Gender bias in IT entrepreneurship: the self-referential role of male overrepresentation in digital businesses

ABSTRACT

Drawing on optimal distinctiveness theory and Jellinek’s concept of the Normative Force of the Factual, we provide an alternative explanation for how the normality of gender imbalance in IT-driven start-ups (i.e., the fact that there are far more men than women) becomes normativity (what everyone expects to see), and eventually an imperative for those desiring to enter the field. We uncover the process used to pressure female founders of IT start-ups into being like the prototypical actor, usually male, and how failing to do so reduces audience support. This process is self-referential and self-enforcing and thus detached from efforts to reduce hurdles and obstacles for female founders. Our article provides recommendations on how to break this self-referential cycle.

KEYWORDS:

• Optimal distinctiveness
• categorising
• gender bias
• female entrepreneurship
• male overrepresentation
• digital business
• IT entrepreneurs

1. Introduction

Information technology (IT) is plagued by a massive and persistent gender imbalance (Gorbacheva et al., 2019; Trauth et al., 2009). While male dominance is not a new phenomenon in IT (Reid et al., 2010; Serenko & Turel, 2021; Von Hellens et al., 2012), many expected that the advent of digital technologies would have a gender-democratising effect on the IT landscape (Pergelova et al., 2019), a promise that has not yet been fulfilled (Ughetto et al., 2020). The underrepresentation of women in the IT sector is still omnipresent, from IT education to IT entrepreneurship and from IT departments to the C-suites of IT giants. Data from Crunchbase indicate that only 17 percent of IT start-up teams have female members (dropping to 8 percent in later stages of the business life cycle; Teare, 2021). Unicorn start-ups founded by women are rare (Pantin, 2021), mirrored by the fact that nearly all IT giants are founded by men (e.g., Google, Apple, SAP, Facebook, Twitter, Uber, Airbnb, and eBay). This gender imbalance in IT entrepreneurship is particularly worrying (Guzman & Kacperczyk, 2019; Ughetto et al., 2020) because new ventures build the bedrock of economic growth. The reality today is that women are not only less likely to become IT entrepreneurs than men, but once they do, they are less likely to outperform their male counterparts (Guzman & Kacperczyk, 2019; Jennings & Brush, 2013; Ruef et al., 2003; Yang & Aldrich, 2014). Consequently, the classic tech-savvy founder who embarks on a journey to revolutionise our lives is male (Ughetto et al., 2020), not only in our empirical reality but also in public awareness (Hamilton, 2014).

While several authors argue that gender imbalance is already worrying, mostly for reasons associated with the downsides of homogeneity (e.g., on creativity and innovation), imbalance, by itself, does not necessarily point to women being disadvantaged (Perez, 2019).¹ Imbalance can be the result of deliberate decisions on the part of women not to enter the field of IT. However, imbalances might also result from obstacles to embarking on such a career, or lack of opportunities, or even hostile work environments (Trauth et al., 2009). In the case of hostile work environments, we must assume an even more worrying gender bias that signals systematic or systemic disadvantages for a particular gender (Perez, 2019).

The main theoretical approaches today tend to address gender biases but not imbalances. They are built around gender essentialism, which is based on the notion of social constructions of gender, or they take a gender-intersectionality perspective (Adam et al., 2006; Gorbacheva et al., 2019). In a nutshell, these approaches focus either on fundamental differences across gender identities based on their biological and psychological attributes or (and more recently) on socially-constructed differences (Marlow & Dy, 2018). It is commonly suggested that disadvantages for female IT entrepreneurs are rooted in their education (Bevelander & Page, 2011), culture and norms (Hechavarría et al., 2018), stereotypes (Gupta et al., 2009), role expectations (Bullough et al., 2022), or work–family conflicts (Shelton, 2006). These conditions differ significantly across gender identities to the disadvantage of women, presenting them with obstacles to the success of their entrepreneurial aspirations in IT-driven environments (Sperber & Linder, 2019). While this line of literature deserves much credit, it suffers from a research perspective that views obstacles in isolation and within a linear cause-and-effect logic. The rationale is that there is a set of core causes preventing women from exercising their full potential, thus blocking them from the success of their male counterparts. This thinking suggests that the solution is straightforward: erase these core causes and gender bias will disappear. Despite the supposed simplicity of this solution, the issue remains unresolved.

What if the current male-dominated IT start-up reality is not just a consequence of a set of core causes? More specifically, we theorise that male dominance contains and enfolds an implicit normative force that can become an imperative transcending the distinction between what we observe and its normative justification. Drawing on optimal distinctiveness theory (Deephouse, 1999; Zhao et al., 2017) and Jellinek’s (1905) concept of the Normative Force of the Factual, we offer an alternative explanation for how entrepreneurship “normality” becomes “normativity” and eventually an imperative for those pursuing an entrepreneurial career. We argue that the strict epistemological separation of “is” and “ought” is less stringent than analytical reasoning proposes and is even more blurred in the normative expectations and assumptions that entrepreneurs face. In public perception, “normality” and “normativity” do not represent fully separated and distinct concepts; they reinforce each other. IT entrepreneurship is male-dominated; hence, it ought to be this way. This perception places pressure on female entrepreneurs and creates a serious obstacle in IT-driven start-up environments. To uncover the mechanism behind how “normality” becomes “normativity”, our research asks, “Why does a male-dominated IT start-up reality become the normative imperative for aspirants of an entrepreneurial career in IT?”

To answer this question, we turn to the literature, which suggests women are much less likely than men to obtain external support from audiences such as investors (Edelman et al., 2021; Marlow & Patton, 2005). Audiences provide both legitimacy and significant resources for the start-up process. From this perspective, IT-driven ventures are a specific category of start-ups. This category happens to be male-dominated (Serenko & Turel, 2021), which leads to the creation of prototypes for the “IT start-up” category that best represent the category’s typical actor (Durand & Paolella, 2013). The reference point for these prototypes is, as we will demonstrate, the factual entrepreneurship reality in which men are vastly overrepresented. By shaping how audiences create prototypes, the overrepresentation of men becomes self-referential. Gender bias is not just a function of the sum of its causes but simultaneously the cause and effect of gender bias. This mechanism establishes normativity and an imperative that requires some degree of similarity to the male-dominated prototype, a demand that female entrepreneurs can hardly meet. Consequently, inequality is caused by a bias and is more than an imbalance. It is systemic not because we are incapable of erasing the causes of male overrepresentation but because there is a self-strengthening and self-referential mechanism detached from the linear logic of a cause-and-effect relationship. Uncovering this mechanism builds the main contribution of our work.

Our study pushes the discourse on gender bias in IT forward by theorising on the self-fulfiling effect of male overrepresentation in IT entrepreneurship (Gorbacheva et al., 2019). Specifically, we provide a missing explanation for why gender bias is more than the sum of obstacles hindering female IT entrepreneurs. Our findings suggest that there are liabilities of distinctiveness for female founders that result from gender differences leading to reduced legitimacy, which, in turn, limits their access to (financial) support. While distinctiveness can be an advantage (Zhao et al., 2017), we demonstrate that being distinct from the prototype in terms of gender is problematic when being male is the normative imperative. For this reason, resolving gender biases requires not only the dismantling of the factual or positive causes of inequality but the breaking of the self-referential cycle. We develop practical recommendations for how to do this.

