Saturday, October 25, 2014

Empirical IP Research Conference: copyright

Effect of Copyright Infringement on Creative Incentives

Facilitator: Chris Sprigman (NYU)

Joel Waldfogel (Minnesota Carlson School of Management)
Topic: The strengths and limits of the natural experiment methodology to explore the effects of piracy on both industry output and creative incentives, Commenting on: Brett Danaher & Michael Smith, Gone in 60 Seconds: The Impact of the Megaupload Shutdown on Movie Sales (2013), Christian Peukert et al., Piracy and Movie Revenues: Evidence from Megaupload, A Tale of the Long Tail? (2013)

Do the rewards/incentives created by IP promote welfare? Welfare is consumer surplus plus revenue minus costs of creation/production/distribution.  Incentives depend on profit, not welfare.  Rather than “is stealing sapping our revenue,” ask whether enforcement actions like HADOPI or the Megaupload shutdown stimulate revenue.  Important question for big actors like MPAA, but relevance for creative incentives is indirect.

Both papers use Megaupload shutdown as discontinuous variation on feasibility of copying and ask about effects on legal purchases: not about profits/creation but revenue.  Want a measure of change in movie copying to isolate its impact on purchase. Not observed, so use country as proxy. High-copying countries like Spain provide a large experiment and lower-copying countries like the US provide an implicit control.

But: If revenues and costs are both falling, it’s not clear that digitization sapped incentives. What’s happening in the motion picture industry?  Digital camera costs falling, even for big budget movies.  Huge growth in production.  Supply responses take time; we need to distinguish between “is the Man getting paid?” and “is copyright fulfilling its function?”

Christopher Buccafusco (Chicago-Kent)
Topic: Translating Piracy’s Effect on Sales into Piracy’s Effect on Creative Incentives, Commenting on Commenting on: Brett Danaher & Michael Smith, Gone in 60 Seconds: The Impact of the Megaupload Shutdown on Movie Sales (2013), Christian Peukert et al., Piracy and Movie Revenues: Evidence from Megaupload, A Tale of the Long Tail? (2013)

Deep heterogeneity in the markets; effects on incentives may themselves be variable.  In the second paper, the Megaupload shutdown is pretty good for blockbusters and not helpful/possibly even harmful for smaller movies that rely on word of mouth for success. These aren’t simple or closed systems.  Need to be able to study the effects as they move around. 

Also, incentives for whom?  Who are the producers we care about? That can affect kinds of works created, who creates them, amount that gets invested in various groups, etc. Extrinsic v. extrinsic motivation: complicated relationship between incentive to create and output of products.  These won’t show up in large scale econometric studies and require different methodologies to answer.

What else don’t we know? These aren’t papers about the costs of production, so we need that. There’s not much about overall welfare.  What questions can experimentalists help answer?  Ultimately these are individuals, who might be better studied in labs or qualitative interviews.  To what extent do behavioral biases affect the kinds of creativity we get?

Peter DiCola (Northwestern)
Topic: Generalizability of Studies in One Creative Industry, Commenting on: Rahul Talang & Joel Waldfogel, Piracy and New Product Creation: A Bollywood Story (2014), Joel Waldfogel, Copyright Protection, Technological Change, and the Quality of New Products: Evidence from Recorded Music Since Napster, 55 J.L. & Econ. 715 (2012)

Waldfogel: estimates of quality of new music since Napster, using within rating system variations.  Data from 2000-2007 (end of decade ratings tend to be harsher on most recent music).  Music did not get worse after Napster by various quality measures, including airplay. 

Bollywood: revenue crash and decline in movie production 1985-2000, adjusted for quality.  But DiCola sees a v shape: revenue craters and comes back in 1992-93, making the institutional story more complex.  Unauthorized use: the causal claim comes from a historical narrative, but there’s VHS and cable. VHS proliferation starts in 1985; cable piracy starts in 1992.  Pro-studio change in 1994-95, and enforcement complaints continue through 1997.  Cable piracy starts when things start to recover.  Need a more precise timeline story, in order to generalize from one industry to another and to create a more formal model.

