Three Essays on Network Analysis and Patent Citation Networks
- Author(s): Hancock, Galen
- Advisor(s): Hall, Bronwyn H.
- et al.
Patent statistics provide insight into a range of economic phenomena in the economics of innovation. Since the earliest research using patent data, it was recognized that patents are highly heterogeneous with a minority of patents having much greater impact and value than others. Citation counts are a key measurement of patents, observable directly from patent bibliographic data, which carries information about patent characteristics. But patent citations form a complex network that carries more information about patents than the number of a patent's immediate neighbors.
This dissertation explores network measures that exploit this structure to extract more information about patents. The first chapter of this dissertation is addressed to network centrality and the measurement of patent value. The centrality of a patent is negatively related, after controls for citation counts, to public company market value. This establishes that centrality measurements carry information about patent attributes beyond citation counts, and suggests a clear market saturation phenomenon that may drive the result.
The next chapter considers whether patent centrality measurements can detect probable holdup in the economic space of patents. Patent pools associated with technical standards, such as the MPEG-2 video compression method, can relate to inventions subject to many competing rights. Centrality measurements are lower for patents that are in the patent pool, suggesting a line of research into the measurement of patent complementarity and substitute relationships.
The remaining chapter concerns technology identification from patent citations. Classifying patents by technology is a common necessity in patent empirics, and existing approaches using administrative classification suffer certain inevitable limitations due to the nature of the classification system. Community detection algorithms can use citations to break patents into natural groups. An empirical analysis shows that the communities thus produced contain more information than administrative classifications and can detect underlying technological changes before the classification system does.