« Prove It | Main | High Stakes High Definition »

August 26, 2008


The application of statistics is relatively simple. Statistics is taught at college for many fields. But properly applying statistics seems to elude many. A recent patent reexamination analysis by lawyers is exemplary: a small sample size of biased data, rendering it rather meaningless. But the statistically-challenged authors reported the results as conclusive. Another study, on civil lawsuit settlements, suffers the same flaw. There at least the authors admit the data base as flawed, but regardless paint a brave face on tainted data.

The civil lawsuit study, reported in the New York Times and ABA Journal, purports to show that, statistically averaged, it's better to settle than go to trial. The problem with the data is intrinsic to the problem studied: the study ignored settlements, because the data was not available, and would have not been statistically helpful anyway. Most settlements remain confidential. So, all the data is based upon suits that went to trial, then used hindsight of settlement offers before trial. It would, of course, be impossible to say, for settled cases, what going to trial would have meant, so that data is conveniently ignored.

The results were that plaintiffs got the short end of the stick by going to trial more often than defendants. The report points to contingency law firms for plaintiff error, and lack of insurance for defendants. Don't give any of that credence.

As with many exercises in cumulative averaging, the numbers ignore reality. Civil lawsuits are a heterogeneous set. They differ by type and circumstance so broadly as to make an apples-and-oranges comparison while pretending to be looking at only raspberries. Fruity and fruitless.

The most interesting point made regards risk skewing: perception of risk based upon situation. People expecting a gain are more risk adverse, but those facing loss more willing to gamble. Risk skewing partly explains conservatism and terrorism.

What isn't so easily explained, other than lack of proper education, itself apparently a rarity, is why there are so many are so statistically challenged folk putting up such worthless numbers in the first place, and then for those numbers to be given credibility by admiring publication. Chances are somebody got paid to fiddle numbers, and somebody got paid to report them. Statistically speaking, stupidity and corruption appear a winning combination.

Posted by Patent Hawk at August 26, 2008 11:07 AM | Litigation


I'm not sure I agree with the statement, "People expecting a gain are more risk adverse, but those facing loss more willing to gamble."

It's still a change in $$. If a case is worth +$100 to me or -$100 to me I would be just as unlikely or likely to gamble.

I think it's more like this - if there's little at stake, you'll gamble. If there's a lot at stake, you'll probably settle.

Thus, the statistics will show cases where there is little at stake. If I may lose $50k of my $500 million company, I'll gamble... if I may lose the company, I'll settle.

Posted by: NJ NY Patent Attorney, Lawyer at August 26, 2008 2:54 PM

This evokes the completely bogus analysis by Professor John R. Thomas on the (allegedly high) grant rate of the first 100 published patent applications. Thomas neglected to point out that most of the first published applications had antecedent patent family members (ie, the first 100 were primarily not first filed cases). Biasing one's sample can lead to unusual results.

See IPBiz posts

Do the published applications of 2001 tell us about patent grant rate?
(July 31, 2007)

More on patent grant rate; the USPTO is NOT a rubber stamp
(August 2, 2007)

Posted by: Lawrence B. Ebert at August 27, 2008 4:41 AM

"People expecting a gain are more risk adverse, but those facing loss more willing to gamble."

I found this statement interesting but incomplete, too.

In the US system the infringement P will generally put at risk litigation costs (a certain loss) and its patent rights, since the validity of the patent will be called into question as a matter of course. The P's expected gains may be 8 or 9 digit damages, injunction (now less likely), enforced royalties, and judicial finding of patent validity, thus strengthening the monopoly and protecting market access or share.

The infringement D will generally put at risk litigation costs (again, a certainty), the 8 or 9 digit damages, and loss of market access or share if the patent is found to be valid.

Seems to me the first level of analysis for both parties will be to assess the ratio of one's bankroll to the expected lawyers' fees, call it L/B. Contingency fees and the American rule against loser pays, as always, tilts the field toward the P. If the L/B cannot work for either party, then the pressure to settle is very strong regardless of risks or gains, perceived validity of the patent vel non, or merits of the case.

Assuming, your L/B suggests you can afford the fight, the second level of analysis would be a comparison of the opponent's L/B to yours. If the litigation is going to burn 90% of your assets win or lose and only 0.1% of your opponent's, your position as P or D is a huge factor. As a P you can hope to recoup the 90% loss with damages; as D you can't so you lose, win or lose.

So I'm not quite sure where I'm going here, but I think it's to a position that what is a win for the P and what is a win for the D are so vastly different that any assessment of "risk skewing" must be done separately for D's and P's. A P with a strong case (strong patent) and good L/B will be willing to risk more than a P with a weak case and poor L/B, and yet both may be less risk adverse than a D facing a crippling L/B, win or lose.

Posted by: Babel Boy at August 27, 2008 8:26 AM

"People expecting a gain are more risk adverse, but those facing loss more willing to gamble."

This statement in the NYT was about a reference to some simple psychological experiments. In short, if you have 2 people in Vegas and the first person is down a grand while the second person is up a grand and you offer each of them to bet 1000 on a flip of a coin, the person down a grand will be more likely to take the bet (trying to erase their losses), while the person up a grand will be more likely to keep their winnings. Or so the theory goes.

Posted by: jonesie at August 27, 2008 12:05 PM

Risk skewing is my own term for what is a well-recognized psychological dynamic. The NYT article gave an example, but pointed to no study on the topic.

While not about risk skewing particularly, recommended reading about risk evaluation is “Against the Gods: The Remarkable Story Of Risk,” by Peter L. Bernstein.

Alas, I don't know a good book to recommend educating about proper application of statistics, particularly in recognizing inherently bad data sets.

Posted by: Patent Hawk at August 27, 2008 12:53 PM

What's that old expression: "There are lies, damned lies and then there are statistics."

Posted by: EG at August 29, 2008 4:13 AM