Founder personality and entrepreneurial outcomes: A large-scale field study of technology startups

Significance

Technology startups are essential to the global economy. Yet, predicting their short- and long-term outcomes remains difficult. This is particularly true in the early stages of a venture’s lifecycle when little to no performance data are available. We offer large-scale, ecologically valid evidence for the importance of a startup feature that is available from the moment of conception: founder personality. Our findings suggest that while some traits consistently predict startup outcomes across all stages (i.e., emotional stability), others reverse their associations with entrepreneurial outcomes as the startup matures from conception to exit (i.e., conscientiousness). By doing so, they offer novel insights into the role of founder personality in technology startups.

Abstract

Technology startups play an essential role in the economy—with seven of the ten largest companies rooted in technology, and venture capital investments totaling approximately $300B annually. Yet, important startup outcomes (e.g., whether a startup raises venture capital or gets acquired) remain difficult to forecast—particularly during the early stages of venture formation. Here, we examine the impact of an essential, yet underexplored, factor that can be observed from the moment of startup creation: founder personality. We predict psychological traits from digital footprints to explore how founder personality is associated with critical startup milestones. Observing 10,541 founder–startup dyads, we provide large-scale, ecologically valid evidence that founder personality is associated with outcomes across all phases of a venture’s life (i.e., from raising the earliest funding round to exiting via acquisition or initial public offering). We find that openness and agreeableness are positively related to the likelihood of raising an initial round of funding (but unrelated to all subsequent conditional outcomes). Neuroticism is negatively related to all outcomes, highlighting the importance of founders’ resilience. Finally, conscientiousness is positively related to early-stage investment, but negatively related to exit conditional on funding. While prior work has painted conscientiousness as a major benefactor of performance, our findings highlight a potential boundary condition: The fast-moving world of technology startups affords founders with lower or moderate levels of conscientiousness a competitive advantage when it comes to monetizing their business via acquisition or IPO.

Entrepreneurs and the new ventures they create are of great importance to both the economy and the diffusion of innovative technologies. Yet, the trajectories of new ventures, especially technology startups, remain highly unpredictable. It is thus no surprise that the determinants of startup outcomes, especially during the early formative stages when few business metrics may be available, are of significant interest to both researchers and practitioners alike (1, 2).

One potential determinant of a startup’s trajectory that has received increasing interdisciplinary attention is that of founder personality. Recent work has identified differences in personality traits between founders and CEOs/employees (3–5) and linked certain founder personality traits to entrepreneurial success (4, 6). For example, Zhao et al. find that conscientiousness and emotional stability predict venture growth (6). However, the ability of prior work to study the relationships between founder personality and startup outcomes in large samples and settings of high ecological validity has been limited by a number of methodological challenges. These include i) problems related to collecting large sample sizes from the field^*, ii) potential self-report biases associated with survey responses that capture both personality and outcomes (i.e., success may mean different things to different founders, 7) and iii) concerns over endogeneity (e.g., to what extent founders’ personalities change as a response to their entrepreneurial experiences). In addition, despite the growing importance of technology ventures to the global economy, high-tech founders have received relatively little attention among personality scholars (8). Because of this, and the fact that the ecosystem for technology startups differs from other industries in a number of important ways (e.g., the heavy reliance on venture capital and the focus on quick exits, 9–11), the generalizability of existing findings to high-tech ventures remains speculative.

Here, we offer large-scale evidence for the association between founder personality and entrepreneurial outcomes among technology startups. In a first step, we estimate founder personality using the well-established Big Five model (e.g., 12, 13), which posits five personality traits: openness to experience, conscientiousness, extraversion, agreeableness, and neuroticism (inversely known as emotional stability). The Big Five personality traits have been extensively validated (e.g., 12, 13) and shown to predict a wide range of consequential life developments including entrepreneurial preferences and outcomes (6, 14, 15). While the assessment of personality traits has traditionally relied on self-report questionnaires, recent research has shown that they can be predicted with high levels of accuracy from the digital footprints people create when they interact with technology (e.g.,  16–18). For the purpose of this study, we predict founders’ personality from the language extracted from their Tweets in the 2 y prior to founding (see Methods for details on the predictive models).

