A $29 Billion US Troll-Tax or Just Another Statistical Smokescreen?

In the past few days, there has been a disproportionate level of attention given to a somewhat academic study of the costs imposed upon the US economy by purported patent ‘trolls’, or 'non-practicing entities' (NPEs).

The study, conducted by Boston University law researchers James Bessen and Michael Meurer, is entitled The Direct Costs from NPE Disputes, and a working draft is available from SSRN.

As it has been presented in the technology media (see, e.g., US patent trolling costs $29b: study and Patent trolling cost the US $29 BILLION in 2011), the study shows that patent trolls impose a huge burden on innovation, and that this is further proof of our ‘broken’ patent system.  This is great headline fodder (or click bait), but does it really add up?

Reading the full paper by Bessen and Meurer raises, for us at least, a number of issues, concerns and questions which are (unsurprisingly) absent from the bulk of the media coverage.  Here are just a few…

1. If it is indeed true that patent trolls exact a $29 billion ‘tax’ on the US economy, then this is certainly cause for alarm.  But does this figure really pass the ‘smell test’, or is it just too implausible to take seriously?  If it is wrong, then this study is adding to the hysteria around purported problems with the patent system without due cause.  When the figures in the study are stacked up against the total number of technology companies operating in the US, and the total R&D expenditure, it is frankly difficult to believe that the results are a true reflection of reality.
2. The raw data for the study comes from RPX Corporation, a ‘patent aggregator’ which offers ‘defensive buying, acquisition syndication, patent intelligence and advisory services’.  Basically, RPX acquires patents (just like a ‘troll’), but with the stated intent of using them to remove trolls from the market, and to assist the victims of trolls.  Companies pay to become RPX ‘members’ for not-insubstantial fees.  The survey data used in the study is from RPX clients, or other associated firms, and the broader litigation data is from RPX’s own database, selected and compiled according to its own criteria.  While the study’s authors are keen to point out that RPX had no say in how they used the data, or presented their research, they are nonetheless completely dependent on information that is unlikely to be free from selection bias.
3. There is no differentiation in the study (because there is no differentiation in RPX’s data) between different kinds of NPE.  RPX uses the term to encompass patent assertion entities (i.e. organisations whose primary business model is to acquire and assert patents in order to obtain settlement and license fees) as well as individual inventors, universities, and non-competing entities (i.e. operating companies asserting patents well outside the area in which they make products and compete).  Not all of these entities are patent ‘trolls’.  Indeed, it may be that the true ‘trolls’, i.e. those entities which make absolutely no contribution to innovation within the economy, are in a minority.
4. The statistical methods employed in the study are opaque, and lacking in any sensible or meaningful assessment of error or confidence.  For all we can determine from the published data, the number ‘$29 billion’ could mean ‘anywhere between $100 million and $100 billion’.  Or it could mean something else entirely.  People who perform these kinds of analyses need to start to understand a simple fact: if you cannot establish the ‘error bars’ on your results, they are meaningless to a statistically-informed reader, and worse than meaningless to the lay person, who may treat them as accurate and precise.

Having said all that, the general project represented by the study is an entirely worthwhile enterprise.  The patent system is supposed to provide a net benefit to society, by providing incentives for innovation.  At the same time, however, it awards monopolies in favour of private interests and, as such, it inevitably imposes costs on the economy.  Ideally, it would be possible to conduct a meaningful cost/benefit analysis, to ensure that the patent system is, in fact, working (roughly) as intended and, if it is not, to assess the reforms that might be required in order to achieve a better outcome.

The Boston University study is, in this context, a step in a productive direction.  What we need now is more, and deeper, analysis.  How precise, and how plausible, is the $29 billion figure?  Can we get better data to produce a more meaningful result?  Can we distinguish real ‘dead weight’ costs to the economy from productive transfer costs associated with an innovation marketplace?

Patentology would love to know what ‘real’ patent trolls really cost the economy, and whether it is worth tampering with a system which actually benefits many stakeholders in order to reduce those costs.  Unfortunately, this study does not provide us with the answers to these critical questions.

DOES A $29 BILLION COST ADD UP?

Bessen and Meurer contend that the direct costs of NPE patent assertions – $29 billion in 2011 – compare with a total US business spend on R&D in 2009 of $247 billion.

Their point is, we assume, that NPEs, including patent trolls, are burdening the innovation system to the tune of greater than 10% of the total amount actually invested.

If true, this is shocking.  But can it possibly be true?

