Do Angel Investors Prefer Dominant Entrepreneurs?

Do Angel Investors Prefer Dominant Entrepreneurs?

Théodore P. Wanner ^a, *, supervised by Dr. S. Murtinu ^a , co-assessed by Dr. M.J. Brand ^a

a

Faculty of Economics and Business, University of Groningen, The Netherlands

THESIS INFO ABSTRACT

JEL classifications:

This Thesis aims to ascertain to what extent nonverbal behavior

(NVB) affects the investment decision of angel investors. To ef- fectively receive funding, an entrepreneur must, especially in the seed and early stage, present not only her business idea but also the quality of the entrepreneurial team. Yet as the team-quality is not directly observable, it is hard to communicate it to investors.

To find out more about how to do so, I analyze the dominance of the NVB of 374 entrepreneurs who participated in the British TV-show Dragon’s Den. I show that more dominant NVB in- creases the likelihood of receiving an investment offer as well as the number of investment offers received. Furthermore, different sources of dominant NVB interact in a way that the importance of specific NVB signals depends on each other.

January 22, 2018 D83

G11 G24 L26 M13

Keywords:

Nonverbal Behavior Startup

Angel Investment Dominance

Word count: 11,990 (excluding references and appendixes).

The author is especially grateful to Samuele Murtinu, whose guidance and input in many fruitful dis- cussions have been exceptionally helpful.

* Corresponding author of: MSc. Small Business & Entrepreneurship, student number s2927020, Nettel-

bosje 2, 9747 AE Groningen, NL. E-mail address: t.wanner@student.rug.nl (T. Wanner).

1. Introduction

Angel investment is considered by both academics and practitioners a suiting source of finance in the early stage of an entrepreneurial firm (Feeney et al., 1999; Mason and Harrison, 1995). According to the Euro- pean Business Angels Network (EBAN, 2017), in 2016, 67% of the early stage investments came from angel investors, while only 26% came from venture capitalists (VCs, henceforth), and the residual 7% from equity crowdfunding. Furthermore, a total of 9.9 billion Euros has been invested in new ventures in 2016 in Europe alone. A large body of literature assessed the screening process of angel investors and the role of the entrepreneurial team in the new venture evaluation (Maxwell et al., 2011). It was found that, espe- cially in the early stages of a firm, the entrepreneurial team plays an important role in the decision-making process of investors (Bernstein et al., 2017; Rajan, 2012). Young ventures have less standardized operations than more established ones, and most of their value lies in the execution of a yet untested business plan.

Only after the first investment round go firms through a standardization process, in which the operations and processes become more generic and the uniqueness of the founding team diminishes, therefore mak- ing it more replaceable (Rajan, 2012). Before this transformation process, the founding team is the core value of the enterprise. The prospects of those new ventures are highly uncertain, and they lack tangible assets that could be used as collateral and are therefore highly risky to finance (Hall and Lerner, 2010).

Given those difficulties, the question about how investors choose businesses to fund does not have an easy answer. Despite often being debated (e.g., Gompers and Lerner, 2001), more in-depth analysis of the factors influencing the investor’s decision to finance a specific venture is scarce. On the mission to eluci- date these factors, prior research analyzed the decision-making process and the decision-making criteria of investors by interviewing them. It has been shown that investors put a value on the entrepreneurial team especially in the early stage, looking for signs of perceived competence, power, likability and trust (Macmillan et al., 1985; Montano et al., 2017). Though the question about how investors recognize the mentioned characteristics remains unanswered and the final decision often results from “gut-feeling” (Kahn, 1987).

Assessing the intuition-based and tacit decision criteria is challenging, and it can often not be observed directly. Yet additional knowledge would help entrepreneurs to prepare for their presentations more ef- fectively.

In this Thesis, I focus on a highly unambiguous construct, namely dominance. Dominant individuals are perceived as more successful than non-dominant individuals, and they often emerge as leaders in groups (Lord et al., 1986). One of the most definite and best-researched ways of assessing the dominance of individuals is nonverbal behavior (NVB, henceforth) (Hall et al., 2005). Scholars have focused on three main sub-constructs of dominant NVB: 1) Paralanguage, 2) speech features, and 3) body language. Para- language is about the physical measurement of the voice. Speech features are about the way an individual speaks, such as the speed, or the complexity of the words used. Body language is probably the most de- veloped NVB. It focuses on (micro-)movements and expressions of the body.

The fact that angel investors (AI, henceforth) primarily focus on the entrepreneurial team, and that the

assessment of the entrepreneurial team is stated to be intuitive, gives the optimal basis to assess the dom-

inance in NVB as an indication of funding probabilities. Additionally, dominance is a singularly positively

associated cue when it comes to determining the competence or capability of individuals (Lord et al., 1986). This leads me to the following research question:

RQ: Does more dominant NVB of the entrepreneurs lead to more successful business presentations in front of angel investors?

I aim to answer the research question using a uniquely constructed dataset, directly coded from the British TV-series Dragon’s Den. I assess the nonverbal behavior of the presenters and measure the relationship with the probability of receiving an investment offer as well as the number of offers. To measure the dominance of the NVB, I use three widely accepted and used sub-constructs, paralanguage, speech features and body language.

In this Thesis, I contribute to the literature at the crossroads between entrepreneurial finance, psychology, human behavior and human decision processes in several ways. First, I use a unique setting to analyze the NVB of the entrepreneurs in a real-life situation. Second, to the best of my knowledge, this is the first study to examine all the three sub-constructs of NVB together. Analyzing them together is of great im- portance because it allows for the testing of interaction effects that have not been tested before, and potentially add significant value to the literature of i) nonverbal behavior and ii) the investment criteria of early-stage investors. Third, I introduce a more sophisticated and more precise measurement of the speech rate in a new variable (Words per Voiced Frame), as discussed later.

Methodologically, I use binary choice models and count-data models. First, the Probit model is used to determine the relationship between NVB and the probability of getting at least one investment offer, while the log- linear Poisson model is employed in a second step to show the effect on the number of offers received. The results show that entrepreneurs with more dominant paralanguage receive more offers than entrepreneurs with less dominant paralanguage. Furthermore, paralanguage interacts with speech features on the proba- bility to receive an offer as well as the number of offers received. Body language significantly interacts with speech features on both the probability of receiving an offer as well as the number of offers received.

Evidence from additional analyses shows that paralanguage also interacts with body language.

The remainder of this Thesis is structured as follows: Section two gives an overview of the relevant liter- ature on early-stage investments and non-verbal behavior, why dominance makes a good construct, how NVB impacts the perceived dominance and how this eventually might influence the decision making of the angel investors. Section three lays out the methodology and the idea behind it. Section four shows the main analysis as well as additional analyses. Section five discusses and concludes.

2. Literature Review

This section shows the prior literature on AIs, the dominance as a construct, how I measure dominance with NVB, and the resulting hypotheses.