In a second step, we link these personality profiles to the following startup outcomes across all stages of the entrepreneurial life cycle: i) whether the startup raised funding, ii) the amount it raised in the earliest funding round, iii) the number of investors it included in the earliest funding round, and iv) whether it exited [via acquisition or initial public offering (IPO)]. The data were accessed using Crunchbase’s research application programming interface (API) (19). The final dataset includes 10,541 US-based founder–startup pairs, their predicted personalities, and their startups’ outcomes (Fig. 1).

Fig. 1.

Data-processing flowchart. Founders’ Big Five personality traits were predicted from their Tweets. The predictive models (ridge regressions based on sentence embeddings) were trained on MyPersonality data which combines self-reported Big Five traits with status updates. Startup outcomes were extracted from Crunchbase.

Results

 Fig. 2 displays the results of a series of logistic and linear regression analyses predicting startup outcomes from Big Five personality traits and controls (see SI Appendix, Table S1 for full model output and SI Appendix, Table S2 for the variable means and zero-order correlations). To facilitate the interpretation of effects, the personality estimates were z-standardized. All analyses control for the year the company was founded, the number of founders, the gender of the founder, US state, and industry-fixed effects.

Fig. 2.

Relationships between founders’ Big Five personality traits (z-standardized) and startup outcomes. (A) %-Change in the likelihood of raising funding. (B) Change in the dollar amount raised in the first round (in Million). (C) Change in the number of investors in the first round. (D) %-Change in the likelihood of exiting (via IPO or acquisition). d*P < 0.05, **P < 0.01, ***P < 0.001.

Whether a startup raised initial funding (Fig. 2A, logistic regression) was positively related to openness and agreeableness. An increase of 1 SD in openness and agreeableness was associated with a 5% higher likelihood of attracting funding. For those startups that raised initial funding, the amount they raised during the startups’ earliest funding round (Fig. 2B, linear regression) was significantly related to conscientiousness and neuroticism. An increase of 1 SD in conscientiousness was associated with an additional $170,000 raised, while an increase of 1 SD in neuroticism was associated with a drop of $90,000. Similarly, the number of investors included in the first round of funding (Fig. 2C, linear regression) was significantly related to conscientiousness and neuroticism. A 1 SD increase in conscientiousness and neuroticism was associated with having 0.21 and 0.20 fewer investors, respectively. Finally, whether the startup exited (via acquisition or IPO; Fig. 2D, logistic regression) was negatively related to conscientiousness and neuroticism. A 1 SD increase in conscientiousness and neuroticism was associated with a 15% and 16% lower likelihood of exiting, respectively. Taken together, the results suggest that founder personality plays an important role across the different phases of a new venture’s development, from initial fundraising to exit.

Discussion

Studying the associations between founder personality and startup outcomes in settings of high ecological validity has been challenging. Using natural language processing to predict the personality of US-based tech founders from their Tweets, we show that founder personality is related to objective outcomes at all stages of the entrepreneurial process (i.e., from the initial fundraising rounds all the way to the exit).

For example, both openness and agreeableness were associated with a 5% increase in the likelihood of raising a first round of funding (but were unrelated to all the subsequent, conditional outcomes). These associations could be explained by different mechanisms that we cannot disentangle with our study design (see limitations below for more details). On the one hand, the effects might be the result of founders’ dispositional motivations and abilities. For example, openness is closely aligned with the archetype of the curious and inventive founder (20, 21). Therefore, open founders may be more likely to develop products that garner attention. Aligned with this proposition, past research has linked inventor’s openness to higher patent citation counts (22). Similarly, agreeable founders might be able to leverage their larger and more supportive networks to identify and persuade investors (23). On the other hand, the results could also be driven by investors’ perceptual biases. For example, investors might show a preference for curious/inventive founders who fit the stereotypical entrepreneurial mold or those who appear friendly and empathetic during early conversations (for evidence highlighting biases in investors’ decision-making, 24, 25).

The two traits that most consistently predicted startup outcomes were conscientiousness and neuroticism. Although neither of the two traits predicted the likelihood of raising an initial funding round, each significantly predicted all subsequent, conditional outcomes (i.e., the amount raised in the first round, the number of investors, as well as the likelihood of exiting via IPO or acquisition).

In line with a large body of literature suggesting that neuroticism is generally related to worse life outcomes (14, 26) and recent research highlighting the negative effect of neuroticism on entrepreneurial outcomes (6), we find that higher levels of neuroticism among founders were associated with a funding gap of approximately $90,000 and 0.2 fewer investors in the first round, as well as a 16% lower chance of exiting. On the one hand, these findings might highlight the value of emotional stability, resilience, and self-confidence in overcoming the many obstacles startup founders are likely to face (6, 27). On the other hand, they could also be indicative of perceptual biases among investors. Given that neuroticism is associated with lower levels of emotion regulation, founders who are perceived as neurotic might be considered less competent and trustworthy by investors, employees, and the public (28).