Firstly, however, we note that the $247 billion figure selected by Bessen and Meurer is the lowest they could have chosen as representative of the US R&D investment.  According to the US National Science Foundation (NSF) figures from which they quote, $247 billion covers only the funds actually expended by businesses from their own pockets.  Once other funding sources are included, the total business R&D expenditure in 2009 was $282 billion.  Universities, colleges and other nonprofits (i.e. potential NPEs) spent a further $72 billion.  Further Federal Government spending amounted to around $46 billion, for a grand total of about $400 billion.

Even so, $29 billion would be a significant overhead on this investment.

The figures in the study indicate that this cost is incurred by around 5,800 defendants, 3,400 being small/medium companies (under $1 billion annual revenue) and 2,400 being large companies (above $1 billion). 

According to NSF statistics, in 2006 there were around 6 million firms operating in the US, of which 99.7% were small businesses (less than 500 employees), although the 0.3% of large business accounted for about 50% of the nation’s employment.

So, if the study’s conclusions are correct, less than 0.1% of US companies are bearing NPE-related patent defence costs amounting to around 10% of the total national R&D spend by all businesses.  If this were true, surely the tiny proportion of businesses bearing this disproportionate cost would be driven out of the market, and NPEs would effectively be killing the goose that lays the golden eggs?  Maybe this is happening, although we are unaware of any hard evidence to support it.

Occam’s razor says – barring very persuasive evidence to the contrary – the most likely explanation for this improbable scenario is that the numerical results from the study are wrong.  And, as discussed further below, the statistics in this study are just not that persuasive.

NOT EVERY NPE PATENT ASSERTION IS A ‘COST’ TO THE ECONOMY

One significant problem with the study – and more particularly with the way it has been reported – is that it makes no distinction between different categories of NPE.  As noted already, the term is used to encompass a range of different patent-holders, including not only patent assertion entities (i.e. organisations whose primary business model is to acquire and assert patents in order to obtain settlement and license fees) but also individual inventors, universities, and non-competing entities (i.e. operating companies asserting patents well outside the area in which they make products and compete).

Within the context of the 34-page academic paper this is fine, since it is at least clearly stated (although the study would be much more interesting if it distinguished between these different types of entities).  However, when it is reported as ‘patent trolls cost US companies at least $29 billion in 2011’, then it completely misrepresents the true situation.  Do Americans really begrudge individual inventors asserting their patent rights over companies that are using their inventions?  Are they really concerned that universities might be successfully transferring technology developed with (possibly public) research funds to commercial enterprises at a fair price?  Should companies that happen to have developed technologies with wide-ranging applications outside their own fields of commercial interest really be labelled ‘trolls’ because they do not believe that others should be able to use those technologies for free?  And while lumping CSIRO in with the patent trolls might bring joy to Ars Technica, it is unlikely to impress the Australian taxpayers who funded the organisation’s Wi-Fi research in the first place.

More importantly, however, not all costs associated with NPE patent assertions are costs to the economy.  To the extent that the patents in question represent innovations resulting from genuine investment in R&D, and/or that licensing income resulting from assertion is reinvested in further commercial and research activity, the costs are transfer costs, not ‘dead weight’ on the economy.  When companies that have not invested in R&D relevant to their business pay license fees to companies that have, this represents a functioning market for innovation, or intellectual assets.

Bessen and Meurer acknowledge that, at least historically, NPEs have played a positive role in assisting inventors to monetise patent assets where they might otherwise not have had the private resources to do so.  However, their overarching thesis seems to be that the cost incurred due to patent trolls is now an excessive burden on the system.

This might be so, but we currently have absolutely no idea how the alleged $29 billion ‘cost’ is divided among genuinely troll-like assertions and other, more defensible, enforcement actions.

SOURCES OF DATA AND STATISTICAL METHODS

A further major concern is that the results of the study involve extrapolations on a massive scale.  Extrapolation based on a small sample of an overall population is a valid statistical method, so long as the sample is sufficiently strongly representative of the whole.

In this case all of the data comes from defensive patent aggregation company RPX Corporation.  A survey was conducted in which companies were invited to participate, all being clients of RPX, or non-clients with whom RPX has some other relationship.  They are therefore, by definition, respondents with a particular pre-existing interest or concern with defending themselves against NPE lawsuits, since that is the primary service provided by RPX.

RPX also maintains a database of NPE-related litigation data.  This is obtained from publicly-available US district court records, however it is RPX’s own researchers who decide which litigants qualify as NPEs, and therefore make it into the database.  Over the period covered by the study (2005-2011) the total number of defences against NPE lawsuits, according to RPX, was 20,565. On average, each company from this complete population defended only 2.2 NPE lawsuits.

