Duane Shelton and Sam Monbo WTEC STI 2012 Montreal September 6-8, 2012 Input-Output Modelling and Simulation of Scientometric Indicators: A Focus on Patents
Outline Patents as important indicators of science Best input-output models, systems with memory Simulation and validation on past data What-if games for the future Conclusions
Why are Patents Important Indicators? Patents are a proxy for the output of applied research, as papers are for basic research Stakes are high: As professors publish or perish, inventors seek serious money from inventions Lots of data is available, sometimes more than for papers There have been big changes in patent regimes
Applications to patent offices of Japan, USA, Europe, S. Korea, and PRC. From Zhou & Stembridge, Patented in China, Thompson Reuters World IP Today, The Chinese took the lead in Big Changes: China is Now Rising in Patents, a New USPTO Regime and Switch to “First to File”
Q: Why the US Patent and Trademark Office? A: All Inventors Like USPTO patents This is the first of 6 pages from the most valuable patent ever. Bell lived in Scotland and the US, but did some of his best work in Canada. Note the 3 weeks at USPTO.
Data from the USPTO is an International Indicator There are now more foreign applications (and grants) at the USPTO than domestic ones. US data needs to be separate, because of its home advantage.
Where Do Those Foreign Applications Come From? About 99% come from the 40 “OECD Group” countries
Input-Output Models Obviously resources are needed to invent and pay for patent applications Which national resource indicators are most strongly correlated with patents: Overall “GERD,” gross expenditures on R&D GERD sources: Government, Industry, Abroad (funding from abroad), Other GERD spending: HERD, BERD, Non-Profit (other than universities), and GOVERD (government labs) Number of researchers
Triadic USPTO Capital vs. Labor GERD Researchers Funding Components Industry Government Spending Components HERD BERD Same Year Correlations: Patents vs. Inputs
Step-Wise Regression for USPTO Patent Share Step Constant Ind T-Value P-Value Gov T-Value P-Value Gov_ T-Value P-Value HERD 0.29 T-Value 1.93 P-Value S R-Sq R-Sq(adj) US omitted because of home advantage
The Best Same-Year Predictor of USPTO Applications is Industry R&D. US is Omitted. USpatents = IndustryR&D p = 0.000, R 2 = 92%, Good, but, can lags improve this?
Not Patents, but Papers From the SCI in 2007 The correlations are very high, and there seems to be very little delay in getting published papers out from R&D investment in.
There is a very high correlation between industrial R&D and USPTO applications (in 2007). And it gets better with a lag of 5 or so years, but the peak is broad. Lags are Important for Patents, Even for Applications
USPTO applications in 2010 vs. industrial R&D in 2002 on a log-log scale. The U.S. would be at (5.38, 5.02) Example of a Lagged Regression
USPTO Input and Output Disposals (grants + abandonments) are the total output. This is an unstable system: arrival rates > departure rates. How can these additional delays be modelled?
The Dreaded Instability Curve for the M/M/1 Queue Traffic intensity is arrival rate / service rate. Delays increase without bound when this goes to 1.
Foreign Industry R&D Funding U.S. Industry R&D Funding Patent Office Applications Backlog QueueServer Grants Multiplier Delay Multiplier Delay Abandonments GPSS/H is used for simulation Model for Applications and Grants from R&D Inputs
Industry R&D is input. Dataset is OECD Group of 40 countries. Fixed ratio based on 5-year average, then actual R&D used. Simulation Results from US and Foreign Sources —Validation to 2008
The actual service rate is used in the queue for this historical data. Thus they should agree closely after the initialization prior to Simulation Results--Continued
The USPTO says their “backlog” is about 750K, but this only counts applications awaiting first action. This estimate uses USPTO data, adds allowances, and deducts design patents. Average Queue Length and Delay Time Usually Do Not Agree So Well
“Traditional Total Pendency” includes time waiting at the USPTO plus time the inventor spends preparing a resubmission. Thus the single queue model is a simplification. This graph suggests that the USPTO fudges this calculation. Delay Time in the USPTO. Big difference here, why?
What-If Results The model was validated through 2008 It can be used to forecast through 2014 at least, since R&D data is available to 2009 But, first there have been big changes at the USPTO in
Recent Changes at USPTO David Kappos, Director in 2009 Patent reform of 2011, “America Invents Act” First to file (instead of first to invent) International harmonization Better access to fees to hire examiners (Congress has a habit of raiding these)
Big Changes in 2009 The model was built through 2008 for an unstable system. If this can be kept up, it will stabilize the USPTO.
Model Allows Forecasts Based on Scenarios. Applications Can be Accurately Forecast Until 2017 #1 Same disposal rate as 2008 #2 Improvement as done in
One trend is that US applications are being crowded out by foreign applications. In 2008 for the first time, there were more foreign grants than domestic ones. The trends are slow, so it will be a long time before no Americans get US patents. The model permits tracking of US vs. foreign applications. Further Work Ideas: Extrapolation to 2017, Plus
Conclusions The simulation shows that lagged R&D investments as an input can accurately model patent applications, grant outputs, and total backlog Until 2009 USPTO was an unstable queue with arrival rate > service rate However, its performance only gradually deteriorates— there is no imminent danger of collapse The simulation allows what-if games, complicated by the big changes in 2009 Further work: forecasts to 2017, US vs. foreign patents, time delay discrepancy
References usptoqueue An queueing analysis of the USPTO usptoqueue Publish or Patent: Bibliometric Evidence for Empirical Trade-offs in National Funding Strategies. JAIST With L. Leydesdorff. Publish or Patent: Bibliometric Evidence for Empirical Trade-offs in National Funding Strategies