In contrast, conscientiousness showed a more complex pattern that varied across the different outcomes we observed. Specifically, conscientious founders raised approximately $170,000 more from 0.2 fewer investors, but had a 15% lower likelihood of exiting. While the positive relationship with the initial amount of capital raised aligns with the existing characterization of conscientiousness as a major benefactor of both individual performance and firm performance (6, 29–31), the reversal to a negative relationship with the likelihood to exit via IPO/acquisition can be interpreted in a number of different ways that should be further explored by future research.

We speculate that these patterns might be driven by different stages favoring different aspects of conscientiousness stemming from founder dispositions (i.e., their intentions and ability), investor preferences, and the interaction of the two. Specifically, at the early stages of venture creation—i.e., prior to the first round of fundraising—high levels of conscientiousness might afford founders the ambition to aim for high-profile outcomes and the ability to develop a well-thought-out business plan to accomplish their goals. The former is supported by prior work linking high levels of planning, forecasting, and logic among founders (i.e., aspects of conscientiousness) to a higher likelihood of targeting acquisition and IPO-based exit strategies (32). Given that nearly all returns to venture capital come from such acquisitions and IPOs (33–35), investors might be more likely to invest in conscientious founders whose goals for the future of the business more closely align with theirs. The latter is based on the assumption that conscientious founders are more likely to excel at the tasks necessary to develop a sound and water-tight business plan (e.g., conduct meticulous market research, formalize financial projections, and formulate a strategic vision, 36, 37). This step is critical for two reasons. First, the formulation of a business plan has been directly linked to venture viability (38). Second, a sound business plan and founder preparedness are two of the major decision criteria for early-stage investors (39). That is, while the public image often associates entrepreneurship with creativity and ideation, investors may initially prioritize signals that a founder is able to reliably execute critical day-to-day tasks that require the level of organization, diligence, and goal orientation associated with high levels of conscientiousness. In addition to their ambition and ability to produce high-quality business plans, conscientious founders might also be afforded advantages through indirect signals they send to investors. For example, research suggests that when deciding whether to invest in an early-stage startup, investors focus their due diligence on signals of past success, including founder’s educational attainment and performance in their previous jobs (40), both of which are related to high levels of conscientiousness (6, 29–31, 37). All in all, the interplay of founders’ dispositions and investors’ preferences might afford conscientious founders an advantage when it comes to raising initial capital during the early venture stages.

However, as the startup matures and moves past the initial stages, the dynamics between founders and investors might change in several ways. First, the same ambition that led conscientious founders to strive for success via IPO or acquisition at the beginning of their journey, might later prevent them from selling their venture once it is successful. For example, founders scoring low on conscientiousness might be more interested in quick financial gains (thereby aligning their goals with those of investors and one interpretation of startup success promoted by management scholars 41–43), whereas highly conscientious entrepreneurs might pursue long-term goals that no longer equate success with selling the company as fast as possible. Specifically, the combination of ambition with the desire to retain control over the future of their business might lead conscientious founders to place greater emphasis on long-term profitability and impact. Second, while the early stages of developing a business plan and gaining investors trust may have benefitted from the meticulous and organized tendencies of the conscientious founder, the later stages might favor founders with the flexibility and ability to adjust to new challenges and opportunities as the startup matures [e.g., through rapid ideation and prototyping rather than rigorous forecasting and strategic planning (10, 44)]. Finally, the advantage afforded to founders scoring lower on conscientiousness could also be driven by shifting investor and market incentives that align with the (in)famous Silicon Valley “move-fast-and-break-things” culture (45). Potential acquirers, for example, might prefer a founder who they view as disruptive and adaptive, as a means to inject their own company with innovative capabilities or culture (46).