2. Theory and hypotheses

2.1. The capital allocation process under uncertainty

While the existence of a significant gender-related funding gap in IT entrepreneurship is not in dispute, there is disagreement regarding the underlying mechanisms causing this gap (Brooks et al., 2014; Eddleston et al., 2016; Verheul & Thurik, 2001). Investor-centred explanations suggest that the capital allocation process disproportionately favours male IT entrepreneurs due to uncertainty in investors’ decision processes (Alsos & Ljunggren, 2017; Kanze et al., 2018). Finance decisions are particularly complex, given that the quality of an early start-up is commonly not observable (Edelman et al., 2021). Uncertainty of finance decisions is characterised by the absence of information on the quality of both the start-up’s proposed business model and the technology used. Such uncertainties include feasibility, prototyping, efficiency, implementation of the organisation and the integration of the technology into the business model, the people (including the partnerships), the knowledge requirement and the human capital, as well as business environment issues like standardisation and availability of use-cases (Mattke et al., 2021; Walsh et al., 2020). Technologies with unproven or not (yet) fully verified potential (e.g., artificial intelligence, blockchain, and machine learning) count as the most difficult to assess (Walsh et al., 2020).

To evaluate the quality of IT start-ups in the absence of reliable data and information about future performance, investors rely on heuristics or simple decision rules and surface perceptions of overt characteristics such as gender. Uncertainty pushes investors to bank on past observed attributes that proved successful because they presumably predict unobservable but important determinants of future start-up success (Conti et al., 2013). One observable start-up characteristic, and probably the most important, is the start-up team and its composition (Klotz et al., 2014). This makes gender imbalance not only a current empirical fact but a key element in how audiences evaluate future teams.

2.2. Categorising the start-up team

To assess the quality of start-up teams, audiences depend on putting teams into categories (Negro et al., 2011). Categories are social constructions that serve as a cognitive representation of “a meaningful consensus about some entities’ features as shared by actors grouped together as an audience” (Durand & Paolella, 2013, 1100). Audiences use these categories to classify actors and assess if actors behave consistently with category expectations (DiMaggio, 1987). Categorisation, hence, is the collective process to assess social behaviour, events, and mechanisms in and around organisations (Boulongne & Durand, 2021). Categorisation and the resulting categories are not axiologically neutral, unbiased, or impartial. Instead, they put pressure on those being categorised and frame the perceptions of those doing the categorising.

Specifically, categories are taken-for-granted systems that an entrepreneur experiences as givens (Cattani et al., 2017). Category membership becomes a question of conformity to audiences’ category expectations, a process in which deviation “from existing norms, expectations, and practices” (Zhao et al., 2017, 93) generally leads to penalties. As such, conformity is inevitably linked to being perceived as a legitimate category member (Durand & Paolella, 2013), while an actor receives legitimacy in the eyes of audiences through complying with collective expectations. Categories shape behaviour because legitimacy is a necessary condition for audience support and paramount for entrepreneurial success (Garud et al., 2014).

Significant evidence further suggests that conformity varies across entrepreneurs, ranging from full membership to partial membership in a particular category (Durand & Paolella, 2013). Full membership means that an entrepreneur shares all category-relevant characteristics with the archetype or the prototype of a category (Boulongne & Durand, 2021). Prototypes are the idealised IT entrepreneur – the one every investor desires to support and who serves as the reference point for categorising all available investment opportunities. Being perceived as a legitimate IT-driven start-up, therefore, entails some level of conformity to such prototypes (Boulongne & Durand, 2021; Negro et al., 2010). However, mental representations of categories are graded in a way that implies that, besides full membership, entrepreneurs can have partial membership in a particular category (Rosch & Mervis, 1975). This suggests that category conformity decreases when partial membership increases. Under competition, we can expect that several entrepreneurial teams qualify for full membership. In this case, there might still be a difference in similarity to the prototype that eventually disadvantages those teams that cannot ensure a necessary level of similarity. This reasoning is supported by extended research on category straddling and multiple category membership, a fact that in general reduces the appeal of any category within which the IT entrepreneur is present (Durand & Paolella, 2013). Following this rationale, one can assume that female founders are just as well equipped with skills, knowledge, and capabilities as their male counterparts and that their pursuit of an equally promising business opportunity as their male counterparts will result in an equal level of conformity. In fact, female founders remain dissimilar to the prototype if this prototype is male.

Audiences such as analysts, critics, brokers, or venture capitalists (VCs) have a stabilising effect on the category boundaries. VCs, for instance, operate as gatekeepers allowing or denying access through their support (Ruef & Patterson, 2009). This suggests that if a male founder team is the prototypical founder team within a category, audiences such as VCs stabilise the category by validating new founder teams according to the properties of the existing prototype. They deny access to the category if they believe a team’s properties drift too far from the category prototype.

In what follows, we argue that the core business of VCs is the assessment and potential financing of uncertain business models. Under uncertainty, the question of whether the founder team is capable of handling the involved risks becomes decisive. Therefore, VCs assess potential teams as a proxy for future success in reference to the archetypical founder team within the respective category. Derivation or drift from this prototype causes a decrease in audience appeal and the eventual loss of VC support. The question that remains is whether being a female founder is already too dissimilar to the prototype. Figure 1 depicts this theorising.

Figure 1. Audiences, business model uncertainty, and drift from the category prototype.

2.2.1. Stereotypes and category belonging

Prior literature suggests that female founders face pressure to resemble prototypical characteristics. Investors see entrepreneurship as a masculine endeavour, and women simply do not fit the basic expectations of how an ideal founder is, behaves, and acts (Balachandra et al., 2019; Kanze et al., 2018). The literature on women in IT comes to the same conclusion but on different grounds. The typical tech-savvy computer scientist is male, independent, socially awkward, unemotional, and rational, with a soft spot for cutting-edge technologies (Berg et al., 2018; Reid et al., 2010). An IT-savvy founder, hence, typically comes from a male-dominated industry and enters – as Balachandra et al. (2019, 117) term it – another “man’s world” through his entrepreneurship. The same journey looks different for women, not because they are less capable of doing the job but because they are entering a world that is not made for them. They are neither archetypical nor prototypical in the category of digital technologies; hence, they are a deviation from audiences’ expectations.

Some authors argue that gender role expectations are the source of what is considered adequate gender-specific behaviour (Eagly & Wood, 2011). This line of reasoning suggests that legitimacy in male-dominated fields requires correspondingly stereotypical gender characteristics and behaviour (Malmström et al., 2017). For obvious reasons, in an industry with significant male overrepresentation, these characteristics are more easily displayed by male founders than by female founders (Eddleston et al., 2016). Moreover, while behaviour can be adjusted – and there is significant literature suggesting that successful women do adjust their behaviour (Carli, 2010) – biological or physical appearance is much harder to adjust. In this regard, Brooks et al. (2014) demonstrated that investors are more likely to support business ideas pitched by male founders. A significant line of literature argues further that women are associated with less aggressive growth strategies, risk aversion, and consequently with less capital-intensive businesses (Orser et al., 2006). Thus, female founders or female teams deviate from the male-dominated prototype, which is commonly seen as more aggressive, growth-oriented, and risk-taking (Malmström et al., 2017). Kanze et al. (2018) argue further that gender is subtle as it hides in speech patterns, nonverbal gestures, displayed social competence, or attractiveness. Distinctiveness from the male prototype is thus deeply embedded in how audiences perceive and interpret female teams.