Pamela Samuelson (UC Berkeley)
Topic: Relevance of Empirical Research to Policymakers, Commenting on: National Academies Of Science Report, Copyright in the Digital Era: Building Evidence for Policy (2013)

Need more studies like DiCola’s recent studies of musicians.  Congress/the Copyright Office are willing to hear about real creator communities.  Questions about how to get more people to register copyrights/record transfers are empirical questions.  Are the fees too high? Would differential pricing help?

Katherine Strandburg: Transformativeness as a potential place for empirical data. Courts might be more receptive than Congress.

Samuelson: could identify target audiences for particular work; new audience makes fair use more plausible. Not everything works that way, but some targeted data could help.

McKenna: we never would have had iTunes without Napster.  Second- and third-order effects of legal rules. When you frame “the effects of getting rid of Megaupload” you need to think about the other effects of the legal rule that took down Megaupload, and that can’t be limited to Megaupload.

Talang: industries can recalibrate themselves.  Do we become too quick to offer a regulatory response?  Spotify: new revenue opportunities; labels are appropriating more of that v. the artists.  Will the market be able to work itself out?  India: ultimately people chose quality over free.

Sprigman: porn industry. Used to depend on the feature; now there’s a shift to the live cam/performance, which can’t be copied. 

Talang: growth in concert revenue in the music industry is another example: it’s an experience to be sold. 

Sprigman: industrial organization issue.  Slate article on Mindgeek and pornography—they own content producers and tube sites with a ton of piracy, including of content owned by the company that owns the tube sites.  Piracy takes from one hand and gives to the other.  Piracy drove down the price of the industry players they acquired; innovative in revolting ways.

Lunney: we haven’t mentioned software: is copyright working well there?  If so, why or why not?

DiCola: the calculations about decay in appeal of music over time would also be interesting to do for videogames.  There are people who use old simulators but most are interested in the next edition.

Samuelson: Breyer’s Uneasy Case for Copyright focused on books, but a little on computer programs. Software isn’t one industry but many; different mechanisms for capturing value, a lot not on copyright—complements, first mover advantage, updates, customization, ads.

Sprigman: platforms v. hosted games—different vulnerabilities to piracy. There must be something that the audiences want that sustains the platforms.

Bechtold: DRM on platforms makes it inconvenient to play pirated Xbox games; inconvenience and time constraints make buying easier, and interactive/online elements are also harder w/pirated or even secondary market purchases.

Strandburg: videogames are interesting because of how people consume them/potential substitutability with other works of entertainment that one might consume, e.g. board games.  What happens when you don’t have certain kinds of protections?

Copyright breakout session: Investigating Copyright Infringement Standards
Facilitators: Ben Depoorter (UC Hastings) & Paul Heald (Illinois)

Depoorter: Cognitive biases may affect infringement inquiries: confirmation bias, jury group effects, hindsight bias, etc. Applications in music: gaps between music theory and law. Some papers argue that music is different, similar story told about pictures (that’s me!).  Auditory sensation, time, musical elements lay observers find hard to detect; musical language is intuited but not understood; still we ask lay listeners to make the determination, but they consider performance characteristics rather than compositional ones.  Jamie Lund’s interesting empirical work: lots of room for false positives given how the cases are litigated.  What is to be done?  ABKCO v. Harrisongs: students look for similarities in My Sweet Lord and He’s So Fine; judge did it right by identifying similarities in musical forms.  Improvements: Lemley suggests increasing the role of experts in improper appropriation test; Pam Samuelson suggested reversing the test and looking first for improper appropriation and then to copying.

Heald: Laroche: mathematical approach to distance between a piece of music and an allegedly infringing piece—how many moves do you have to make to get from one to another.  When they map distance results to songs on the Columbia/USC music infringement database, they don’t get much consistency with the infringement findings, but that doesn’t mean their model is wrong.

For further research: redo Lund’s experiment with composers instead of lay listeners.  If © is a tort, we need a model for determining whether someone w/ a property right has suffered actionable damage.  What we ought to care about is whether composers are behaving reasonably when they borrow from other composers.  Reasonableness doesn’t require perfect rationality, just whether the appropriator is within the normal bounds of musical appropriation norms. 

We can also test for what composers think the law is.  A lot of compositional norms are distorted by an erroneous perception of the law.  Organists: is it ok for them to buy a piece of music and copy it to lay it out on the keyboard so they don’t have to turn pages since they are busy with hands and feet? Some think they can’t do that so they buy 8 copies.

