 indicate age quintiles of the MBA student at the time of the audio recording. MBA students in the highest age quintile are captured in the intercept. LogSalary is the natural logarithm of the student’s self reported base salary, in dollars. LogBonus is the natural logarithm of the student’s self reported signing bonus in dollars, where we set logBonus to zero in the event the student does not receive a signing bonus. Bonus is an indicator variable that equals one if the student self reported receiving a signing bonus and zero otherwise. Nationality fixed effects indicate whether the student is from the United States/Canada/Mexico, South America or Asia. Graduation year fixed effects are indicator variables for calendar year 2009 and 2010. Remaining control variables are defined in Table 6. Robust standard errors are presented in parenthesis below the coefficients. *, **, *** represent statistical significance at better than the 0.10, 0.05, and 0.01 levels, respectively. Pane l A: Association be twe e n Vocal Pitch and Employe r Pre stige Dependent Variable: FRanked FRanked FRanked logMeanF0 -0.923 ** -0.940 ** (0.425) (0.473) Deep 0.302 (0.191) AgeQ1 -0.121 (0.263) AgeQ2 0.001 (0.246) AgeQ3 -0.134 (0.276) AgeQ4 -0.136 (0.263) logMBA_gpa -2.490 * (1.417) Grad_School 0.000 (0.197) Fin_Conc 0.145 (0.176) Intercept 4.615 ** 7.687 *** 0.128 (2.041) (2.848) (0.084) Nationality Fixed Effects Graduation Year Fixed Effects N Pseudo R2 Model Type

FRanked

0.307 (0.201) -0.120 (0.265) 0.029 (0.247) -0.132 (0.278) -0.102 (0.263) -2.628 * (1.413) 0.020 (0.197) 0.131 (0.175) 3.333 * (1.823)

RF_Rank -57.689 ** (25.237)

RF_Rank -62.015 ** (27.295)

RF_Rank

20.707 ** (10.443)

298.457 ** (120.671)

-6.099 (14.682) -2.608 (13.700) -12.889 (15.805) -12.151 (14.301) -36.763 (76.229) 0.728 (11.235) 7.153 (9.627) 365.710 ** (158.066)

RF_Rank

17.661 *** (5.699)

21.358 ** (10.738) -6.053 (14.875) -1.456 (13.863) -12.973 (16.003) -10.328 (14.396) -43.853 (75.811) 1.788 (11.217) 5.807 (9.587) 75.110 (97.362)

NO NO 282

YES YES 282

NO NO 282

YES YES 282

NO NO 282

YES YES 282

NO NO 282

YES YES 282

0.0121 Probit

0.0364 Probit

0.0066 Probit

0.0328 Probit

0.0027 Tobit

0.0066 Tobit

0.0020 Tobit

0.0059 Tobit

37

Panel B: Association between Vocal Pitch and Base Compensation Dependent Variable: logSalary logSalary logSalary logMeanF0 -0.016 -0.056 -0.011 (0.065) (0.065) (0.067) Deep AgeQ1

0.000 (0.041) -0.041 (0.036) -0.078 (0.042) -0.091 (0.044) 0.483 (0.204) 0.081 (0.030) -0.027 (0.024)

AgeQ2 AgeQ3 AgeQ4 logMBA_gpa Grad_School Fin_Conc

0.003 (0.040) -0.038 (0.035) -0.067 (0.040) -0.081 (0.042) 0.475 (0.198) 0.082 (0.030) -0.031 (0.024) 0.001 (0.000) 10.856 (0.385)

* ** ** ***

RF_Rank Intercept

Nationality Fixed Effects Graduation Year Fixed Effects N Adjusted R2 Model Type

11.587 *** (0.313)

11.116 *** (0.391)

logSalary

logSalary

logSalary

-0.006 (0.034)

-0.001 (0.033) 0.000 (0.041) -0.038 (0.037) -0.077 (0.042) -0.088 (0.045) 0.474 (0.203) 0.082 (0.030) -0.028 (0.024)

-0.017 (0.032) 0.003 (0.040) -0.036 (0.035) -0.066 (0.040) -0.080 (0.042) 0.472 (0.196) 0.082 (0.030) -0.031 (0.024) 0.001 (0.000) 10.813 (0.260)

* * ** ***

* ** ** ***

*** ***

11.513 *** (0.013)

10.865 *** (0.267)