2.1. Angel Investors & Decision Making

Angel investors are wealthy individuals who privately provide risk capital to new and growing businesses to which they have no family connection (Mason and Harrison, 1995). They tend to invest in earlier stages, more frequently and less formally than VCs (Maxwell et al., 2011). They normally have business experi- ence, and next to providing capital, they provide management experience and contacts (EBAN, 2018).

Especially in those earliest stages of the new venture, human assets are of great importance, due to the

high uncertainty and the lack of more measurable non-human assets (Rajan, 2012). This focus on the entrepreneurial team is of great importance and has attracted complex experiments to find out how the investors assess the team and whether it can be viewed as a rational strategy. For example, Bernstein et al.

(2017) conducted a randomized field experiment to identify the most important characteristics of a new venture to the AIs. They use businesses from AngelList and communicate via E-Mail with almost 4500 investors, spanning 21 different start-ups. They provide evidence that the team is not merely a signal of quality, but that investing based on the team is a rational strategy. They confirm the model of Rajan (2012) which predicts that especially in the early stages of a firm, before it goes through a standardization process which makes the founding team replaceable, human capital is of vital importance. This does not mean that AIs ignore market factors, such as the potential growth of the sector, prior revenue and protectability of the product, but that their focus is on the entrepreneurial team. (Hall and Hofer, 1993).

Measuring human capital in a business is difficult, and the investors are dependent on mental shortcuts (i.e., heuristics) and intuition, which make the decision process somewhat opaque (Maxwell et al., 2011).

Heuristics are the decision-rules used to reduce complex judgmental tasks to relatively simple cognitive operations (e.g., reduce complex signals to an overall feeling of a certain personality characteristic) (Holcomb et al., 2009). In highly uncertain and complex situations, heuristics is (unconsciously) used to make judgments that should lead to acceptable outcomes (Busenitz and Barney, 1997). For example, Kahn (1987) analyzed the decision-making process of institutional investors, while taking actuarial models to predict success into account. However, the question how investors differentiate between a successful and not successful entrepreneur was left unanswered, while most of the investors stated that they use “gut- feeling” to decide. Paul et al. (2007) conducted a study in which they interviewed 30 AIs about their in- vestment process, and found that background and character of the entrepreneurs are of great importance.

Feeney et al. (1999) and Mason & Harrison (1996) found that substantial weight of the decision is given to entrepreneur’s abilities and track records, and that balancing risk & reward is key to the investment decision. Feeney et al. (1999) additionally found that personality measures such as openness, honesty, realism and integrity are of utmost importance. Haines et al. (2003) show in their empirical study with Canadian angel investors that a strong work ethic, sound business understanding and realistic notion of the venture valuation are important criteria as well. Maxwell et al. (2011) found in a study with 150 inter- actions between potential investors and entrepreneurs that AIs use an “elimination-by-aspect method”

(Tversky, 1972), and therefore eliminate entrepreneurs who do not fulfill certain requirements of the AIs.

As mentioned above, due to the evaluation difficulty, they use heuristics to trim the evaluation set to a more manageable size to conduct their analyses (Maxwell et al., 2011). Other studies found that confidence of the entrepreneur in the own project is critical as well (Feeney et al., 1999; Paul et al., 2003 Mason et al., 2004).

Therefore, understanding how to communicate the quality of the entrepreneurial team is vital information for the entrepreneurs seeking investment.

2.2. Dominance

Communication between human beings can be studied from different angles, and organized along differ-

ent dimensions (Fiske, 1991). The variety of disciplines not only make the study more promising, but also

more challenging. Despite the many areas and disciplines of research, two major fundamental constructs have been found in human relationships: dominance and affiliation (Tusing and Dillard, 2000). Tusing &

Dillard (2000) found that dominance and affiliation reflect two major and utterly important aspects in human history: how to get ahead (i.e. dominance) and how to get along (i.e. affiliation). In this Thesis, I focus on dominance. Dominance can directly influence the perceived capabilities of the entrepreneurs. It can be conceptualized in different ways, such as social skills (Burgoon and Dunbar, 2000; Byrne, 2001), or as the enactment of certain nonverbal or verbal behaviors (Rosa and Mazur, 1979). Dominance, therefore, is reflected in the communication, verbally and non-verbally.

In prior literature, the term dominance has often been used with inconsistent or no definitions, and often in relation to the term power (Hall et al., 2005). I use the definition of dominance used by Hall et al., (2005, p. 898), who give the following as an exemplar: “dominance can be defined as a personality trait involving the motive to control others, the self-perception of oneself as controlling others, and/or as a behavioral outcome (success in controlling others or their resources)”. Opposing, power is defined relatedly: “(…) the capacity or structurally sanctioned right to control others or their resources (…)”. Power does, according to Hall et al., (2005) not necessarily imply prestige or respect. From the definitions, one can see that dominance is more about the motive of the person, while power is the capability.

As a personality trait, dominance is one of the strongest predictors of influence in face-to-face settings (Anderson and Kilduff, 2009). Despite not being clear why this is the case, the fact seems uncontested. In a meta-analysis, Lord et al. (1986) showed that dominance predicts who emerges as the leader in groups, even more than intelligence and any other personality characteristics. Dominant individuals are consist- ently perceived as more competent than they are, and seem to progress faster in social hierarchies (Anderson and Kilduff, 2009).

A primordial reason why dominance leads to beneficial outcomes goes back to the basic premise of mod- ern evolutionary theory, which predicts that all types of organisms strive to maintain to pass on their gene pools (Mayr, 1985). All living organisms are naturally more attracted to the ones more likely to survive.

Human beings used “heuristics”, mental shortcuts, to quickly assess each other’s dominance, as well as detect potential threats, which might have led to death. Higher dominance led to a higher survival rate.

Despite nowadays (mostly) not being a fight of life and death for human beings anymore, the natural predisposition still exists. Non-verbal cues, such as vocal cues have been useful in decoding potential sources of threat or find the most dominant peer group (Tusing and Dillard, 2000).

Prior literature has used the construct verticality (vertical dimension, V, interchangeably) to conceptualize

the (relative) dominance in social relationships. For example, Hall et al. (2005) used the verticality construct

as a measure of dominance and perceived dominance in their meta-analysis about dominant NVB. It pro-

poses that the dominance in a dyad relationship (such as entrepreneur-investor) can be put on a vertical

axis from low to high and that both individuals take in a position on this axis. In this setting, the vertical

relation is related to dominance, power, status and hierarchy. Hall et al. (2005) state that there are three

major sources of power and dominance: 1) personality, 2) role/rank and 3) social class. Either of them on

its own or multiple ones in any combination define a difference in dominance between two interacting

parties. This means that a person with high rank- or social class-dominance, does not necessarily have high

personality dominance. While the dominance of the role/rank and the dominance of the social class can be assessed by definition, the personality dominance is more challenging to communicate. Therefore, these vertical relations between individuals are often researched with nonverbal behavior (Hall et al., 2005;

Henley, 1977). When it comes to non-verbal modality, it has been shown that for the individuals in a dyad relationship in the lower power position (in this case the entrepreneur has lower role-power), body language matters more than for the individuals in the higher power position (the AI) (Hecht and LaFrance, 1998).