We theorise that if audiences perceive the (male) prototypical founder as possessing all the stereotypical characteristics of being successful, women are consequently perceived as incapable of successfully undertaking IT business venturing (Jennings & Brush, 2013). The existing normality in IT entrepreneurship is the source of stereotypical ascriptions of IT entrepreneurs (Gupta et al., 2009), which, in turn, affects the construction of the prototypical actor. Thus, female entrepreneurs do not receive less funding because they are female but because, from the investors’ perspective, they cannot demonstrate the characteristics that justify the allocation of financial means (Alsos & Ljunggren, 2017; Marlow & Patton, 2005). We call this pressure for women to conform to a certain prototype while being unable to reach it the “gender effect”. Building on our earlier argument, this effect for female entrepreneurs in IT is even higher if presenting start-ups with the uncertainty of business models and technology. Uncertainty facilitates prototype conformity because no other information is available that substitutes for categorisation. In other words, audiences’ subjective framing of category expectations becomes more relevant under uncertain circumstances. Thus, we argue that greater uncertainty even increases the gender effect. Our first hypothesis states:

Hypothesis 1: The greater the uncertainties within a business model, the more significant the gender effect is.

2.2.2. Creating and maintaining similarity

While our first hypothesis assesses the roots of the gender effect, our second hypothesis focuses on the persistence of this effect. Every financial support decision is the result of a complex process that proceeds through a sequence of selection and decision rounds. Because similarity, once proven, tends to persist over subsequent decision rounds, we theorise that if a team has been perceived as similar enough to the prototype, audiences will focus on the level of distinctiveness to alternative category members with equal levels of similarity to the prototype (Zuckerman et al., 2003). The literature suggests that legitimacy ascriptions are relatively stable over time (Elsbach & Sutton, 1992). This fact has a significant implication for our reasoning; there is a strong distinction between creating and maintaining similarity. In other words, maintaining the perception that a particular female founder’s team actions and behaviour are as desirable, proper, and appropriate as those of her male counterpart (Suchman, 1995) is distinct from developing and building this perception (DiMaggio & Powell, 1983; Elsbach & Sutton, 1992). This notion is the other side of the “categorization coin” – without it, the phenomenon of female IT start-up teams would be incomplete and insufficiently explained.

Specifically, the quality of persistence over time provides an opportunity for female teams. When audiences have validated a sufficient level of similarity, this perception will persist and give space to female founders to present their distinctiveness. Several studies have provided evidence for this persistence over time and the stability of such audience perceptions. For example, legitimacy based on expertise is long-lasting because expert identities, though hard to build, are relatively stable (Croidieu & Kim, 2017). In other words, once an expert, always an expert. This feature of certain audience ascriptions allows individuals, for example, to exploit their expertise in media, TV, and other channels for a long time after building their status as an expert. The systematic difference between building and maintaining similarity and the temporal duality of acquiring and maintaining a perception of prototype similarity means that if female founders have received audience validation in a first support round (i.e., are perceived as having sufficient similarity to the prototype), audiences will support them in subsequent support rounds. In other words, the gender effect, once established for the first time, disappears in the following support rounds. Our second hypothesis thus reads as follows:

Hypothesis 2: The gender effect in the first support round disappears in subsequent support rounds.

3. Analysis

3.1. Sampling strategy

3.1.1. Sample selection

Testing our theorising required a data sample that fulfilled three requirements. First, we needed to identify a business model around technology with significant uncertainties because business-model uncertainty, we argue, is one driver for the relative importance of founders and founder teams. Second, to anticipate a male-dominated prototypical actor, there had to be massive male dominance within the start-up scene of this technology. Third, we required an audience that legitimates entrepreneurial action based on a decision process that potentially references category expectations in the absence of actual information about future success.

We identified the most relevant audience that operates under significant uncertainty as being investors who focus on blockchain technologies. Blockchain technologies have severely disrupted many industries (Grover et al., 2019), enabling new digital business models with enormous potential but high uncertainty regarding future success (Ying et al., 2018). Finally, the data indicate that blockchain and the surrounding start-up environment of this technology are indeed massively male-dominated with a minimal female presence not only in real businesses but also in the media and public debate (Elsbach & Stigliani, 2019).

We used three data sources to build our dataset. First, we used Crunchbase to identify start-ups built around blockchain technologies. Crunchbase, an open-source database maintained by TechCrunch, contains information about start-ups, investors, founders, trends, milestones, and other related information, including the business or revenue model of these ventures. Our data are not longitudinal but provide a relation between founder characteristics and finance decisions over time. We used data on founders and founder teams as well as investors, investment rounds, and capital raised. We examined the founders’ profiles to collect data on gender, education, prior experience, etc. (Liang & Yuan, 2016). We also used company websites and public company data to extract information on the firm, its product, and its business model (see, Table 1).

Table 1. Data structure.

Data source	Variable	Meaning	Measurement purpose	Coding
CB	Dependent Variable (model 1)	Financial support (amount)	Measures support expressed as USD amount	Numeric in USD
CB	Dependent Variable (model 2)	Support rounds	Measures whether investors ignore the “gender effect” in subsequent support rounds	Support round 1 = 0; successive rounds 2 to n = 1.
CB or other published media	Moderator	Business model uncertainty	Measures the level of uncertainty within a business model.	1 = “transaction fees”; 2 = “professional services fees and service level fees”; 3 = “cryptocurrency speculation”
CB	Mediator	Audiences	Numeric value	Measures one or more support received from either business angels, corporate venture capitalists, venture capitalists.
CB/CW	Independent Variable	NVT Gender distribution	Measures the male/female dominance within a start-up team	(Simpson, 1949) Index for diversity normalised between 0 and 1.
CB/CW	Control (NVT)	Team size	Number of team members	Numeric value (between 2 and 7)
CB/CW	Control (NVT)	Education (level)	All members having a bachelor’s degree or above	Binary measure, yes = 1; no = 0
CB/CW	Control (NVT)	Education (heterogeneity)	Degree of heterogeneity in educational background	(Blau, 1977) Index of Heterogeneity
CB/CW	Control (NVT)	Industry experience (years)	Average years of industry experience	Numeric value
CB/CW	Control (NVT)	Industry experience (heterogeneity)	Experiential heterogeneity	(Blau, 1977) Index of Heterogeneity
CB/CW	Control (NVT)	Managerial experience	Average years of managerial experience	Numeric value
CB/CW	Control (NVT)	Start-up experiences	Average number of founded businesses	Numeric value
CB/CW	Control (company)	Growth stage	Measures the actual growth	1 = “testing” (no revenues); 2 = “launching” (some revenues, not exceeding costs)’; 3 = “scaling” (generate revenues)
	Control (company)	Age	Average company age	Numeric value
CB/CW	Control (company)	Location	Location in entrepreneurship hot spot or in the periphery	Binary measure, yes = 1; no = 0
CB/CW other publish media	Control (company)	Investor-Company proximity	Business is headquartered in the same area as the investor	Binary measure, yes = 1; no = 0
CB/CW	Control (technology)	Blockchain type	Categorical measure for the type of blockchain	Private blockchain = 1, public blockchain = 2, consortium blockchain = 3
CB/CW	Control (technology)	Consumer focus	Measures the type of consumer served	1 = “B2B”; 2 = “B2B and B2C”; 3 = “B2C”
CB/CW	Control (service)	Security	Is there a product/service focus on …	Binary measure, yes = 1; no = 0
CB/CW	Control (service)	Privacy	Is there a product/service focus on …	Binary measure, yes = 1; no = 0
CB/CW	Control (service)	Compliance	Is there a product/service focus on …	Binary measure, yes = 1; no = 0
CB/CW	Control (service)	Scalability	Is there a product/service focus on …	Binary measure, yes = 1; no = 0
CB/CW	Control (service)	Customisation	Is there a product/service focus on …	Binary measure, yes = 1; no = 0
CB/CW	Control (service)	Liquidity	Is there a product/service focus on …	Binary measure, yes = 1; no = 0

