What is the Probability of Receiving a US Patent?

Following my post last week, I confidently proclaimed at a conference yesterday that the true abandonment rate was about 9%. I was--kindly, thankfully--corrected by the target of the comment, who just so happens to be a co-author of a working paper called What is the Probability of Receiving a US Patent?, by Michael Carley, Deepak Hegde, and Alan Marco at the US PTO. Marco is chief economist, and Hegde is at NYU Stern Business School, but also an Edison fellow at the PTO. I had read the paper before, but hadn't focused on the abandonment rate aspect.

They use data that's generally unavailable to the rest of us (some for good reasons, some for not - this should really be made public) to trace the outcomes of each initial patent application.  This differs from the methodology of Cotropia, Quillen & Webster, who cannot trace each application, but can merely count them up from reports for a given year.

The differences between the studies are not that important for the key measure of that study - how many patent applications are there? Cotropia, et al., want to adjust a report of applications each year so they subtract out continuations to get the real count of new applications.

But it turns out that that this matters for abandonments, because today's abandonments are yesterday's (or four years ago's) applications. And so you have to trace all the way through to get the real number. Indeed, we won't know the real abandoment rate for applications this year for many years - the Carley, Hegde, Marco study only looks at applications between 1996 and 2005, to ensure that most of the cases have resolved. It turns out the real abandonment rate is in the 30% range, depending on technology.

Now, to be clear, Cotropia, et al., never claim otherwise. To quote from their study: The number of Refiled Continuing Applications in 2014 (260,265) was 91% of the number of applications reported as abandoned in 2014 (287,526)." They were clear about how they calculated the total, though one can see (I hope) how one (me) could have misread this to be reporting an abondonment rate.

So, what do we know about the probability of getting a patent? First, it declined from filings in 1996 to 2005. This long supports a theory I've had, which is that our litigation explosion today is caused in part by overgranting (particularly in software) in the 1990s. There just wasn't a lot of prior art at the time, and it was hard to search, and so we got more patents.

This is borne out by the technology differences. Allowance rates dropped steeply in computers in 2000 (more steeply than other fields, though they all dropped), picked up a bit for a year, and then continued to keep dropping.

Second, technology affects probability greatly. More than 80% of initial electric/electronic patents eventually issued (which means a real abandonment rate of about 20%, though some of those cases are still open). For drugs, the probability of getting a patent drops to a low of 60% (which means a real abandonment rate of 40%). Computers and communication (which includes software) falls in the middle, at 69%.

The paper is chock full of details, showing which fields use continuations more often, which fields get patents on the first try, versus later applications, correlations with factors that might affect applications, like GDP, foreign v. US applicants, and more. They even show where in the process patents are most likely to be abandoned (hint, after final rejection, not appeal).