Higher personality dominance of an entrepreneur diminishes the gap between the entrepreneur and the AI in the vertical relationship and the entrepreneurs are perceived as more competent and capable. As shown in the following section, this personal dominance is perceptible from non-verbal communication, such as paralanguage, speech features, and body language.

2.3. Non-Verbal Communication

For long, human interaction has been studied in human and evolutionary history (Apicella et al., 2007), with a focus on attractiveness (Borkowska and Pawlowski, 2011; Eagly et al., 1991; Hill et al., 2017; Puts et al., 2005) and trust (Montano et al., 2017). Only recently, there has been more emerging literature about the connection between nonverbal behavior and possible economic outcomes (e.g., Banai et al., 2017;

Mayew and Venkatachalam, 2012; Hall et al., 2005). The perception of power by subordinates, but also peers and supervisors are critical determinants of managerial success (Aguinis et al., 1994; Ragins and Sundstrom, 2005; Yukl et al., 1993). To assess the abilities of the entrepreneurial team, the AIs might use heuristics to assess the dominance of the team and perceive not only what the entrepreneur is saying but how she is saying it. Individuals unconsciously show high or low dominant NVB without explicitly invoking it (Hall et al., 2005).

I use three main sub-constructs of NVB (i.e. communication channels): Paralanguage, speech features, and body language.

2.3.1. Paralanguage

One of the most lucidly and widely applied paralinguistic measurement in prior literature is the fundamen- tal frequency (F

) (e.g. Apicella et al., 2007; Banai et al., 2017; Gregory and Gallagher, 2002; Klofstad et al., 2012; Puts et al., 2005). F

is the physical measurement deduced from the number of vibrations made by the vocal folds to produce a vocalization from the mere airflow. The vibrations occur when the air pressure in the glottis is high enough to burst the folds, releasing pressure by air streaming out. Once the pressure is low, the folds close, building up new subglottal pressure. It is the speed of this rhythmical repetition that determines the depths of a voice. Fewer cycles equal a lower voice. The perceived pitch (how high or low voice sounds) is its perceptual correlate (Pisoni and Remez, 2005; Tusing and Dillard, 2000).

The average F

is around 110 Hz for men’s and 200 Hz for women’s voices (Levelt, 1989, p. 426). The measurement Hz stands for “Hertz” and is defined as one cycle (vibration) per second. Next to the fun- damental frequency, the standard deviation of the fundamental frequency (F

), measures the variation of the pitch. This deviation is how “monotone” or how “animated” a person speaks (e.g. Banai et al., 2017;

Klofstad et al., 2012; Gregory and Gallagher, 2002). Having started mostly in human behavioral sciences,

the academic analysis of the pitch has been linked to attractiveness and other physical measures (e.g. facial

asymmetry, levels of testosterone etc., (Montano et al., 2017; Klofstad et al., 2012; Hill, et al., 2017; Ekman, et al., 1985; Saxton et al., 2015)). For example, it has been found that men with lower-pitched voices are perceived as more attractive and socially more dominant than men with higher voices (Feinberg et al., 2005; Collins, 2000; Sell et al., 2010; Tigue et al., 2011; Gregory, 1994; Puts, 2006; Wolff and Puts, 2010).

Also for women, lower voices have been found to appear more socially dominant (Borkowska and Pawlowski, 2011; Feinberg et al., 2005). Furthermore, F

has been found to correlate negatively with the levels of testosterone (Dabbs and Mallinger, 1999; Harries et al., 1997). Also, dominant men have higher testosterone levels (also called the dominance hormone (Cuddy, 2012)) than subordinate men, leading to the conclusion that voice is likely a cue to dominance (Mazur, 1998; Swaddle, 2002; Tigue et al., 2012). An- other, smaller stream of literature assessed the pitch in economic settings. For example, Banai et al. (2017) analyzed 51 actual election outcomes and found that vocal characteristics, especially fundamental fre- quency and its variation affect the actual election outcomes, giving low-voiced males an advantage over higher-voiced males. Gregory & Gallagher (2002) found similar results. Tigue et al. (2012) found that lower pitched voices were associated with favorable personality traits in more cases than higher-pitched voices.

Moreover, people preferred to vote for politicians with lower-pitched voices. A study from Mayew (2013) found that the fundamental frequency and its variation of the voices of CEO’s from S&P 500 companies negatively correlate with compensation, tenure length and size of the company they manage. Furthermore, prior literature found that the voice influences perceived trustworthiness, finding that leaders with lower pitched voices are perceived as more trustworthy (Klofstad et al., 2012; Tigue et al., 2011). The prior literature generally agrees that a lower F

(deeper voice) generally leads to a more dominant personality perception, which then leads to more successful economic outcomes. Yet, one study found a negative effect of F

on perceived dominance (Tusing and Dillard, 2000).

Also for F

, somewhat more ambiguous meanings have been attached. While most scholars found that higher variation of fundamental frequency leads to worse economic outcomes (Banai et al., 2017; Mayew et al., 2013), a minority argues that a more “animated” and “warm” voice tone is perceived as more char- ismatic, leading to better economic outcomes (Howell and Frost, 1989; Towler, 2003).

2.3.2. Speech features

Speech features are different from paralanguage in the sense that they are no physical measurement, but a

characteristic of how an individual speaks. For example, speech rate, as commonly measured in words per

time (minutes or seconds), has been positively associated with dominance and competence (Cowan et al.,

1997; Roodenrys and Hulme, 1993), and negatively with benevolence (Smith et al., 1975). Nevertheless,

the effect seems to vary with culture. Hall, Coats & LeBeau (2005) found in a meta-analysis that Korean

listeners anticipate lower speech rate with higher dominance, while they found the opposite for western

countries. Nevertheless, all the investors in the setting of this Thesis are from the UK, and higher speech

rate should be associated with higher dominance. Apple, Street & Krauss (1979) increased and decreased

the speed of a recorded speech, therefore artificially raising and reducing the words per minutes, while not

affecting the distribution in which the words are spoken. They find that slow speakers were judged as less

truthful, less fluent, less persuasive and more passive, while the opposite counts for fast speakers (Apple

et al., 1979).

Another speech feature is the complexity of words. Despite never having been associated with dominance directly, it bears the signs of dominant behavior. Word complexity has mostly been analyzed in connection with the word-length effect, saying that longer words need more time to rehearse, and therefore having a negative impact on the memorability (Cowan et al., 1997; La Pointe and Engle, 1990). This would decrease the impression of the presentation (Cowan et al., 1997), and therefore reduce the influence and lower the position on the vertical axis. I presume that the word-complexity has a negative impact on the impression, and hence on the probability to receive an investment offer and the number of offers.

