Changes
Jump to navigation
Jump to search
Huang Murray (2009) - Does Patent Strategy Shape The Long Run Supply Of Public Knowledge (view source)
Revision as of 19:07, 27 March 2013
, 19:07, 27 March 2013no edit summary
*This page is referenced in the [[Patent Thicket Literature Review]]
*This page is listed on the [[PTLR Core Papers]] page
Knowledge-based firms seeking competitive advantage often draw on the public knowledge stream (ideas embedded in public commons institutions) as the foundation for private knowledge (ideas firms protect through private intellectual property [IP] institutions). However, understanding of the converse relationshipthe impact of private knowledge strategies on public knowledge productionis limited. We examine this question in human genetics, where policy makers debate expanding IP ownership over the human genome. Our difference-in-differences estimates show that gene patents decrease public genetic knowledge, with broader patent scope, private sector ownership, patent thickets, fragmented patent ownership, and a genes commercial relevance exacerbating their effect.
==Review==
===Measures of thicket===
Patent thickets are measured by:
*The number of patents claiming the same gene.
*A fragmentation index based on Herfindahl-type indices of the number of owners of different patents reading on same gene.
===Sample===
*1,279 human gene patent-paper pairs can be identified (same gene sequence in patent and academic paper) were identified from the universe of 4,270 human gene patents.
**12,830 annual observations of papers from 1998-2006.
**Academic papers are from Thomson ISI Web of Science database.
**Patent-paper pairs matched a human gene sequence, names of patent inventor to paper author, and dates of patent application and paper publication.
===Results===
Patent thicket results:
:''"a modest 2 percent increase owing to the denser patent thickets than existed for genes with only one patent...Although the 2 percent difference provides weak support for Hypothesis 5, the underlying relationship between gene patent grant and long-run public knowledge production is clearly not linear in the number of gene patents and does not increase smoothly with thicket density."''
*The number of citations to patent-paper pairs declines by 2% for denser patent thickets - citations are significantly reduced by 9% compared to 7% when number of patents claiming the gene is 5-7 patents versus 1 patent.
:''"strong support for Hypothesis 6. Specifically, over and above the baseline decline in expected citations of 5 percent (also see model 3, Table 6), there is an incremental 7 percent decline (p<.05) in follow-on knowledge production for every unit increase in fragmentation of the patent thicket (relative to the mean and in absolute terms). For example, for any given patent, an increase in ownership from one to two organizations for a claimed gene would result in an incremental 3.5 percent decline in the forward citations of the paired paper."''
Other results:
:''"...for the production of public knowledge (in human genetics), the main impact of patenting arises through private sector gene patents."''
*Granting of a patent significantly negatively reduces citations of patent-paper pairs by 17% or 5% baseline and difference-in-differences models respectively.
*Patent scope modestly reduces citations of patent-paper pairs by 1%.
*More useful patents (OMIM, cancer, disease genes) significantly reduce citations.
*Private patent ownership reduced citations of patent-paper pairs by 6-9% depending on specification.
===Social Welfare Consequences===
:''"We show that although the relationship between patent thicket size and knowledge production is noisy (the results were directionally as we predicted), increasing ownership fragmentation contributes significantly to the negative impact of patent grant. This finding suggests that even though ownership concentration is an important source of competitive advantage for owners of a patent portfolio, fragmentation is more problematic for follow-on contributors to public knowledge, presumably because of the complexities and costs of navigating and negotiating with many patent assignees in a fragmented patent thicket."''
:''"...a firm’s patents, and the patent landscape that emerges, contour and stifle public knowledge over the long run."''
===Dependent Variable and Model===
The dependent variable is the count of the number of follow-on public/peer-reviewed academic publications citing to the original gene paper in a given year.
*The dataset includes years before and after a patent has been granted for a patent-paper pair (58% of observations had a granted patent). Thus, a differences-in-differences approach can be used to measure the impact of control variables before and after a patent is granted.
*A negative binomial model is used for analysis.
Measures of patent thicket used as controls are:
*The number of patents claiming the same gene, measured in 5 categorical variables for 1, 2-4, 5-7, 8-10, or 11-20 patents.
*Fragmentation of patent-gene ownership:
**A Herfindahl index for gene ownership fragmentation is defined as the sum of the squares of the shares of a gene held by individual owners of patents to a gene;
**For patents that read on multiple genes, the fragmentation index is further calculated over all genes included in the patent.
**The fragmentation index is 1 minus the above indices, so that the increases in the measure indicate more fragmentation
Control variables for patent characteristics are:
*An indicator of whether a patent was in force;
*An indicator of whether the patent was granted in the year of the observation;
*Scope of patent, as measured by an interaction between the patent being in force with patent scope net of average patent scope, where scope is the number of national classes in which a patent is categorized (following Lerner, 1994);
*Strength of patent, as measured by an interaction between the patent being in force with the number of patent claims net of the average patent claims in the data.
*Some specifications include interactions with measures of patent importance (cancer gene, OMIM gene, disease gene).
Other control variables include:
*Year of academic paper publication;
*Number of authors;
*Number of author affiliations (measured by author address);
*Indicator for US author address,
*Indicator for public institution address;
*Indicator for private address;
*Impact factor to proxy for journal quality, a measure published by ISI;
*Fixed effects for paper, paper-age and citaion year.
