Guest post by Michael Risch, Professor of Law, Villanova University School of Law. Professor Risch also recently joined the Written Description blog as a regular author. The full article, forthcoming in the Iowa Law Review, is available here.
This follows my last guest post about my article A Generation of Patent Litigation and is the third in my Patent Troll Myths series of studies of the ten most litigious NPEs from 2000-2010. To recap, I gathered data on 1313 randomly selected patent cases distributed over a 25 year period in roughly the same proportion as the 917 cases filed by the most litigious NPEs over the same time period. The number of cases grew substantially starting in 2004. This led to 792 nonNPE patents versus 352 NPE patents, which indicates that the NPEs asserted the more patents per case. This article expands on the last one by looking at the technology categories for each patent as well as the initial source of the patent that wound up in litigation.
The results of my analysis confirmed much of what we already knew, but the data allowed me to demonstrate it. In short, patent litigation is a complex system made up of at least three layers: inventors and their assignees, patent plaintiffs, and technology. There are surely more layers, like defendants and licensees, but these three layers have some of the most relevance to patent quality. The problem is that most of our discourse examines one—or maybe a second—layer at a time, but rarely all three. Thus, we have NPEs versus producers, software versus pharma, individuals versus corporations. We rarely have data that includes all three of these layers in one place. When they are included, they are usually considered control variables rather than additional explanatory measures.
This study seeks the interconnection between the layers. I’ll give a few examples in this post, but there is a lot more detail in the paper.
Consider, for example, initial assignees. Both the random plaintiffs and the most litigious NPEs obtained a majority of their patents from product companies. And a substantial percentage of those companies were public for both groups (though about twice as many for the random plaintiffs). But not all product companies are created equal. Among the random companies, the initial assignees were bigger, better funded by venture and stock market investors, had more employees, and earned greater sales.
What does this mean? Any quality differences we might see between the NPEs and random plaintiffs might relate to the size and types of companies obtaining those patents. It also means that the technologies we see the NPEs enforcing might be the types of technologies that require less investment.
In fact, we do see different types of technologies. The following table shows the top five patent classes for each group with a comparison to the percentage held by the other group. The differences are stark. The paper shows the top 13 categories, and shows that 66% of the NPE patents are in the top 13 classes, while only 30% of the random group’s patents are in the top 13 classes.
|Top NonNPE Classes||Top NPE Classes|
|362||Illumination||4.17%||0.28%||360||Mag. Info. Storage||0.00%||8.24%|
|348||Television||3.66%||6.53%||705||Fin. Bus. Meth.||1.01%||6.82%|
|349||Liquid Crystal||2.53%||0.00%||709||Data Process. Trans.||1.89%||4.55%|
When we break down by technology and by plaintiff type (two different layers), we see differences that weren’t apparent before. The graphs below show two categories, e-commerce, which has higher invalidation rates, and electric circuits, which has lower.
In electronic commerce, the nonNPE group saw no challenges – at all. Among the litigious NPEs, however, nearly half of the patents were challenged. When there was a decision on the merits, patents were completely invalidated about half as often as they were held valid. But much of the time, challenges were denied or pending at dismissal.
For electric circuits, however, patents were challenged at about the same rate. But this time it was the nonNPE group that was more likely to reach a decision on the merits – with validation more than twice as much as invalidation when there was a decision on the merits, but also a decision on the merits almost three-fourths of the time. Among the litigious NPE group, however, all of the challenges were denied or pending at settlement, and none went through to final judgment.
These are just two technology categories. The paper compares several others, including optics, chemistry, and medical instruments. It also considers results for different types of software (and for non-software).
As a final test, I ran a series of regressions to test the likelihood that a patent would be adjudicated to have any invalid claim. The full model is presented in the paper, but a few of findings stood out.
First, patents coming from failed startups had the highest correlation with invalidity, regardless of who enforced the patent.
Second, patents left unassigned at issuance were more likely to be invalidated whether asserted by either group (though individual obtained patents fared better when asserted by the random plaintiffs). However, the same was not true of patents assigned to inventor-owned companies. The data does not allow a causal inference, but there appears to be something about inventors starting their own companies that improves validity outcomes later.
Third, once source of the patent and type of plaintiff is controlled for, invalidity differences appear for only some types of technology. This is consistent with the graph I show above, but more rigorous. Thus, for example, cryptography patents are invalidated about one-third as often when we consider the source of the patent and type of plaintiff as compared to just looking at the average cryptography patent without patentee/plaintiff type. On the other hand, optics patents are invalidated at about the same rate, whether or not we consider the source or plaintiff type.
Fourth, these findings continue for software. Though software patents are, on average, invalidated more often than other patents (a finding consistent with other studies), when the type of patentee and plaintiff is considered, whether the patent covers software is no longer statistically significant.
This last point is ultimately the point of the article. When we consider all of the layers of the system rather than just the averages on any one layer, the picture gets far more complex. My data can’t answer every question, of course. After all, I only studied the most litigious NPEs. But even this sample shows the complexity. There’s much more I could write, but I’m out of space. The full article is here if you are interested in reading more.