Notes: * Published company data includes company website but also further documents such as investor presentation or media coverage. CB = Crunchbase, CW = company website.

3.1.2. Sample characteristics

We ensured a representative sample that balanced male and female teams as much as possible. We identified a total of 107 start-up teams but no solo founders. Fifteen of the teams were all-male. As we could not identify an all-female team, we ensured that the remaining 92 teams represented different degrees of gender distribution. Teams ranged from two to seven members. We found no significant differences in education among members, nor in industry and managerial experience, but there was slightly higher start-up experience in teams with a high number of male founders. We further excluded investors who focused exclusively on female financing. Means, standard deviations, and bivariate correlations for the principal variables and controls are presented in Table 2. The correlation matrix suggests that there are indeed significant associations in the hypothesised direction between the model variables.

Table 2.: Correlation Matrix.

	Mean	SD	1	2	3	4	5	6	7	8
	9	10	11	12	13	14	15	16	17	18	19	20	21	22
Financial Support	1.253	7.010	1.000
Support Rounds	0.234	0.428	0.114**	1.000
Business model uncertainty	1.284	0.738	-0.075*	0.033	1.000
Audiences	1.290	0.761	0.092*	0.231*	0.148	1.000
Gender distribution	0.287	0.357	-0.428	0.011	0.149	-0.126*	1.000
Team size	3.263	1.440	-0.141	0.348*	0.074	0.043	0.193*	1.000
Education (level)	0.061	0.239	0.063*	0.164	0.061	0.070	0.012	0.109	1.000
Education (heterogeneity)	0.052	0.221	0.053*	0.164	0.062*	0.061*	0.026	0.121*	0.446	1.000
Industry experience (years)	0.114	0.318	0.137	0.297	0.016*	0.145*	-0.069	0.109	0.244	0.271
Industry experience (heterogeneity)	0.024	0.154	-0.054	0.061	0.054	-0.237	0.187*	0.080*	0.065*	0.070
Managerial experience	0.036	0.187	0.075**	-0.006	0.090*	-0.134	0.024*	0.143	0.106	0.082*
Start-up experiences	0.226	0.418	-0.188*	0.008	0.095*	-0.156	0.358*	0.054	0.059*	0.049
Growth stage	1.144	1.087	0.066	0.143*	0.023	0.031*	-0.036	0.047	0.042	0.054
Age	4.367	3.682	-0.205	0.072	0.107*	0.017*	0.194	0.275*	0.037*	0.032
Location	0.868	0.997	-0.117	-0.049	0.066	-0.145	0.102	0.018	-0.045	-0.035
Investor-Company proximity	0.280	0.449	-0.161	-0.058	0.035*	-0.258	0.278	0.053*	0.006	0.002
Blockchain type	0.076	0.265	-0.071	0.164*	0.033*	0.048**	0.103	0.296	0.052*	0.062*
Consumer focus	0.215	0.745	-0.005	0.010*	0.007	0.060	0.016*	0.000	-0.001	0.000
Security	0.020	0.139	0.021*	0.128	0.008	0.051	0.015	0.055	0.218	0.274
Privacy	0.257	0.114	-0.099*	0.119	0.137**	0.027*	0.151	0.055	0.107	0.109*
Compliance	0.017	0.129	-0.043*	0.030*	0.045*	-0.193	0.149**	0.068	0.051	0.060
Scalability	0.023	0.149	0.062	-0.005	0.082*	-0.115	0.008	0.116*	0.066*	0.064

Notes: ** Correlation is significant at the 0.01 level (2-tailed); * Correlation is significant at the 0.05 level (2-tailed).

3.2. Measures and quality criteria

3.2.1. Dependent variables

We used two dependent variables: one for model 1 (financial support) and one for model 2 (support rounds). The rationale behind the choice of these variables is that investors assist in the formation of IT start-ups by offering support that can be classified in terms of scale (i.e., how much they invest in a particular business) measured in USD (Tian, 2011) and duration (i.e., how long they support a venture) measured in support rounds (Cong et al., 2021). The first variable, financial support, showed a mean of $1.25 million with a high standard deviation of 7.01. The second variable, support rounds, is coded binarily (first round = 1; any subsequent support rounds = 2). In our sample, 38% of all ventures received funds in more than one round. We conducted two analyses: an OLS regression for funds received and a logistic regression for financing rounds.

3.2.2. Independent variables

Business model uncertainty is measured on three levels. The first and least uncertain business model is based on transaction fees, representing the most conservative revenue stream (51% of our sample). The second type of business model combines professional service fees and professional service agreements (27%). It is more difficult to predict future revenues for these models than for transaction fees or subscription fees, thus we assume greater uncertainty than in the first category. Finally, the third type contains start-ups built around cryptocurrency speculations (Canh et al., 2019) – i.e., ventures that bet on the price increases of issued cryptocurrency (22%). Since greater uncertainty may lead to higher returns, we acknowledge that all three types of business models can be interesting investment objects. Moreover, there are specialised investors with particular investment scope on relatively uncertain and speculative investment objects. This operationalisation is consistent with our research interests because we are not interested in making statements about which business model received more support or is favourable but rather whether the support changes once a certain gender distribution is given.

3.2.3. Mediator and moderator

Our mediator evaluates whether audiences such as business angels, VCs, and corporate VCs potentially supported start-ups with uncertain business models (measured numerically). Our rationale is that support is provided relative to the business appeal for each audience and that this appeal decreases with increasing gender distribution. High values indicate that a start-up has received support from multiple audiences within one audience type or across multiple audience types, or both. Thus, we used audience support as a proxy for positive validation of the founder team (Chliova et al., 2020; Navis & Glynn, 2010). Increasing values, therefore, indicate that a founder team in the eyes of the audience has sufficient similarity to the category prototype and is seen as a legitimate investment object.