2.3.3. Body Language

Examples of body language are the body posture, hand and arm movements, or facial expressions, (Aguinis et al., 1998). Body features are signals that entail some sort of information about the current emotional state of the person speaking. Prior literature found that dominance, self-confidence, the ability to convince, power and competence affects the body postures directly (Carney et al., 2010). The body posture further- more affects the self-evaluation (Brinol et al., 2009). Different scholars found that expanding body Pos- tures (power-poses) positively correlate with the level of testosterone and negatively correlate with the level of cortisol in the blood (Huang et al., 2011; Riskind and Gotay, 1982). Carney et al. (2010) found similar results, and more remarkably, could show that it works in both ways, meaning that expanding body pos- tures lead to higher testosterone levels, and people with higher testosterone levels have more expanding body postures. Expansive body postures manifest themselves with widespread limbs, enlarged occupied space by spreading out and straight back with the chest out (Brinol et al., 2009; Carney et al., 2010). A contracting body posture is when the limbs touch the torso and the entrepreneur minimizes the occupied space by collapsing the body inward and the back is curved. This can have substantial impact on the dominance of the entrepreneur. Nevertheless, the concept of power-posing has also received some criti- cism, arguing that heterogeneous outcomes of past literature leaves power-posing alone as a hypothesis, currently lacking empirical support (Simmons and Simonsohn, 2017). Hand movements, like body move- ments, signal dominance if the hands are enlarging the appearance of the body (Cuddy, 2015). The effect of those power-posing hand movements has led to the opinion that expanding body language is a tool to improve performance and even life outcomes (Blodget, 2013; Cuddy, 2012). Hand movements are extract- ing if the individual uses them for gesticulation, while more contracting signals are if the speaker holds them in front of the belly without much movement.

2.4. The Hypotheses

The constructed hypotheses find their foundation in the following reasoning: NVB is an unambiguous and well-researched way of communicating dominance. Dominance is uniquely positively associated with perceived competence and capabilities that lead to superior economic outcomes. Furthermore, the poten- tial of the entrepreneurial team plays a major role in the decision-making process of the AIs. Concluding, dominant NVB inevitably increases the perceived potential of the entrepreneurial team and hence in- creases the probability of receiving an investment offer as well as the number of offers received. I therefore hypothesize the following.

Hypothesis 1: More dominant NVB leads to a higher probability of receiving an investment offer.

Hypothesis 2: More dominant NVB increases the number of offers received.

Despite the fact that the three sub-constructs are all signaling the same personality-dominance as described by Hall et al. (2005), they are different in their communicated nature (paralanguage is acoustic-semiotic

, speech features are acoustic-linguistic

, and body language is visual). Therefore, they might interact in a way in which they reinforce or substitute each other. However, the current theory only explains the indi- vidual effects of the three sub-categories of NVB. This Thesis is the first to measure their potential inter- actions to the best of my knowledge. The following two hypotheses are therefore of explorative nature.

Hypothesis 3: The different communication channels of NVB affect the effect on the probability of receiving an investment offer of one another.

Hypothesis 4: The different communication channels of NVB affect the effect on number of investment offers received of one another.

3. Methodology

This section lays out the data collection, how I operationalize the variables and how I analyze them to get the most profound insights.

3.1. Data Collection

Dragon’s Den is a reality TV show from the British Broadcasting Corporation (BBC, henceforth), in which entrepreneurs with any kind of background have the possibility to present their business ideas to five wealthy British individuals (angel investors, called “Dragons” interchangeably). The presented firms can be both running ventures or mere business ideas. The Dragons then have the possibility to negotiate the terms and offer a business deal. This provides a beneficial setting for my analysis, since I analyze real-life speech features and body language. The AIs and the entrepreneurs encounter each other for the first time in the show, and the AIs do not know the nature of the business ideas presented. The show is recorded in real- time, without any actors. I analyze 374 entrepreneurs, presenting 267 business ideas from the 10

to the 14

season of the TV-show.

The show comes with several rules about the process, which positively affect the validity of my analyses.

The most relevant ones are as follows. 1) The Entrepreneur must not be interrupted by a Dragon in the first three minutes of the pitch (if the pitch takes that long). They state their name, the name of the business (idea), the required investment size and the percentage of equity they are willing to give away. 2) The investment-round is over when all five Dragons declare themselves “out” or the entrepreneur accepts at least one offer. 3) The negotiated investment amount must be equal to or larger than the figure initially required. 4) Dragons may share an investment. 5) The entrepreneur can refuse investment offers. For any further regulations of the show, please consult the BBC webpage (BBC, 2017). The pitches take place in a highly standardized process. Figure 1 shows a scene of a business presentation. After the presentation, the AIs can ask questions and renegotiate the valuation of the company. A typical negotiation takes about

1 Acoustic-semiotic because paralanguage is an acoustic signal for which one does not need words to perceive it.

2 Acoustic-linguistic because speech rate cues are acoustic and depend on the words that are spoken, and how they are spoken.

10 minutes. The businesses can range from mere business ideas up to established ventures. The Dragons may change per season. Over the five seasons I analyze, there are four female and six male investors.

Figure 1: Business presentation by Jacob Thundil, presenting his business "Cocofina", with which he produces a wide range of coconut products. The five seated individuals are the Dragons, the man standing the entrepreneur. Jacob Thundil asks for GBP 75’000 in exchange for 5% of his company. The Dragons negotiate the terms, and Sarah Willingham invests together with Nick Jenkins for 10% equity each.

3.2. Operationalization of the dependent variables

Two dependent variables are used for the analyses. One is the number of offers the entrepreneur receives from the investors after her presentation. The categorical variable !

can therefore take on six different natural numbers, ranging from 0 “no investment offer” to 5 “every Dragon makes an offer”.

The other dependent variable is binary and deduced from the number of offers. It equals 0 if the entre- preneur does not receive an offer and 1 if the Entrepreneur receives one or more offers, and therefore can be stated as follows:

!

= 1 '( !

∈ 1,2,3,4,5

0 '( !

= 0 ⁽¹⁾ Or written out:

!

= 1 = 0ℎ2 230425423264 4272'829 :0 ;2:90 1 <((24

0 = 0ℎ2 230425423264 4272'829 3< <((24 ⁽²⁾

This leaves the probability that an individual i receives an offer from investor j is P[O

=1]=p

, and the

probability that an individual i does not receive an offer P[O

=0]=1-p

. Only under the assumption that

the investors are fully independent of each other, the probability of the binary variable equals the joint

probability of not receiving any offer for the value 0, and 1 minus this probability for the value 1. The

joint probability of not receiving an offer equals the product of the individual probabilities and can be written as follows:

!

= 0 = = =[!

= 0]

A

?BC

= 1 − 5

(3)

It is important to see that since the AIs might have similar investment criteria, the probability of receiving an additional offer might increase with the number of received offers.

3.1. Operationalization of the independent variables

The variables of the three sub-constructs are coded independently and before the dependent variable is coded. If there is more than one entrepreneur presenting, the average of the independent variables of each entrepreneur is taken.