We presume that the survey invitations were extended mainly to companies with particularly high levels of experience defending NPE law suits.  This is evidenced by the statistics: survey respondents comprised 82 companies which defended an astonishing 1,184 NPE lawsuits in the period from 2005 to 2011, i.e. an average of 14.4 lawsuits per company, or over six times the population average!  In addition to being selected by RPX, the respondents were also self-selecting – 250 companies were invited, but about two thirds (168) elected not to respond.

Of the 82, only 46 companies responded to questions regarding patent assertions by NPEs which did not result in litigation.  This data is therefore even more selective.

It is clear from the above analysis that the survey sample used as the basis for the study is not representative of the total population.  Any extrapolation from this sample thus needs to be conducted with great caution!

The strategy employed in the study to extrapolate from information obtained from 82 respondents to a set of final figures for the total cost of NPE patent assertions was essentially to work out typical costs of defending NPE litigation among the sample group, and then extend this based on the total number of NPE lawsuits, i.e. to assume that the 5% of cases covered by the survey are representative of the other 95%.  By way of ‘nuance’, companies were divided into ‘small/medium’ and ‘large’, based on revenues, with the cut-off set at $1 billion per annum.  There were notable differences between the two categories, although one assumes that in reality there is not some magic ‘jump’ in litigation costs once revenues hit $1 billion!

For the non-litigation costs, the study’s authors assumed that the costs incurred by the 46 respondents, relative to the costs of litigation within the sample group, were also broadly consistent with the overall population.

These are significant assumptions, especially given the inherent bias in the sample set.  Where standard errors are provided for the statistical measures (which is not often) they are large – typically between 25% and 50% (or more) of the corresponding mean values.  Furthermore, it is very difficult to interpret these errors, given that the cost distributions are highly skewed, with lower costs being more heavily represented, but with the average costs relatively high due to a small number of extremely expensive cases.

Bessen and Meurer recognise that there are potential biases in their sample, and have therefore attempted to verify their results by comparison of their costs with figures obtained in other surveys, such as the American Intellectual Property Law Association (AIPLA) bi-annual member survey.  They find that the average costs of litigation are broadly consistent, however it is not at all clear that this addresses the kinds of bias likely to be present in the self-selecting survey group of troll-targets.  It does not, for example, alter the fact that a small number of high-cost defences are skewing their mean values upwards, or that extrapolation of results by a factor of twenty has the potential to introduce similar multiplying factors to the uncertainties in the data. 

Statisticians will recognise that the mean alone is a weak indicator of whether two independent samples are drawn from the same population, especially when – as in this case – the distributions are known to be highly skewed.  In other words, the authors’ efforts to verify their assumptions and cross-check their data are a valiant effort, but they are far from adequate.

CONCLUSION

It is a pity that the authors did not carry statistical errors through all of their calculations, and provide error bars, or preferably a confidence interval, on their final figure of a $29 billion cost to US companies from NPE assertions in 2011.  We suspect, however, that a 90% confidence interval would cover a range from something very much lower than $29 billion, up to a much higher figure that all would agree is ludicrous.

The simple fact is that if the survey sample were larger, e.g. if the 168 non-responding invitees had participated, any resulting changes in the calculated mean costs would impact directly and proportionally on that final $29 billion figure.

In the absence of a full statistical analysis, resulting in a statistically-meaningful range within which the actual cost can be said to lie, with some quantitative probability, we would go so far as to suggest that the $29 billion figure is essentially meaningless.  It is almost certainly not correct, and we do not even have an indication of the size of the surrounding range within which the true figure may lie.

However, as we said up-front, Bessen and Meurer’s study is a step in the right direction.  It would certainly be valuable to determine the real cost imposed upon the US economy by patent trolls.

And we agree wholeheartedly with a number of their conclusions.  For example, their suggestion that policy reform should address troll-like behaviour rather than merely status as an NPE is, in our view, spot on the money.  Trollism is presently a unique feature of the US system, and there are a number of reasons why this might be so which Bessen and Meurer identify:

1. too many patent claims of indeterminate scope, especially in the more ‘abstract’ fields, such as business methods and some types of computer-implemented inventions;
2. disproportionate awards of damages, making trolling a potentially highly profitable venture; and
3. reduced financial risk associated with assertion of questionable patents, due to inadequate use of fee-shifting, i.e. whereas in other jurisdictions (including Australia) courts will normally order the losing party to pay the winning party’s costs, this is the exception rather than the rule in the US.

There are a few other features of the US system that we would add to this list.  But we will save these for a subsequent article another day.

Tags: Economics, Innovation policy, Patents and society, Trolls