Our moderator is gender distribution operationalised as the index value between 0 and 1, where 0 represents an all-female team while 1 indicates that a team consists only of men. The reasoning behind this operationalisation is that if the prototype is male-dominated, the degree to which a new venture team is dominated by women is an indication of drift from this prototype. Gender distribution thus is operationalised as a proxy for drifting away from a male prototype. We used Simpson’s (1949) index for diversity to measure the degree of concentration by assigning an index value of 1 to teams with no gender heterogeneity (0 otherwise) and coded the degree of concentration in mixed teams in terms of proximity to either all-male or all-female teams (mean = 0.287, sd = 0.357).

3.2.4. Controls

We chose control variables that captured new venture success to assess if these controls explain audience support better than our variables of interest do. The literature on new venture teams suggests that factors such as team size, education level, education heterogeneity, industry experience, experiential heterogeneity, managerial experience, and start-up experience play a role in investment decisions (Colombo & Grilli, 2005; Unger et al., 2011). We further assessed firm-level characteristics that are known to affect investors’ decision-making such as growth stage, age, location, and proximity to the financier. We also assumed that blockchain technology might influence investment decisions and be controlled accordingly (Niranjanamurthy et al., 2019). Finally, we considered service aspects of the blockchain such as consumer focus (B2B vs. B2C), security, privacy, compliance, scalability, customisation, and liquidity (see, Table 1).

3.3. Hypotheses testing

We evaluated the boundary conditions under which the drift from the prototype, or what we call the gender effect, leads to penalties in terms of audience support. To test our hypotheses, we applied Conditional Process Analysis (CPA; Hayes, 2013) using standardised Ordinary Least Squares (OLS) regression as well as logistic regression. Specifically, we used Model 14 from the PROCESS template to test a first-stage moderated mediation model, which allows the effects from the independent variable on the mediator to be moderated (Hayes, 2013). Such models require estimating coefficients in two regression equations, whereby the mediation M analysis focuses on estimating the indirect effect of X on Y through the intermediary moderator variable W. CPA is suitable to test these relations as it is a general modelling strategy that attempts to describe the conditional nature of mechanisms (such as audience support and drift from the prototype) and estimates how these mechanisms transmit their effects to other variables (i.e., audience support; Hayes & Rockwood, 2020).

According to our first hypothesis, audiences more heavily penalise derivation from the male-dominated prototype in IT start-ups with more uncertain business models. Table 3 presents the results from the CPA. Using the PROCESS macro in the first multiple regressions, we tested whether business model uncertainty affects audiences’ decision-making. We identified a small but significant positive effect (β = 0.073, p < 0.05, CI: 0.009, 0.138), indicating that uncertainty appeals to some audiences. In terms of support, however, our results show a significant link between the mediated path from the audience to the amount of finance provided (β = 0.125, p < 0.01, CI: 0.101, 0.168). Thus, some audiences value uncertainty, probably because it is linked to possible higher returns. No significant direct effect, however, was detected (β = −0.005, ns, CI: −0.032, 0.022), meaning that the amount of finance provided can only be explained through the mediated effect.

Table 3. Standardised OLS regression coefficients with confidence intervals estimating the effects of business model riskiness on audiences.

	Model 1: OUTCOME AUDIENCES
	coeff.	se	t	p	LLCI	ULCI
Constant	2.252***	0.100	22.496	0.000	2.055	2.448
Business model riskiness	0.073*	0.033	2.222	0.027	0.009	0.138
Controls
Size	−0.053	0.047	−1.131	0.258	−0.144	0.039
NVT (Education level)	0.166***	0.047	3.529	0.000	0.074	0.259
NVT (Education heterogeneity)	0.041	0.051	0.795	0.427	−0.060	0.141
NVT (Industry experience)	0.166***	0.046	3.596	0.000	0.076	0.257
NVT (Industry experience heterogeneity)	−0.694***	0.094	−7.384	0.000	−0.879	−0.510
NVT (Managerial experience)	0.031	0.020	1.521	0.129	−0.009	0.070
NVT (Start-up experience)	−0.193	0.149	−1.293	0.197	−0.486	0.100
Venture (Location)	−0.272***	0.052	−5.230	0.000	−0.374	−0.170
Venture (Growth stage)	−0.033	0.036	−0.898	0.370	−0.104	0.039
Venture (Age)	−0.223***	0.024	−9.206	0.000	−0.270	−0.175
Venture (VC proximity)	−0.313***	0.051	−6.091	0.000	−0.414	−0.212
Technology (Blockchain type)	0.095	0.055	1.723	0.085	−0.013	0.204
Technology (Consumer focus)	0.083	0.048	1.717	0.086	−0.012	0.178
Product features (Privacy)	0.082	0.085	0.972	0.331	−0.084	0.248
Product features (Scalability)	0.017	0.021	0.784	0.433	−0.025	0.059
Product features (Customisability)	−0.141	0.150	−0.943	0.346	−0.435	0.153
Product features (Liquidity)	0.095	0.215	0.441	0.659	−0.327	0.516
Model summary
Observations	107
R	0.541
R square	0.292
Mean squared error	0.457
F-statistic	53.297
p-value	0.000

Notes: ***, * Indicates statistical significance at the 1% and 10% levels; coeff. are standardised OLS regression coefficients; bold value represents the hypothesised relations; level of confidence for all confidence intervals in output: 95 percent; number of bootstrap samples for percentile bootstrap confidence intervals: 10,000; a heteroscedasticity consistent standard error and covariance matrix estimator was used.

Our main interest is the gender effect and how it changes the relationship between business model uncertainty and audience support. We computed an interaction term as the product of uncertainty and gender, which appeared to have a significant negative relationship to audience support (β = −0.365, p < 0.001, CI: −0.424, −0.307), controlling for compounding variables. This supported the theory behind hypothesis 1. The model itself is reliable in terms of the proportion of the variance explained (R squared 0.367; see, Table 3 and Table 4).

Table 4. Standardised OLS regression coefficients with confidence intervals estimating the moderating and mediation effects of business model riskiness, gender distribution and audiences on finances received.