Paralanguage

Paralanguage contains the fundamental frequency F

and its variation F

. The variables are constructed using highly sophisticated computer software to assess the physical characteristics of the voice. There are several kinds of software available. In this Thesis, I use the Dutch computer software PRAAT, as in (e.g., Banai et al., 2017; de Jong and Wempe, 2009; Mayew and Venkatachalam, 2012; Montano et al., 2017).

PRAAT is one of the most complex but powerful software for speech analysis (Owren, 2008). I follow the methodology of Banai et al. (2017), and I adhere strictly to the software’s authors guide from the University of Amsterdam when constructing the variables (Boersma and Weenick, 2017).

I first convert the video file and extract the business presentation. I use the whole presentation, starting from the first word of the entrepreneur and I end with the last word. Exceptions are when 1) the entre- preneur gets interrupted or 2) there is any irrelevant sound source, such as a noisy product presentation. I then have 267 different videos. Because there might be multiple entrepreneurs presenting a business idea, I split the videos if necessary again to have a separate video for each of the entrepreneurs. There are five business ideas presented by three, 97 by two, and 165 by only one entrepreneur. This leads to a total of 374 sequences to analyze. Sticking to Banai et al. (2017) I use at least 5 seconds of pitch per entrepreneur.

I convert the mp4 file into an mp3 file using Movavi video software (Movavi, 2017). Since I collect additional information, I depart from the methodology of Banai et al. (2017) who collect only the F

and F

using a script to extract a random 5-second sample from the audio voice report option. I now follow the meth- odology of Mayew et al. (2012), who analyze the periodicity of the audio file using the autocorrelation method, with a pitch floor of 75 Hz and a ceiling of 600 Hz, as recommended by the authors of PRAAT (Boersma and Weenick, 2017). From this extracted pitch, I excerpt the fundamental frequency and its variation. I also extract the total number of frames as a measurement of the total length, and the total number of voiced frames as the number of frames in the pitch in which the entrepreneur speaks. The mentioned measures are directly observable as calculated by PRAAT (Boersma and Weenick, 2017).

Since there is a significant difference between the fundamental frequency of women’s and men’s voices, I

take the absolute deviation of the gender average as a measure, as done by Borkowska & Pawloski (2011). I

therefore have a negative Hz number for below-average voices and a positive value for above-average voices. Figure 2 shows a pitch sequence of one of the analyzed presentations. The blue areas in the top chart are voiced frames, the white areas are unvoiced frames. The blue line in the bottom chart is the pitch in Hz.

Figure 2: Pitch sequence of the software package PRAAT. The top chart shows the vibrations. The blue area is the spoken frames; the white area is the unspoken. The blue line in the bottom chart shows the Pitch. The pitch ranges between 75 and 600 Hz.

Speech analysis

Speech analysis consists of the speech rate of the entrepreneur and the complexity of the words she uses.

As of the speech rate, prior literature has taken the number of words of an entrepreneur divided by the total time of the pitch (e.g. Apple et al., 1979; Tusing and Dillard, 2000). Taking the number of words per minute does not account for the distribution of the words in the analyzed time frame. A person who speaks very fast but makes long breaks might leave a different impression than a person that speaks slower but without pause, yet they might have the same number of words per minute. Furthermore, entrepreneurs who make deliberate pauses in speaking would have similar values to individuals who forget their lines. I therefore assess the total number of frames of the presentation, as well as the total number of voiced frames, as shown in Figure 2. The measurement will therefore be the number of words per time spoken (Words per Voiced Frame, Words per VF, interchangeably), coming close to the understanding of rate of speech (Tusing and Dillard, 2000). This method of measuring is new to the literature to my knowledge, as words per minute is more commonly applied.

Word complexity has been measured in various ways. E.g. Cowan et al. (1997) measured attributes such

as syllable numbers per time, syllables per word, number of stresses, ease of articulation and vowel type. I

adhere to the measurement of a majority of prior literature and use the average number of syllables per

word e.g. (Cowan et al., 1997; La Pointe and Engle, 1990; Hulme and Tordoff, 1989). This ensures an

objective measurement method, and therefore increases the reliability of the Thesis. Syllables as well as

words are counted for the entire recorded pitch to get an accurate measurement. To reach that, I code the content of the pitch and write it down. I use an automatic syllable counter and word counter. Numbers are written in the way they are spoken, meaning that the words “20% of my business for £ 100’000” are coded as “twenty percent of my business for one hundred thousand pounds”, leading to a total of 15 syllables and 10 words, and therefore a word complexity score of (15/10) = 1.5.

Body Features

Figure 3 shows an example of a high power-pose (left) and a low power-pose (right) (Holland et al., 2017).

I split the measurement of the poses, and use Body Posture (Posture, henceforth) as a first variable and Hand and Arm Movements (Hands, henceforth) as a second variable. The reason for this is that the en- trepreneurs can use the hands for articulation or to demonstrate products and thereby be expanding, while otherwise showing a contractive body posture. The variable is binary, with the value 1 for an expansive and 0 for a contractive body posture. As can be seen from Figure 3, the more dominant pose comes along with wider limbs and straighter back and protruded chest, while the low power pose has either closed limbs or even crossed limbs and chest a curved back (Carney et al., 2010). There are three values for the Hands variable. 1 stands for not showing the hands, meaning they are either in the pockets or behind the back, 2 is showing the hands, but without or with very little movement, and 3 is using the hand for artic- ulation or demonstration. I measure the body language variables in the beginning of the presentation, sticking to the criteria of Holland et al. (2017).

Figure 3: An example of a high-power (left) and a low-power pose (right). Since the entrepreneur stands during the pitch, the one on the right is a more typical low-power pose in this setting. The picture is taken from Holland et al. (2017).

3.2. Control variables

A first control variable is the company valuation in British pounds (GBP, £). It entails the required invest-

ment as well as the offered stake in the company. The natural log is taken from the valuation. Gender is

taken as a second control variable, as there might be differences in the investment behaviors of the AIs

depending on gender (Brooks et al., 2013; Hecht and LaFrance, 1998). Gender is a binary variable with

the value 0 for female entrepreneurs and 1 for male entrepreneurs. Last control variable is the number of presenters. Data on age, wealth, education, ethnicity and marital status are of interest, but not available.

An overview over all the variables is provided in the Appendix A – Descriptive Statistics.

3.3. Statistical Analyses

To answer all four hypotheses, I conduct multiple analyses, using one variable per sub-construct, and one interaction per regression. As there are three sub-constructs, I end up with a first regression without in- teraction effects, three for each of the interaction effects between two of the variables, and a last regression with all the interactions to increase the robustness. In the main analyses, I use the F

for paralanguage, Syllables per Word and Words per VF for the speech analysis and the Posture for body language. In additional analyses section, I also add the F

for paralanguage and the Hand Movements for body language. I do not use multiple variables per sub-construct, as they might act as confounders and cause spurious associations.

As discussed above, there are two outcomes of interest to being able to reject or confirm the hypotheses, namely the probability of receiving investment offers, and the number of investment offers the entrepreneur receives, tested with the Probit model and the Poisson model, respectively. I use the software package STATA 15.