Taken together, our findings identify a potential boundary condition of conscientiousness as an unequivocal benefactor of performance (6, 29–31). Specifically, they suggest that the fast-moving world of technology startups (which substantially differs from other industries, 47, 48), rewards different behavioral dispositions at different startup stages. Importantly, our findings are purely descriptive in nature. They suggest that founders low in conscientiousness are afforded a competitive advantage in the current ecosystem when it comes to acquisition and IPO, not necessarily that they should be. For one, the current focus of technology startups on quick growth at the expense of careful planning and due diligence might create societal externalities (49) that make advocating for lower levels of conscientiousness in the technology space problematic. In addition, prior research has shown that “moving fast” is associated with both high levels of success and high levels of failure (9). Consequently, future research should explore whether the effects of conscientiousness might follow a nonlinear (potentially curvilinear) relationship. That is, both extremely low levels of conscientiousness (i.e., complete lack of planning or unethical deviation from norms and laws) and extremely high levels of conscientiousness (i.e., inability to adapt quickly and course correct when necessary) might be detrimental to long-term startup success.

Irrespective of the specific mechanism underlying the relationship between founders’ conscientiousness and the propensity to exit, the discrepancy between the amount invested in startups run by conscientious founders and the likelihood of these startups rewarding investors by exiting, might suggest that venture capitalists are not allocating resources efficiently. While we cannot directly test this conclusion in our data, future research should investigate whether adding personality traits to predictive decision-making models (50) could increase the effectiveness and profitability of investments.

Extraversion was the only personality trait that did not show any significant relationships with startup outcomes. This finding stands in contrast to prior literature which has found a weak relationship between founder extraversion and startup performance (6). However, the difference might be explained by our focus on technology startups. The public depiction of technology founders as “introverted geeks” (e.g., the TV series Silicon valley) and the prominence of many successful introverted tech founders (e.g., Bill Gates, Mark Zuckerberg, or Steve Jobs) might offset any positive effect extraversion typically has (51).

Our study has several important limitations. First, we cannot directly speak to the causality of our effects or disentangle the mechanisms by which the effects operate. Although the Tweets were collected prior to the founding of the startup to reduce the risk of reverse causality, we cannot rule out confounds that might have impacted both. In addition, we cannot distinguish between founder and investor effects. That is, the personality traits of entrepreneurs might impact outcomes through dispositional differences in motivation and ability associated with the entrepreneurs or perceptual biases that arise when investors evaluate the entrepreneurs and their startups. Such perceptual biases could also arise from founders with certain personality traits misrepresenting or exaggerating the effectiveness or success of their product. For example, prior work suggests that people low on conscientiousness are more likely to deploy strategic positioning and dishonest behavior (52, 53). While we partly alleviate some of these concerns by focusing on objective outcome metrics, these metrics might themselves be biased by investors’ responses to exaggeration and deceit. Second, our context is limited to tech startups which operate in an ecosystem that substantially differs from other industries. For example, the failure rates of startups in the technology space are much higher than those in other industries (47, 48). This could partly be driven by the fact that the technology landscape changes more rapidly, making it more difficult for founders to keep up. These differences are critical when evaluating our findings for conscientiousness. For example, the effects of conscientiousness on long-term outcomes might look different in an environment that is more stable and benefits from the organized and reliable nature of conscientious founders all the way to the end. Moreover, what constitutes success may differ across industries. Although exits via IPO or acquisition have previously been considered as a marker of startup success in the technology space (41–43), there are numerous other potential outcomes of importance such as impact, profitability, or simply survival. For example, survival is a common success metric in the adjacent space of nontechnology ventures. In the technology space, however, it is less clear whether survival necessarily equals success. As has recently been argued, technology startups often seek explosive growth or fast and efficient failure (9, 10). Third, our sample is limited to entrepreneurs who use social media, and the accuracy of their profiles is likely a function of the amount they post. Given that personality itself is known to influence people’s likelihood of joining social networks and posting (54), our measure of personality introduces a confound that is difficult to account for. For example, prior research has linked extraversion to more active engagement with social media platforms (54). While this might impact the reliability of our personality estimates (i.e., more data means more precise estimates), we do not expect it to systematically impact the relationships between our personality estimates and outcome measures. Fourth, while many startups have multiple founding members, we treat founders as independent observations and only study the dyadic relationship between a given founder and their startup’s outcomes. Future research could explore how the coherence or complementarity of personality profiles among all founding members of a company impacts short- and long-term outcomes. Fifth, we limited our analyses to the Big Five personality traits. As prior work has suggested, more specific psychological traits, such as self-efficacy, internal locus of control, need for achievement, and innovativeness, might provide additional predictive power (3, 4). In addition, future research would benefit from breaking down the main personality traits into facets or even specific conscientiousness-related behaviors. For example, our interpretation of the effects observed for conscientiousness is complicated by the fact that different facets could influence startup outcomes differentially or even in opposing directions. While facets such as achievement orientation may benefit founders across all stages, others such as orderliness may switch their relationship as the startup matures (4). In addition, it is possible that certain behavioral expressions of conscientiousness are beneficial to startup success (e.g., the use of structured A/B testing mechanisms, 10) while others are not (e.g., rigid adherence to existing plans that might become outdated). Given that we cannot distinguish between different facets of conscientiousness, it is also plausible that our measure more closely aligns with some facets than others. Consequently, our findings only speak to the relationship between entrepreneurial outcomes and the particular aspect of conscientiousness that is captured in large-scale language models. Future work wishing to disentangle facets will require new datasets that combine facet-level self-report data (e.g., using the BFI-2, 55) with text data gathered from social media profiles as the foundation for more granular predictive models. Finally, our results are conditional on individuals selecting into entrepreneurship, a decision that in and by itself is known to be related to personality traits (3, 4).