	Model 2a: OUTCOME FINANCE SUPPORT RECEIVED (amount)
	coeff.	se	t	p	LLCI	ULCI
Constant	0.375***	0.061	6.176	0.000	0.256	0.494
Business model riskiness	−0.005	0.014	−0.345	0.730	−0.032	0.022
Mediator
Audience	0.135***	0.017	7.806	0.000	0.101	0.168
Moderator
Gender distribution	0.171	0.071	2.419	0.016	0.032	0.309
Interaction
Audience x Gender distribution	−0.365***	0.030	−12.223	0.000	−0.424	−0.307
Controls
Size	0.059**	0.022	2.724	0.007	0.017	0.102
NVT (Education level)	0.015	0.019	0.799	0.425	−0.022	0.052
NVT (Education heterogeneity)	0.008	0.019	0.385	0.700	−0.031	0.046
NVT (Industry experience)	0.073***	0.019	3.870	0.000	0.036	0.111
NVT (Industry experience heterogeneity)	−0.022	0.093	−0.239	0.811	−0.205	0.160
NVT (Managerial experience)	0.014	0.008	1.696	0.090	−0.002	0.031
NVT (Start-up experience)	0.120	0.154	0.780	0.436	−0.182	0.422
Venture (Location)	−0.005	0.021	−0.247	0.805	−0.047	0.036
Venture (Growth stage)	−0.012	0.017	−0.713	0.476	−0.046	0.021
Venture (Age)	−0.047***	0.011	−4.258	0.000	−0.068	−0.025
Venture (VC proximity)	−0.004	0.024	−0.167	0.868	−0.051	0.043
Technology (Blockchain type)	0.017	0.021	0.808	0.419	−0.024	0.058
Technology (Consumer focus)	0.014	0.018	0.774	0.439	−0.021	0.049
Product features (Privacy)	−0.049	0.035	−1.391	0.165	−0.118	0.020
Product features (Scalability)	−0.009	0.009	−1.034	0.302	−0.027	0.009
Product features (Customisability)	−0.016	0.106	−0.151	0.880	−0.225	0.193
Product features (Liquidity)	0.133	0.165	0.807	0.420	−0.191	0.457
Model summary
Observations	107
R	0.606
R square	0.367
Mean squared error	0.083
F-statistic	54.849
p-value	0.000

Notes: ***,** Indicates statistical significance at the 1%, 5% and 10% levels; coeff. are standardised OLS regression coefficients; bold values represent the hypothesised relations; level of confidence for all confidence intervals in output: 95 percent; number of bootstrap samples for percentile bootstrap confidence intervals: 10,000; a heteroscedasticity consistent standard error and covariance matrix estimator was used.

We further calculated the conditional effects of audience support at three different levels of gender effect (model 1; Table 4; Figure 2). Our results indicate that our theorising holds for all-male teams (minimum gender distribution) but that adding women to an all-male team is beneficial. This is indicated by the results of the minimum distribution in all-male teams (β = 0.135, p < 0.001, CI: 0.101, 0168) and for the observed (not the theoretical) mean gender effect (β = 0.052, p < 0.001, CI: 0.101, 0168). The second effect is especially interesting because it seemingly contradicts our reasoning. Having a few female team members can be an advantage if we are assuming a well-known fact in innovation management: diversity drives creativity (Van der Vegt & Janssen, 2003). We tested this effect with a second analysis. In particular, the t-statistic was 0.034 (df = 2) among teams with few female members and between high- and low-risk business models. The corresponding two-tailed p-value was p < 0.01. Because in our operationalisation, riskier business models are also the more innovative ones, we can conclude that there are indeed effects of diversity on creativity and innovation that might play a role (Van Beers & Zand, 2014). Unfortunately, this effect is not strong enough to turn the gender effect around. In line with our theorising, gender effects become negative for one standard deviation above mean or larger (β = −0.073, p < 0.001, CI: −0.100, −0.045). This indicates that having a few women in a founder team shows no negative effects and can even be beneficial. As the female distribution surpasses a certain threshold, however, the gender effect sets in. This threshold indicates the boundary of the category. Hence, crossing this threshold leads to a founder’s appearance as too dissimilar to the prototype. We hence conclude that our empirical findings support our theorising.

Figure 2. Visualisation of conditional effects at three different levels of gender distributionnotes: *** indicates statistical significance at the 1% level; CE = confidence level. The black line indicates how audiences support increases or decreases with business model risk at three different levels of female team members.

According to our second hypothesis, once female founders are seen as sufficiently similar to the prototype, the disadvantage due to the gender effect disappears in further support rounds. We test this assumption in model 2 (see, Table 5). Business model uncertainty again has a non-significant and marginalised effect on support rounds (β = −0.055, ns, CI: −0.262, 0.152). However, we found a very strong and significant effect between audiences and support rounds, indicating that investors tend to stay with their investment decision in subsequent support rounds (β = 0.415, p < 0.001, CI: 0.180, 0.651). When taking the gender effect into account, this relationship changes but does not become negative. In line with our theorising, a predominantly female team has no negative influence on audiences’ decisions to support an IT start-up in subsequent financing rounds (β = 0.054, p < 0.05, CI: 0.463, 0. 571). However, while we can conclude that there is no negative gender effect in the second and subsequent support rounds, we must acknowledge that female teams still perform less successfully than their male counterparts. While there is no negative gender effect in absolute terms, there still exists a relative disadvantage for female founders. This means that we have identified one effect that limits the access of female founders to external financing, but we have to conclude that there might be more than just one disadvantaging effect in play, a fact mirrored in multiple other studies (e.g., Brooks et al., 2014; Verheul & Thurik, 2001). The model in Table 6 shows a pseudo-R squared of 0.251 (Nagelkrk), indicating the model provides sufficient explained variance in the outcome.

Table 5. Conditional effects of audiences’ support at three values of gender distribution (minimum, the mean, and one standard deviation above the mean).

Gender distribution ^#	coeff.	se	t	p	LLCI	ULCI
0.000 (minimum)	0.135***	0.017	7.806	0.000	0.101	0.168
0.226 (mean)	0.052***	0.014	3.768	0.000	0.025	0.079
0.567 (1 SD above mean)	−0.073***	0.014	−5.112	0.000	−0.100	−0.045

Notes: *** Indicates statistical significance at the 1% level; coeff. are standardised OLS regression coefficients; level of confidence for all confidence intervals in output: 95 percent; number of bootstrap samples for percentile bootstrap confidence intervals: 10,000; one SD below the mean is below the minimum observed in the data for gender distribution, so the minimum measurement on gender distribution is used for conditioning instead.

Table 6. Logistic regression coefficients with confidence intervals estimating the moderating and mediation effects of business model riskiness, gender distribution and audiences on finance rounds (round 1 and subsequent rounds).

	Model 2b: OUTCOME FINANCE SUPPORT RECEIVED (rounds)
	coeff.	se	Z	p	LLCI	ULCI
Constant	−1.385***	0.398	−3.483	0.001	−2.164	−0.606
Business model riskiness	−0.055	0.106	−0.516	0.606	−0.262	0.152
Mediator
Audience	0.415***	0.120	3.453	0.001	0.180	0.651
Moderator
Gender distribution	−0.455**	0.164	−2.774	0.006	−0.777	−0.134
Interaction
Audience x Gender distribution	0.054	0.264	0.205	0.838	−0.463	0.571
Controls
Size	0.961***	0.153	6.270	0.000	0.661	1.261
NVT (Education level)	−0.110	0.161	−0.686	0.493	−0.425	0.205
NVT (Education heterogeneity)	0.273	0.174	1.566	0.117	−0.069	0.615
NVT (Industry experience)	1.224***	0.175	6.995	0.000	0.881	1.567
NVT (Industry experience heterogeneity)	0.372	0.506	0.736	0.462	−0.619	1.364
NVT (Start-up experience)	0.192	0.801	0.240	0.811	−1.379	1.763
NVT (Managerial experience)	0.037	0.070	0.532	0.595	−0.101	0.176
Venture (Location)	−0.032	0.165	−0.196	0.845	−0.355	0.290
Venture (Growth stage)	0.353	0.124	2.849	0.004	0.110	0.596
Venture (Age)	0.001	0.085	0.010	0.992	−0.166	0.168
Venture (VC proximity)	−0.350	0.499	−0.703	0.482	−1.328	0.627
Technology (Blockchain type)	0.381	0.205	1.855	0.064	−0.022	0.783
Technology (Consumer focus)	0.130	0.161	0.807	0.420	−0.185	0.445
Product features (Privacy)	0.249	0.411	0.605	0.545	−0.557	1.055
Product features (Scalability)	−0.070	0.072	−0.963	0.336	−0.212	0.072
Product features (Customisability)	−0.956	0.672	−1.422	0.155	−2.273	0.362
Product features (Liquidity)	−1.994	0.908	−2.197	0.028	−3.772	−0.215
Model summary
Observations	107
−2LL	1228.995
ModelLL	221.710
McFadden	0.153
CoxSnell	0.185
Nagelkrk	0.251
p-value	0.000