3.4. Probit Model

In regressions with binary dependent variables, the calculated probability p must range in the interval [0,1].

Therefore, a linear regression cannot be used, since the linear estimations can range outside this interval (Hill et al., 2011). The Probit model uses the non-linear, s-shaped cumulative distribution function between the independent and the dependent variable. The Probit model calculates the likelihood to receive an offer from at least one of the investors. The probability is

P !

= 1 F

= P !

> 0 F

= F x

β (4)

While !

is the binary variable of receiving an investment offer, !

the latent variable of the number of offers received, F the cumulative distribution function, K the vector of the coefficients to be estimated, and x

a vector of variables (Hoetker, 2007).

3.5. The Poisson Regression

For the second, discrete dependent variable, !

, representing the number of investment offers received, the Poisson regression is applied. The Poisson regression is a special case of the negative-binomial regres- sion (NBR, henceforth). It assumes that the mean and variance are the same (linear), while the variance in the NBR is a quadratic function of the mean (Hoef and Boveng, 2007). If the variation of the data is greater than the mean, it is called overdispersion. The NBR is more flexible, since it does not assume so. That is, the NB regression takes this overdispersion into account by adding a term q, that is:

L:4':372 '3 0ℎ2 MN 42O4299'<3 = P + 1

R ∗ P

(5)

µ being the expectation and q being the term responsible to account for the overdispersion. In the case of

the Poisson regression, q equals ∞, therefore diminishing the overdispersion part (therefore leading and

assuming equidispersion). However, asymptotically, the Poisson regression is always consistent even though

the true distribution is not Poisson, while the NBR is inconsistent if the true distribution is not negative

binomial. Rejecting the true distribution being a Poisson and applying the NBR is therefore not an option,

and a good way to check for robustness is by applying both models and comparing the results. If they are similar, the findings are reliable. Therefore, the Poisson regression is applied and reported while the NBR is conducted and referenced.

3.6. Validity & Reliability

Validity: The use of a reality TV show for academic research can raise questions concerning the validity of the results, as being on TV might evoke certain changes in behavior. Nevertheless, the consequences of the decisions are real, meaning that the decisions must be equally thought through as in real business situations. Furthermore, prior research which is deemed highly valid has used reality TV before for anal- yses (Maxwell et al., 2011). For example, Gertner (1993) used the game-show Card-Sharks to study indi- vidual risk-taking behaviors. Metrick (1995) uses the television game-show Jeopardy as a natural experiment to analyze behavior under uncertainty. Maxwell et al. (2011) analyze the Canadian version of Dragon’s Den to investigate the decision-making process of AIs. Levitt (2004) analyses data from the television show Weakest Link to distinguish between taste-based and information-based theories of discrimination (gender, race). De Roos and Sarafidis (2006) analyze the Australian version of Deal or No Deal to assess the risk- aversion of the contestants. Furthermore, a behavioral economic study from Post et al. (2008) in which they examine and compare the risky choices of candidates of Deal or No Deal as well as related classroom experiments, and showed that the contestants in TV reflect similar behaviors as in real-life.

Another concern is the pre-selection of the sample of entrepreneurs allowed to present their business in the show. As stated in the official rules of the show, everybody can apply from the age of 18 years on (BBC, 2017). Furthermore, all applicants are assessed by the same criteria, including the strength of the idea, the robustness of the business plan and the projected turnover. Furthermore, screening processes in which the number of applicants is limited to present are common in the investment procedure and therefore does not alter the validity of the Thesis (Tyebjee and Bruno, 1984).

Reliability: The dependent variables are highly objective and unambiguous, and do not impair the relia-

bility. Also the variables for paralanguage and speech features are objective, as they are measured accurately

with a highly sophisticated computer software. For the body language variables, Posture and Hands, I use

inter-rater reliability to increase the reliability of the sub-construct (Aken et al., 2012). A co-student, who

finished her MSc. Degree at the University of Groningen in July 2017, assesses 20% of the data and codes

it. The Cohen’s Kappa is applied as a measurement of reliability. A Kappa is between -1 and 1, while values

above 0 stand for an agreement which is higher than chance and below 0 lower than chance. According

to Altman (1999), who adapted measurement rules from Landis & Koch (1977), the values from 0.21-0.4

are regarded to be fair, values from 0.41-0.6 are moderate, values from 0.61-0.8 are good, and values from

0.81-1 are very good. The agreement for both Posture and Hands are slightly above 0.7 and therefore re-

garded to be good (Hands k=0.718, p<0.001; Posture k=0.709, p<0.001).

4. Analysis

This section shows the descriptive characteristics of the data, the regression outputs and the arrangements to increase the robustness of the analyses.

4.1. Descriptive Statistics

In total, there are 157 of the 267 presentations without an offered investment, 37 with only one offer, 33 with two offers, 19 with three offers, 12 with four offers and 9 with five offers. Table 1 shows the corre- lation of the variables. The Number of Offers obviously strongly correlate with the binary variable Dummy Offer. F

significantly and positively correlates with F

which is in line with prior literature also in the magnitude (r=0.4049, p<0.01) (Banai et al., 2017). Also, the more complex the words of an individual, the fewer words per voice frames a person speaks. Hands correlates positively with Posture. The reason for this could be that individuals with expansive hand movements tend to have expansive postures too.

Correlation Table

Table 1: All data are directly coded from the TV-show Dragon’s Den. The variable Offer is binary with the values [0,1]. F0 stands for the fundamental frequency, F0SD for its standard deviation. The number of observations for all the variables is 267. * p < 0.10, ** p < 0.05,

*** p < 0.01.

Dummy

Offer Number

of Offers F

F

Words

per VF Syllables

per Word Body Posture

Dummy Offer 1

Number of offers 0.8150

1

F

-0.0519 -0.0693 1

F

-0.0287 -0.0619 0.4049

1

Words per VF -0.0265 -0.0316 -0.2389

0.0495 1

Syllables per Word -0.0127 -0.0669 -0.0214 -0.1212

-0.3088

1

Body Posture -0.0499 0.0066 0.0582 0.0385 0.037 0.0175 1

Hand Movement 0.0636 0.0336 -0.0003 0.0562 0.0443 -0.0405 0.4224

Descriptive Statistics I

Table 2: Descriptive statistics of the dependent and independent variables. All data are directly coded from the TV-show Dragon’s Den.

The number of observations for all the variables is 267.

Variable Mean Std. Dev. Min Max

Investment Binary Variable 0..4119 0.4931 0 1

Number of Offers if funded 2.3000 1.2674 0 5

Words per Voiced Frames

0.0548 0.0084 0.0258 0.1045

Syllables per Word 1.5648 0.1010 1.0175 1.9565

Body Posture 0.4413 0.4628 0 1

Hand Movement 2.1572 0.5676 1 3

*words per voiced frames, equals 2.6468 words per second, with a standard deviation of 0.3660, a minimum value of 1.6470 and a max value of 4.1030

Table 2 shows the descriptive statistics of the independent and dependent variables. More than 41% of

the business ideas are receiving at least one offer. The average business idea that gets offers receives 2.3

offers, with a standard deviation of 1.2674. The Pitch record length in sec. is the number of seconds of the

business presentation that is analyzed. Therefore, it does not represent the total length of the presentation.

