2 questions re swpat paper

["Jim Bessen" <jbessen@researchoninnovation.org>]

Bob Hunt and I are preparing to release a new extensive revision of our
working paper "An Empirical Look at Software Patents." We will also release
a short, less technical version.

Some criticisms have been raised about the earlier version and these are
answered in the new version. However, Erik Josefsson has encouraged me to
post answers to 2 questions in the interim:

Q1: Is your definition of "software patent" flawed? Wouldn't it be better to
use a definition based on IPC classes?

Answer: Our new version goes through a very careful analysis and validation
procedure. We validate our definition against two databases of patents that
were individually read and classified (one by us and one by independent
researchers). Although our method makes mistakes, its rate of false positive
(patents classified as SW that are not) is 16% and false negatives (patents
not classified as SW that are) is 22%.

We also considered a definition based on IPC classes developed by Graham &
Mowery. Their false positives were 30% and false negatives 74%, much worse.
Also, we ran all of our regressions and tables using Graham & Mowery and
found the same main results, although the standard errors were greater (as
would be expected from a cruder measure). The general problem with IPC
classes is that they concern functional relationships and not technologies.
For example, a class contains inventions involving "visual representation of
data" and this includes software, computer hardware, and pre-computer
mechanical and electrical signs.

So our definition is not perfect, errors can be found, but it is pretty damn
good and better than the alternatives and doesn't seem to introduce any
biases.

Q2: "One empirical result in the paper reveals a statistically significant
 positive correlation between patents and R&D." Doesn't this refute your
theory?

Answer: This is sophistry. Our evidence shows that software patents and R&D
are substitutes. The incentive theory holds that they are complements, so
our evidence rejects the incentive theory (that less costly patents will
produce more R&D).

This question reveals a lack of understanding of the basic
economic concepts of complements and substitutes. Consider that coffee and
tea are substitutes. Some households will consume more of both (e.g.,
wealthy households) and some will consume less of both (e.g., poor), so they
will be positively correlated across households. Nevertheless, they are
still substitutes.

The substitution effect is observed relative to price changes, that is, a
decrease in the price of tea relative to the price of coffee will lead
households to consume less coffee RELATIVE to the amount of tea they
consume. Wealthy households may still consume more coffee than poor
households, but it is less than they consumed earlier. The relative
consumption can be more clearly measured by looking at ratio of coffee
consumed to tea consumed in each household. Alternatively by the share of
expenditures spent on coffee (this is a little more complicated)--if
coffee's share decreases with a drop in the price of tea, coffee and tea
cannot be complements.

We are, in effect, saying that the price of SW patents decreased over time.
Large companies have both more R&D and more patents overall. But firms that
relatively increased their use of software patents (and therefore had a
relatively larger drop in their average cost per patent) should have had an
increase in the share of expenditures spent on R&D if patents & R&D are
complements. This was not the case; they are substitutes.

Conclusion: the incentive theory can be rejected for software patents. The
simple intuition is that firms that were better able to take advantage of
software patents should have increased their R&D if SW patents provided an
incentive. Instead, they decreased R&D intensity.

--Jim Bessen