Notes: *** Indicates statistical significance at the 1% and 10% levels; coeff. are standardised logistic regression coefficients; bold values represent the hypothesised relations; level of confidence for all confidence intervals in output: 95 percent; number of bootstrap samples for percentile bootstrap confidence intervals: 10,000; a heteroscedasticity consistent standard error and covariance matrix estimator was used.

3.4. Robustness test

For audience support to be an explanatory factor, four criteria needed to be met (Baron & Kenny, 1986; Preacher & Hayes, 2008). First, business model risk should predict audience support significantly. Second, audience support should predict the amount of finance received (Model 2a) as well as the persistence of support over different support rounds (Model 2b). Third, this relation should be (negatively) moderated by the increasing gender distribution (i.e., the increased number of female founders in a team). Fourth, after controlling for compounding variables, business model risk should have a zero impact on Model 2a and Model 2b. We used the Sobel test (MacKinnon et al., 2002) to determine whether audience support carried the influence from business model risk to the amount of finance received and the persistence of support over different support rounds. Results for Model 2a showed the explanatory power of audience support as slightly below the required significance level of p = 0.05 (Sobel Z = 2.131, std error = 0.004, p = 0.033). For Model 2b, our results are slightly above this requirement (Sobel Z = 1.863, std error = 0.016, p = 0.062), indicating a non-significant explanatory power for one of our robustness relations. Because the significance level of the Sobel test is known to be less reliable when the sample size is small or when data are not normally distributed, the results of our Sobel test are not fully reliable. Because of these issues, we focus on the direction of effects that generally support our reasoning (Preacher & Hayes, 2004).

4. Discussion

Some authors argue that gender-based biases are a simplistic explanation for the lack of funding or support for female entrepreneurs (Heilman et al., 2004). More specifically, Balachandra et al. (2019) suggest that investors are not reluctant to finance female entrepreneurs because they are women but because they are seen as too feminine. We not only support these statements but add theoretical grounding to them.

4.1. The cause of male overrepresentation is male overrepresentation

Male overrepresentation in most IT fields is not the result of obstacles preventing women from becoming IT entrepreneurs. The cause of male overrepresentation is male overrepresentation – that is, it is a self-referential process that does not reference external conditions such as women’s education, their experiences, the quality of their start-ups but only the already existing status quo. This status quo shapes audience perception of the ideal or prototypical IT start-up team as young, white, tech-savvy, male college graduates. IT entrepreneurship is widely equated with a “lifestyle” that is highly homogeneous and by no means representative of race, gender, or cultural and ethnic background. In fact, we clearly identified the pressure for female teams to conform to the normative standard represented by male-dominated category prototypes. Women experience these normative standards as givens, resulting in the core challenge to balance competing pressures from the need to be “like” the prototype but at the same time be “different from” those peers to highlight unique features and advantages (Deephouse, 1999; Durand & Calori, 2006).

4.2. IT start-ups have properties that strengthen this self-referential process

IT start-ups are highly uncertain and, as some authors argue, even more uncertain than start-ups in other industries (Kuester et al., 2018). Investors put significant emphasis on searching for indications that help to reduce uncertainty compared to the approaches used in more stable and predictable industries. As a consequence, both ascribed and overt characteristics of IT founders play a special role in investors’ decision-making processes (Greene et al., 2001). Audiences have an ideal or prototypical founder in mind who best represents all category-relevant expectations, features, and characteristics (Negro et al., 2010). These prototypes represent the reference point for decision-making. As a sociocognitive category, they do not have to exist but are utilised as a point of comparison. Thus, it is similarity and not equality that matters, while more similarity or less similarity means that a founder team drifts closer or further away from the reference point. This reference point is severely male-dominated – a fact that poses serious challenges to female founders. Given this categorisation process and the need to use the founder as a reference point for decision-making, the condition of high uncertainty strengthens self-referentiality. As IT has been associated with substantial uncertainties since its beginning, uncertainties in the start-up process of IT-based ventures might even strengthen the process of self-referentiality. As a result, we experience a higher and more persistent male overrepresentation in IT than in industries with more stable and predictable business models.

4.3. Female founders are disadvantaged because they are perceived as too distinct from the prototype before they can demonstrate their potential

We further conclude that similarity generates legitimacy, but legitimacy by itself is not sufficient for audience support because, under competition, audiences can choose whom to support. Thus, we do not suggest that female founders are disadvantaged because they are not male. In fact, we propose that some level of distinctiveness from the prototype is needed (Zhao et al., 2017; Zuckerman, 1999). Distinctiveness matters because it allows a team to stand out and attract attention (Taeuscher et al., 2021; Zhao et al., 2017). Thus, new venture teams need to forge a unique identity under strong pressure to conform (Brewer, 1991); or, in the words of Deephouse (1999), they need to be “as different as legitimately possible”. Our findings support the notion that being female is already too distinct from the prototype for many audiences. Optimal distinctiveness theory suggests that there is a clear hierarchy between both requirements (Zhao et al., 2017). Only when sufficient similarity is given do questions on how a founder differs from other founders within the same category become relevant (Zuckerman, 1999). This sequence of evaluation is a major obstacle for female founders because they receive few opportunities to showcase their distinctiveness and potential compared to their male counterparts.

4.4. Gender imbalance could lead to gender bias

Gender imbalance has the potential to turn into gender bias because gender imbalance shapes how audiences create prototypes. Following our reasoning on how overrepresentation of men becomes self-referential, we must assume that higher gender imbalance leads to gender biases as soon as category prototypes are built around the “normality” or the empirical status quo within a social context. Normality has an implicit force to shape and create normativity. Thus, what might start simply as the collective and deliberate decision of one group not to participate in a particular activity or field could turn into a normative standard over time that unfairly prevents those who deliberately opted to make a different decision, as is commonly done. Gender imbalance and gender bias, hence, are not detached from each other but linked through the self-referentiality of overrepresentation.

