Shortest paths to success: Network indicators of performance in innovation ecosystems

Abstract

In this thesis I show how various theories and methodologies borrowed from complexity science, organisation science, and network science can be suitably integrated to provide a comprehensive and interdisciplinary approach to the study of innovation processes. I study the network foundations of success in innovation ecosystems and I conduct several empirical investigations to identify those network characteristics that are expected to correlate with positive outcomes and success. I assess the extent to which the diversity and the strength in the networks of relationships boost the performance and success of scientists and early-stage firms. To this end I analyse two large-scale data sets about scientific publishing and start-up firms by making use of already existing topological network measures and by proposing novel measures to characterise the degree of interdisciplinarity and access to diverse pools of knowledge in scientific collaborations. Results provide empirical support to the idea that collaboration sustains innovation and performance by facilitating knowledge diffusion, acquisition and creation. First, results indicate that the networks of interaction between start-ups have a strong impact on the firms' longterm success. Second I find that, while abandoning specialisation in favour of moderate degrees of interdisciplinarity deteriorates scientific performance, very interdisciplinary scientists tend to outperform specialised ones. Additionally, I address the computational challenges related to the size of the data sets used and their time-varying nature. In particular I focus on the scalability challenges of incremental graph algorithms. The thesis contributes in this direction by proposing new efficient algorithms and data structures to handle and to analyse large graphs whose nodes and edges change rapidly over time. These efforts have been collected and made available to the public in the form of a web platform (http://lab.startup-network.org/) and an open-source python package, NetworkL (https://networkl.github.io/).