The average words per voiced frame are 0.0548. This equals 2.6468 words per second, or 158.81 words per minute. The average speaker uses around 1.56 syllables per word. No variable except for company value is statistically different for businesses that received an investment offer and businesses that did not receive an investment offer. For more information on the differences in mean, consult the Appendix A – Descriptive Statistics. Table 3 gives more information about the businesses and ideas. The average invest- ment amount is almost GBP 100’000. The average offered stake is around 15% and the average valuation of the businesses GBP 930’000. The value of the business is determined by the entrepreneurs, and might not represent a realistic valuation.

Descriptive Statistics II

Table 3: Descriptive statistics of the investment offers and the team. All data are directly coded from the TV-show Dragon’s Den. The number of observations for all the variables is 267.

Variable Mean Std. Dev. Min Max

Investment Amount 97'828 150'191 25'000 2'000'000

Stake required 0.1504 0.0821 0.0100 0.5100

Company Value 930'424 1'726'097 100'000 20'000'000

Number of Entrepreneurs 1.4007 0.5279 1 3

Pitch record length in sec. 57.8195 22.2683 10 119

Team heterogeneity 0.1223 0.3270 0 1

Voiced frames in % 0.4840 0.0611 0.3240 0.6398

Table 4 shows the average F

and F

for women and men. The mean is about 213Hz for women and 144Hz for men. This is in both cases higher than the historical mean as described by Levelt (1989), who found that the average for women is about 200Hz and for men around 110Hz. The reason for the higher frequency might be nervousness, which has been found to increase the voice (Apple et al., 1979). The average F

is for men about 53Hz and for women around 56Hz. The difference is tested using a t-test and is statistically not significant, while the difference of the fundamental frequency F

is on a 1%-level.

Gender Difference in Voice

Table 4: The average Fundamental Frequency and its deviation for male- and female-entrepreneurs. The data are directly coded from the TV show Dragon’s Den. The variables are not used in the analysis, since the deviation of the mean is used only. T-tests have been conducted to find the statistical significance of the differences in mean. ^* p < 0.10, ^** p < 0.05, ^*** p < 0.01.

Men Women

Mean Std. Dev. Mean Std. Dev. Difference in Mean

Average F0 144.4074 21.2676 212.6978 28.8635 68.2904***

Average F0SD 53.2685 17.7085 56.0116 14.1318 2.7432

3 Additional t-tests have been conducted to assess the difference in the likelihood to get funded between women and men, but no statistically significant differences were found.

4.1. Regression Results

Table 5 to 8 show the regression results of the Probit and Poisson regression of the different variables.

For each of the regressions, one variable of each category (paralanguage, speech features and body lan- guage) is used. Table 5 & 6 show the Probit & Poisson analysis, respectively, of the first three variables, F

, Words per VF, and Posture. In Table 7 & 8, I switch the variables Words per VF with Syllables per Word and apply the same models as in tables 5 & 6. Due to the non-linearity of the Probit analysis, coef- ficients are difficult to interpret, and marginal effects should be reported

. Yet there is no interpretable marginal effect of interaction effects, as the marginal effect of one variable depends on the level of the other variable (Hoetker, 2007). Hence, I report the coefficients in the regression tables (5 & 6) and plot the marginal effects in graphs, as proposed by Hoetker (2007). I plot the graphs on the 1%, 10%, 50%, 90% and 99% level. All regression-tables have the same structure. Every table contains one paralinguistic cue (F

or F

), one body language cue (Body Posture or Hand Movements), and one speech feature (Words per VF or Syllables per Word). The first columns (of all the regression tables, Probit and Poisson) show the regressions without an interaction term. Columns 2 show the interactions between paralanguage and body language, columns 3 between paralanguage and speech features, columns 4 between body lan- guage and speech features, and column 5 with all the three interactions from columns 2-4 in one regression.

The coefficients are reported with their corresponding standard errors.

Hypothesis 1 (H1) predicts that more dominant NVB leads to a higher probability of getting investment offers. As can be seen from tables 5 and 7, none of the single variables in column 1 is significant, and therefore H1 cannot be confirmed. H2, however, predicts that more dominant NVB leads to a greater number of investment offers. F

is significantly negative in both tables 6 & 8, columns 1. As lower F

signs higher dominance, H2 can be confirmed for paralanguage, while it cannot be confirmed for body language and speech features. H3 predicts that the variables interact with each other, so that they influence each other’s relationship on the probability of receiving an investment offer. As can be seen in table 5, column 2, F

positively interacts with Words per VF. Its associated marginal effect graph, as shown as the top right interaction in figure 4, shows that entrepreneurs who have a low variance in F

, and therefore a more dominant paralanguage, have a significantly higher probability of receiving an investment offer if they speak slowly. The opposite is true for individuals who have a more animated voice tone. Furthermore, as shown in table 5, column 4, the interaction between Posture and Words per VF is also significantly positive. Nonetheless, as shown in the bottom left chart in figure 4, there is no level of Words per VF in which posture would make a significant difference on the probability of receiving an investment offer.

Moreover, column 4 in table 7 shows a significantly negative interaction effect between Posture and Syl- lables per Word, again confirming the interaction between paralinguistic NVB and speech features. As shown in the top-left chart of figure 5, if the Posture is contractive (0), the effect of the word complexity measure is positive, while it is negative if the Posture is expansive (1). Yet there is no level of Syllables per Word in which the probability of receiving an investment offer is significantly different with expanding or contracting Posture. Hence, H3 can be confirmed for the interactions between speech features with both

4 Marginal Effects of the Probit & Poisson of the main analyses without interaction is reported in table 23, Appendix B – Additional Evidence.

paralanguage and body language. A similar picture can be drawn on the number of offers received, as hypothe-

sized in H4. As shown in table 6, column 3, F

positively interacts with the Words per VF. Furthermore,

column 3 in table 8 shows the interaction of F

with the Syllables per Word, confirming the interaction

in table 6 (column 3). Moreover, Posture interacts negatively with the Syllables per Word in column 4 in

table 8. Therefore, H4 can be partly confirmed for the interaction of body language with speech features

and fully for the interaction of paralanguage and speech features.

Table 5: Probit analysis on the probability of receiving an investment offer without and with the interaction effects. F0SD is the standard deviation of the fundamental frequency (F0) of the entrepreneur in the recorded pitch. N is the number of observations. The data are directly coded from the British TV-Show Dragon’s Den. For more information about the variables, please consult appendix A.

Coefficients are reported throughout. Standard errors in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01.

Coefficient Coefficient Coefficient Coefficient Coefficient

(S.E.) (S.E.) (S.E.) (S.E.) (S.E.)