Taken together, our findings contribute to an advanced understanding of how founder personality impacts the outcomes of technology startups across different stages. Notably, the methods introduced in this paper could also be used to explore selection into entrepreneurship and within-subject changes in personality. For example, a growing literature in personality psychology suggests that personality traits not only change over the course of a person’s lifespan (56, 57) but are also responsive to major life events (58, 59). Future research could build on these findings to investigate how founder personality changes as a function of environmental dynamics such as startup failure and experimental interventions such as entrepreneurial training programs (51). As a consequence, the current research provides a fruitful starting point for exploring the role of personality traits in the entrepreneurial process in settings of high ecological validity.

Methods

The study was reviewed by Columbia University’s ethics review board and deemed exempt.

Predicting Founder Personality.

Founder personality was estimated using five predictive models that translate a founder’s Twitter posts into Big Five personality traits (compare to 18, 60). Automated personality predictions from Twitter have been shown to provide valuable insights in the context of vocational preferences and outcomes (61, 62).

Training Predictive Models on MyPersonality Data.

The predictive models were trained with data from the MyPersonality project (63). MyPersonality was a Facebook application active between 2007 and 2012 that allowed users to take validated psychological assessments and receive immediate feedback on their responses. The assessments also included a 100-item measure of the Big Five personality traits (International Personality Item Pool, 64). After receiving feedback, users could voluntarily share their Facebook profile information (including Facebook status updates) with the research team.

Prior work has shown that the Big Five personality traits can be accurately predicted from the language found in status updates (18, 65). Notably, these computer-based predictions not only align with people’s self-views but also predict external criteria with the same level of accuracy as self-reports (18).

In this study, we used a subsample of n = 65,598 users in the United States (i.e., people who set their home country to United States or indicated US English as their primary language). We trained five predictive models––one for each personality trait––using the following procedure: First, we removed users with fewer than 50 status updates to increase the robustness of our models. This resulted in a sample of 47,521 users with an average number of 3,531 ± 3,375 SD words. Second, we transformed each status update into a 4,096-dimensional sentence embedding using Facebook’s InferSent (66), an embedding model trained on web and social media text. We chose to use word embeddings over other preprocessing methods (e.g., bag-of-words, n-grams, or topics) because they capture the semantic meaning of text and are thus better suited for transfer learning (i.e., applying models built on one set of data to a new task or domain) (67). For each user, we extract the element-wise mean vector, resulting in an input matrix of 47,521 users × 4,096 features. Third, we split the data into training and test sets (test size = 10% of the total sample, n = 4,753 users) and trained the models using ridge regressions (68). Ridge regressions are similar to linear regressions but add a penalty term for the squared magnitude of the coefficients (“L2 regularization”). To improve the predictive performance of the model and reduce the risk of overfitting, we tuned the model using 10-fold crossvalidation on the training data. Model performance was measured as the Pearson’s product–moment correlation between self-reported and predicted personality scores in the test set (separately for each personality trait). Aligned with prior research (18, 60), the accuracies were found to range from r = 0.30 for agreeableness, r = 0.34 for conscientiousness and neuroticism, all the way to r = 0.40 for extraversion and openness.

Applying Predictive Models to Founders’ Twitter Data.