4.5. The normative standard in IT cannot be proven wrong by referencing empirical counterarguments

One key contribution of our paper is the uncovering of how the normative force of male-dominated prototypes exerts implicit power on female founders or founder teams. While empirical statements must fit the world, normative statements do not have to fit the world (Jellinek, 1905). In fact, it is the opposite. If the normative standard or normativity implies the prototype in IT entrepreneurship is male but the founder in question is female, the female founder is “wrong” because she fails to conform to the normative standard. Because norms are not facts, such normative standards cannot be proven wrong on empirical grounds. Self-referentiality here means that the “truth value” of such a statement is rooted in audiences’ expectations, norms, and values. Hence, they are not intended to make a statement about the entrepreneurial reality of IT start-ups themselves. Instead, they tell us what ought to be, not what is. This is unique to normative standards. On the contrary, a statement that 80% of all IT founders are women is an empirical statement with a truth value that can be proven wrong; in other words, if 80% of IT founders are not female, then this statement does not fit the world. We do not expect the world to change but hope that the false statement is corrected. The criticality of the gap between “is” and “ought” is that normative statements about the prototype do not refer to something that is supposedly the case. Thus, they cannot be corrected, as is the case for correcting false empirical statements. Consequently, all efforts to erase existing empirical obstacles for female entrepreneurs are less promising while the cycle of self-referentiality remains unbroken.

4.6. Practical implications and recommendations

The interesting aspect of our research is the demonstration that male overrepresentation functions without external influences. The cause for such systemic issues is not the external obstacles but the prototype itself. We believe that this effect, even if not fully explored yet, exerts a pressure that cannot be compensated for by better training or other support programmes for women. Better training, for example, can increase distinctiveness but does little to help create similarity. Many female founders do not, however, receive the opportunity to demonstrate their potential because they fail before they can do so. According to our research, this is the fundamental problem.

Our first recommendation, hence, is to put less emphasis on the distinctiveness elements in one’s persona. It is not the case that female founders are incapable of founding and running IT businesses; instead, the challenge is to expose similarity that places an unfair disadvantage on them. The practical recommendations derived from our study are contrary to what we currently observe in practice. Policymakers, associations, and institutions indeed place an emphasis on educating and training female founders in the hope of making them equally successful as male founders. At the same time, we observe that these initiatives have not solved the problem. To overcome obstacles for female founders, initiatives need to target the category prototype and not the skill set of female founders. Our reasoning indicates that this cannot be done directly. Prototypes are normative “facts”, not empirical ones; hence, they are immune to counterarguments rooted in empirical observations on the appropriateness of female founder skills, knowledge, and capabilities. Nonetheless, there are opportunities to change the prototypical actor. This change must come from the periphery and not from the centre (Safadi et al., 2021).

Niche markets, special applications, and ventures at the boundaries of the category in which women are less underrepresented should be targeted to create alternative prototypes at the periphery of the IT entrepreneur category. One decisive property of categories is that they lose relevance when boundaries become fuzzy and ill-defined. This offers opportunities to establish alternative prototypes because the steady infiltration of a category causes new expectations. As blurred category boundaries present the risk of a category drifting into insignificance (Negro et al., 2010), this offers an opportunity to break the male dominance in IT start-ups.

A further practical implication refers to the relationship between gender imbalance and gender bias. As we have seen, it is difficult to change prototypes, especially when self-referentiality allows them to persist over a long time. This fact emphasises the importance of preventing the emergence of prototypes that might lead to an unfair advantage in the first place.Because the notion exists that biases follow imbalances, we recommend targeting the gender imbalance, even if such an imbalance is not critical per se. Imbalance, however, can trigger self-referentiality and stabilise gender imbalance to the extent that it becomes a bias. This happens if male-dominated prototypes emerge due to the lack of alternative role models.

This recommendation relates to our generally rather critical or reluctant position towards the showcasing of female founders’ potential and success. Showcasing the potential of female founders is a double-edged sword. On the one hand, it presents an alternative to the prototype, challenging category expectations. Admittedly, the steady and continuous challenging of audience beliefs might lead to the reformulation of category expectations. However, and this is the critical aspect, emphasising the occasional success of an IT start-up led by female founders lets them appear as the exception to the norm. This strengthens the prototype, but it is exactly this that needs to be prevented. We, therefore, argue that the negative effects of glorifying the success of a few female founders might be counterproductive. Thus, we recommend only showcasing the success of female founders if there are significant cases available. Single or occasional examples are not sufficient to challenge the male-dominated prototype in IT.

Finally, we wish to warn of a potential misunderstanding of our research. We do not advise female founders to seek success by acting like the prototype (Balachandra et al., 2019). Increasing similarity by copying male behaviour is risky because women who behave in masculine ways risk creating the perception of being overly assertive, which results in being viewed negatively with the consequence of possible backlash (Heilman et al., 2004). Our study by no means implies such recommendations.

4.7. Limitations and further research

Some limitations of our study provide avenues for further research. Specifically, our research has not done justice to the overall group of female founders that we have treated as a homogeneous group. More importantly, we have adopted a binary concept of gender that does not reflect the plurality of gender identities (D’Ignazio & Klein, 2020). While it is likely the case that every identity that drifts away from the prototype faces similar obstacles as described in our study, we admit that “drifting away” could have various meanings when gender identities alter. Obstacles might then be affected during this alteration. Today, we know almost nothing about these processes, which has significant consequences for our theorising on gender biases or entrepreneurial success.

A further limitation results from our methodological design, which has blind spots because it only allows us to identify net effects and probabilistic relations. Methodologically, we are blind to qualitative elements such as meaning, sense-making, and constructivist insights into how IT start-up reality is created, perceived, and maintained. We added dependencies that underly the phenomenon; to draw a more comprehensive picture, qualitative research is needed. For example, audiences do not only attribute expectations to categories, but categories provide a framework for seeing, interpreting, and making sense of what is observed. This interconnection could provide deeper insights into the categorisation process of the male founder and its persistence over time.

A further avenue for future research is the normative force of the factual. We believe that this force is not only responsible for gender biases but also biases regarding race relations, and so on. While this is a powerful force, we know little about the processes that facilitate this effect. With our quantitative approach, we were only able to scratch the surface.

We have identified a fair number of start-ups with female involvement but could not identify an all-female team with a blockchain-based business model. While this strengthens our point on the underrepresentation of women and existing gender imbalance, it does not allow us to compare all-male to all-female teams, which consequently weakens our results. We also received mixed results regarding the conditional effects of our gender effect, which we cannot fully explain. Our results indicate that some degree of female involvement can even be an advantage for receiving audience support. A deeper understanding of this effect would help contextualise our theorising and potentially allow us to add a further layer of complexity to craft a more precise theory. Unfortunately, our dataset lacks relevant observations to allow us to explore this matter in more depth.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ We acknowledge that a binary view of gender is too narrow to represent the plurality of gender identities (D’Ignazio & Klein, 2020). Hence, we do advocate for a non-binary conception of gender. In what follows, we focus on the binary view of the gender identity spectrum for the purpose of identifying one root cause of gender bias in IT entrepreneurship. The arguments presented will, however, hold for every gender identity that is distinct from the gender identity of the prototypical actor