F

-0.0040 -0.0035 -0.1034

-0.0045 -0.0960

(0.0049) (0.0068) (0.0386) (0.0049) (0.0398)

Posture -0.1866 -0.1872 -0.2055 -2.1570

-1.6500

(0.1784) (0.1785) (0.1802) (1.1597) (1.2177)

Words per voiced frame -6.7722 -6.7351 -7.7510 -19.9142 -17.2762

(9.4244) (9.4320) (9.8385) (12.1799) (12.7813)

F

× Posture -0.0010 -0.0069

(0.0105) (0.0112)

F

× Words per VF 1.8153

1.7284

(0.6982) (0.7319)

Posture × Words per VF 35.5932

26.0403

(20.6690) (21.7367)

Log Company Value -0.1973

-0.1968

-0.1970

-0.2176

-0.2068

(0.1040) (0.1042) (0.1038) (0.1050) (0.1051)

Gender 0.1616 0.1615 0.1352 0.2045 0.1667

(0.1980) (0.1980) (0.2000) (0.2010) (0.2031)

Number of Entrepreneurs 0.3444

0.3438

0.3505

0.3635

0.3615

(0.1499) (0.1500) (0.1508) (0.1508) (0.1517)

N 267 267 267 267 267

Table 6: Poisson analysis on the number of investment offers received without and with the interaction effects. F0SD is the standard deviation of the fundamental frequency (F0) of the entrepreneur in the recorded pitch. N is the number of observations. For more information about the variables, please consult appendix A. The data are directly coded from the British TV-Show Dragon’s Den.

Coefficients are reported throughout. Standard errors in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01. ⁵

Coefficient Coefficient Coefficient Coefficient Coefficient

(S.E.) (S.E.) (S.E.) (S.E.) (S.E.)

F

-0.0074

-0.0017 -0.0787

-0.0075

-0.0781

(0.0042) (0.0057) (0.0277) (0.0042) (0.0295)

Posture -0.0299 -0.0620 -0.0130 -0.8382 -0.3014

(0.1452) (0.1473) (0.1451) (0.9227) (0.9553)

Words per VF -5.8196 -5.8432 -2.4247 -11.2766 -4.1145

(7.6167) (7.6916) (7.4407) (9.8376) (9.9134)

F

× Posture -0.0130 -0.0177

(0.0092) (0.0095)

F

× Words per VF 1.2987

1.4347

(0.4963) (0.5370)

Posture × Words per VF 14.7865 4.5878

(16.6467) (17.1391)

Log Company Value -0.0839 -0.0800 -0.0895 -0.0909 -0.082

(0.0838) (0.0847) (0.0828) (0.0845) (0.0845)

Gender 0.0806 0.0840 0.0340 0.0931 0.0329

(0.1626) (0.1628) (0.1632) (0.1631) (0.1654)

Number of Entrepreneurs 0.3232

0.3213

0.3252

0.3284

0.3208

(0.1139) (0.1140) (0.1145) (0.1142) (0.1147)

N 267 267 267 267 267

5 In addition to the Poisson regression, a Negative Binomial regression has been conducted. The results are in line with the results of the Poisson regression, indicating a good fit of

Figure 4: Marginal effect graphs for three interactions as well as Words per Voiced Frames individually. The graph shows the marginal ef- fects of the Probit analyses on the probability of receiving an offer. The x axis shows the percentile of F0SD in the two upper charts and the Words per Voiced Frames in the two bottom charts, and the y axis shows the probability of receiving at least 1 investment offer. The dashed lines are their corresponding confidence intervals.

Table 7: Probit analysis on the probability of receiving an investment offer without and with the interaction effects. F0SD is the standard deviation of the fundamental frequency (F0) of the entrepreneur in the recorded pitch. N is the number of observations. The data are directly coded from the British TV-Show Dragon’s Den. For more information about the variables, please consult appendix A. Coefficients are reported throughout. Standard errors in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01.

Coefficient Coefficient Coefficient Coefficient Coefficient

(S.E.) (S.E.) (S.E.) (S.E.) (S.E.)

F

-0.0039 -0.0035 -0.1014 -0.0051 -0.0763

(0.0049) (0.0068) (0.0811) (0.0050) (0.0861)

Posture -0.1878 -0.1883 -0.1582 5.3431

5.0773

(0.1784) (0.1786) (0.1803) (2.8245) (2.8910)

Syllables per Word 0.4061 0.4002 0.4052 1.7348 1.6148

(0.7861) (0.7900) (0.7800) (1.0659) (1.0550)

F

× Posture -0.0008 -0.0023

(0.0105) (0.0112)

F

× Syllables per Word 0.0615 0.0456

(0.0510) (0.0535)

Posture × Syllables per Word -3.5368

-3.3535

(1.8029) (1.8495)

Log Company Value -0.1962

-0.1957

-0.1965

-0.2079

-0.2046

(0.1040) (0.1042) (0.1050) (0.1040) (0.1048)

Gender 0.1483 0.1483 0.1243 0.1713 0.1499

(0.1972) (0.1972) (0.1983) (0.1986) (0.2000)

Number of Entrepreneurs 0.3534

0.3530

0.3594

0.3694

0.3708

(0.1493) (0.1494) (0.1496) (0.1505) (0.1507)

N 267 267 267 267 267

Table 8: Poisson analysis on the number of investment offers received without and with the interaction effects. F0SD is the standard deviation of the fundamental frequency (F0) of the entrepreneur in the recorded pitch. N is the number of observations. For more information about the variables, please consult appendix A. The data are directly coded from the British TV-Show Dragon’s Den.

Coefficients are reported throughout. Standard errors in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01. ⁶

Coefficient Coefficient Coefficient Coefficient Coefficient

(S.E.) (S.E.) (S.E.) (S.E.) (S.E.)

F

-0.0084

-0.0021 -0.1496

-0.0093

-0.1043

(0.0043) (0.0058) (0.0631) (0.0043) (0.0663)

Posture -0.0209 -0.0548 0.0167 4.9148

4.9775

(0.1451) (0.1471) (0.1466) (2.1681) (2.1773)

Syllables per Word -0.8462 -0.9259 -0.6572 0.3896 0.3399

(0.6250) (0.6224) (0.5980) (0.8532) (0.8099)

F

× Posture -0.0144 -0.0153

(0.0094) (0.0098)

F

× Syllables per Word 0.0896

0.0642

(0.0397) (0.0410)

Posture × Syllables per Word -3.1721

-3.2197

(1.3925) (1.4036)

Log Company Value -0.0779 -0.0721 -0.0784 -0.0845 -0.0801

(0.0835) (0.0845) (0.0850) (0.0827) (0.0850)

Gender 0.0671 0.0711 0.0404 0.077 0.0489

(0.1613) (0.1615) (0.1614) (0.1616) (0.1624)

Number of Entrepreneurs 0.3340

0.3312

0.3427

0.3430

0.3493

(0.1129) (0.1131) (0.1128) (0.1122) (0.1123)

N 267 267 267 267 267

6 In addition to the Poisson regression, a Negative Binomial regression has been conducted. The results are in line with the results of the Poisson regression, indicating a good fit of