Learning Algorithms for Discrimination Free Innovation Funding Activities

We investigate whether digital traces can be used to predict early stage startup survival. Based on common survival factors from the entrepreneurship literature, we mined the digital footprints of 542 entrepreneurs and their ventures. Using a context-specific text mining approach, we performed a bootstrapping simulation in which we predict 5-year survival for different survival rates that range from 50% to 10%. Our results indicate that we can predict 5-year survival with an accuracy of up to 91%. With this study, we will provide an evidence-based taxonomy of digital traces for predicting early stage startup survival, identify the most important digital traces for doing so and benchmark our predictive approach against the actual investments of 339 business angels.

Research indicates that interactions on social media can reveal remarkably valid predictions about future events. In this study, we show that online legitimacy as a measure of social appreciation based on Twitter content can be used to accurately predict new venture survival. Specifically, we analyze more than 187,000 tweets from 253 new ventures' Twitter accounts using context-specific machine learning approaches. Our findings suggest that we can correctly discriminate failed ventures from surviving ventures in up to 76% of cases. With this study, we contribute to the ongoing discussion on the importance of building legitimacy online and provide an account of how to use machine learning methodologies in entrepreneurship research.