NO NO 282

YES YES 282

YES YES 282

NO NO 282

YES YES 282

YES YES 282

-0.0034 OLS

0.0744 OLS

0.1364 OLS

-0.0034 OLS

0.0725 OLS

0.1374 OLS

38

* ** ***

*** ***

Panel C: Association between Vocal Pitch and Bonus Compensation Dependent Variable: Bonus Bonus logBonus logMeanF0 -0.992 * -0.577 -3.018 * (0.538) (0.574) (1.747) Deep AgeQ1 AgeQ2 AgeQ3 AgeQ4 logMBA_gpa Grad_School Fin_Conc

0.052 (0.328) -0.114 (0.296) 0.311 (0.353) -0.01 (0.332) -0.193 (1.657) 0.011 (0.259) -0.501 ** (0.200)

RF_Rank Intercept

Nationality Fixed Effects Graduation Year Fixed Effects N Pseudo R2 Model Type

5.711 * (3.296)

0.136 (0.362) -0.102 (0.331) 0.531 (0.398) 0.091 (0.363) -0.087 (1.831) 0.004 (0.285) -0.71 *** (0.217) 0.018 *** -0.004 3.067 (3.610)

0.281 (0.869) -0.304 (0.793) 0.816 (0.814) -0.063 (0.851) -1.637 (4.422) 0.131 (0.607) -1.189 * (0.632)

23.528 ** (10.317)

logBonus -1.689 (1.623)

0.360 (0.809) -0.221 (0.747) 1.121 (0.794) 0.218 (0.797) -1.838 (4.070) 0.162 (0.581) -1.300 ** (0.587) 0.038 *** (0.006) 15.844 (9.745)

Bonus

0.147 (0.250) 0.058 (0.330) -0.050 (0.300) 0.316 (0.358) 0.059 (0.330) -0.365 (1.656) 0.050 (0.260) -0.519 *** (0.199)

1.212 (2.156)

Bonus

-0.076 (0.273) 0.153 (0.361) -0.036 (0.326) 0.56 (0.403) 0.157 (0.352) -0.257 (1.868) 0.043 (0.286) -0.719 *** (0.219) 0.018 *** (0.003) 0.572 (2.355)

logBonus

0.268 (0.638) 0.277 (0.888) -0.166 (0.820) 0.852 (0.838) 0.078 (0.855) -2.058 (4.452) 0.182 (0.617) -1.232 * (0.645)

9.769 * (5.908)

logBonus

-0.220 (0.593) 0.358 (0.815) -0.111 (0.760) 1.165 (0.807) 0.323 (0.794) -2.089 (4.109) 0.185 (0.586) -1.317 ** (0.596) 0.040 *** (0.006) 8.250 (5.485)

YES YES 282

YES YES 282

YES YES 282

YES YES 282

YES YES 282

YES YES 282

YES YES 282

YES YES 282

0.1040 Probit

0.2493 Probit

0.0193 Tobit

0.0465 Tobit

0.0934 Probit

0.2462 Probit

0.0169 Tobit

0.0457 Tobit

39

Table 8: Vocal Pitch and Cognitive Ability Proxies for Male MBA Student Sample This table estimates OLS or probit regressions associating voice pitch with three proxies for cognitive ability: MBA grade point average, whether the student entered MBA school with an advanced degree and undergraduate grade point average. LogMeanF0 is the natural logarithm of vocal fundamental frequency in Hz, measured with Praat from MBA student speech samples. AgeQ1, AgeQ2, AgeQ3, and AgeQ4 indicate age quintiles of the MBA student at the time of the audio recording. MBA students in the highest age quintile are captured in the intercept. Nationality fixed effects indicate whether the student is from the United States/Canada/Mexico, South America or Asia. Graduation year fixed effects are indicator variables for calendar year 2009 and 2010. Robust standard errors are presented in parenthesis below the coefficients. *, **, *** represent statistical significance at better than the 0.10, 0.05, and 0.01 levels, respectively.

Dependent Variable: logMeanF0 AgeQ1 AgeQ2 AgeQ3 AgeQ4 Intercept

Nationality Fixed Effects Graduation Year Fixed Effects N Adjusted or Pseudo R2 Model Type

logMBA_gpa 0.020 (0.021) -0.009 (0.011) 0.007 (0.011) 0.007 (0.013) -0.003 (0.013) 1.173 *** (0.099)

Grad_School -0.359 (0.542) -1.098 *** (0.299) -1.265 *** (0.282) -0.452 (0.288) -0.238 (0.277) 1.293 (2.562)

logUNDER_gpa 0.048 (0.039) 0.009 (0.035) 0.025 (0.028) -0.010 (0.029) -0.019 (0.034) 0.839 *** (0.248)

YES YES 282

YES YES 282

YES YES 188

0.0511 OLS

0.1505 Probit

0.0851 OLS

40

Suggest Documents