We used the predictive models trained on the MyPersonality data to estimate founders’ personality from their Tweets (n = 10,541 founder–startup pairs, 12.23M Tweets, 170.33M words). Tweets were accessed via the Twitter API for the 2 y prior to founding. We only considered founders who had tweeted at least 50 times. The average number of Tweets for each founder was 1,162 ± 1,961 SD and the average number of words was 16,017 ± 27,382 SD. Past research has shown that personality can be accurately predicted from Tweets (69), and that predictive models trained on Facebook data offer high predictive accuracy and validity when used on Twitter data (70). In addition, Twitter is known to be an important communication platform among startup and investor communities (71).

We followed the same preprocessing procedure as described in the section “Training Predictive Models on MyPersonality Data” to transform a founder’s Tweets into a 4,096-dimensional element-wise mean vector. Using the predictive models described above, we then predict founder personality for each of the five personality traits (see SI Appendix, Fig. S1 in SI for distributions of the predicted personality traits).

Extracting Entrepreneurial Outcomes.

We used the Crunchbase API to connect founders to startups and the following four startup outcomes (19): i) whether the startup raised funding, ii) the amount it raised in the earliest funding round, iii) the number of investors it included in the earliest funding round, and iv) whether it exited (via acquisition or IPO). Crunchbase is a repository of startup data that is among the most extensively cited in the technology entrepreneurship literature (43). Its data are collected through a variety of methods to ensure both accuracy and coverage. These include 1) a broad variety of publicly available news sources, 2) proprietary information from close relationships with venture capital firms and entrepreneurs, 3) information from partnerships with other digital sources of data (such as early-stage job board AngelList), and 4) crowd-sourcing (which is subsequently manually validated by Crunchbase staff, 72, 73). Crunchbase’s coverage of early-stage technology startups tends to be better than comparable databases (such as PitchBook, VentureSource, and PwC MoneyTree, 11). The database includes information on founders, startups, venture capitalists, funding rounds, acquisitions, and IPOs. Importantly, records of startups cannot be deleted from Crunchbase once created, and there is hence no bias of whether and how long startups survive (accordingly, a large portion of startups on Crunchbase never raise funding, 11). Moreover, Crunchbase has been used extensively to study startup outcomes among technology ventures. Similar to the outcome metrics we use in our analyses, common early-stage metrics include whether a startup raises funding, continues to raise funding, and the amount raised during early financing rounds (43, 72, 74). Similarly, whether or not a startup achieves an IPO or is acquired has previously been considered a marker of late-stage performance among venture-backed technology startups (41–43), although this perspective is not universally supported by all organizational scholars (75). In some instances, the metrics are combined into a single outcome measure (73, 76).

For the purpose of our analyses, we extracted data for each US-based startup founded between 2010 and 2014 which had at least one founder associated with a valid Twitter URL. For each of these startups, we accessed the following outcomes: whether the startup raised funding, the amount of money it raised during its earliest funding round, the number of investors it raised money from during its earliest funding round, and whether the startup was acquired or IPO’d. Amount of money raised and number of investors raised from during earliest funding round were winsorized prior to analysis. Consistent with prior research on technology entrepreneurships, acquisitions significantly outnumber IPOs in our data (e.g., 43). To control for potential confounds in our analyses, we also extracted founder gender, the number of founders, the year founded, location (US state), and industry. We include as controls all industries representing greater than or equal to 1% of industries associated with startups in our sample.

The startups included in the analyses were between 8 and 12 y old. While this might be considered relatively young in some industries, the majority of startup exits (acquisitions and IPOs) and closures in the technology space occur within 8 y (43, 77). In addition, most venture capital funds are structured over 10 to 12-y periods, during which time all assets must be liquidated and all distributions paid out to limited partners (the investors of the venture capital fund, 78).

Analytical Strategy.

We ran a series of logistic or linear regression analyses to predict each of the four outcomes. Given that the five personality traits are correlated, we included all the five traits in each model simultaneously to measure the unique impact of each personality trait on startup outcomes. All analyses control for the year the company was founded, the number of founders, the gender of the founder, US state, and industry-fixed effects.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

We have deposited the predicted personality scores and all analysis code to replicate the main result in an Open Science Framework repository (https://osf.io/s9hu6/) (79). Given the proprietary and sensitive nature of the raw Facebook profiles used to train the personality models, we do not publicly release these data and only make them available upon reasonable request from the second author. The Crunchbase and Twitter data used in our analysis is publicly available through their respective APIs.