Samuelson: role of experts.  Long period when experts were considered inappropriate for certain issues; are experts good/bad idea?  Also ideas about getting courts to move away from total concept and feel.

Jeanne Fromer: norms within areas differ—appropriation artists think differently than other groups. Experts about what?  Norms among artists; consumer understanding; something else?  Collecting data is hard because it’s hard to tell whether courts or judges are doing the same thing, or even the same thing from case to case—it’s not even a multifactor test.

Beebe: cases are borderline uncodable. Confirms slipperiness of concept of similarity in human cognition. 

Heald: efficient results require incredible flexibility of the infringement test in order to deal with the otherwise huge/rigid rights granted by a valid copyright; thus he’s not confident that much rationalization is possible.

Note that computer software doctrine is much more deferential to how programmers behave; not the lay observer. Would like to see that notion in all copyright law. Why care what lay auditor/observer sees instead of how a reasonable actor behaves towards relevant others?

Irina Manta: market substitution would still be relevant.  What about a study of copyright infringement surveys in TM infringement style?  What do people think about similarity in works and does it correlate with results in litigated cases?  If there’s a disparity we’d then want to know why/which we thought were right.

DiCola: The harm comes from the consumers—it’s either the primary part of the case or the result.  Heald is saying that the breach comes from D’s behavior, but harm has to be connected in some way to the wrongful behavior.

Samuelson: may need to distinguish reproductions from derivative works. Infringement test right now gets used the same between reproduction/derivative work.  Students ask what the difference is and she tries to give them a set of factors, but it’s difficult. Castle Rock: implausible to view SAT case as a case about substantial similarity and reproduction. TV series is not a quiz.  If it’s plausible at all, it’s a derivative works right case, and she doesn’t find it very convincing even as such. Good teaching case for discussing the difference.

DePoorter: if jury is substitute for survey, it’s harder to debias juries. Could try to mimic lab setting; valuable contributions would be in manipulating debiasing interventions.

Manta: currently parties are running their own unscientific surveys on the jurors. It would be fairer to do more; it doesn’t take much to skew people’s perceptions of similarity in our studies.  In a lab study you wouldn’t know anything about the litigation, who the parties are, etc.

DePoorter: what about not telling subjects that the D copied?

Manta: telling them that there was copying had a clear effect, as did telling them that a lot of labor went into the original; telling them there was market substitution did not have an effect. Subjects care about labor even though they’re not supposed to.

McKenna: even if people say works are similar, you have to figure out what’s triggering the similarity assessment, and a lot of things are supposed to be irrelevant. Controls are used in TM, at least theoretically. © control is harder to figure out.

Manta: true; our studies looked at smaller works like images rather than books/movies.

McKenna: Gucci v. Guess—sued over diagonal arrangement of overlapping Gs, beige on beige; Gucci claims all this as its trade dress, even though diagonals and beige on beige are common. Control: yellow on blue, arranged in square, with nonstylized letters—they changed all the elements, and the judge still said that was a fine control. Hard to have confidence in good © controls.

DiCola: evidence from criminal law suggests that debiasing is incredibly difficult/practically impossible.  Getting people not to think about labor may be impossible—how do we change their minds? 

Manta: but surveys help with that because people surveyed aren’t being told about labor; when it gets to the courtroom, you should limit what you allow attorneys to say.  Then it would be harder to ignore the survey evidence.

Fromer: courts have ignored this by sending it to jury.  Forcing the court’s hand by bringing in evidence and requiring its confrontation—in patent, inventors are forced to articulate what they think they’ve created up front. In copyright, we don’t have that, and there are good reasons, but it means that when you get to litigation you haven’t articulated what’s important about this work up front. We need a point in the © system as in patent and TM for forcing articulation of features P considers important.

Friday, October 24, 2014

Empirical IP Research Conference: TM counterfeiting

Trademark breakout session #2: Counterfeiting and Its Effect on the Market for the Genuine Article

Facilitator: Scott Hemphill (Columbia; Visiting Professor, NYU)

Why we care: innovation policy; enforcement policy.  Debate over magnitude: 10s of billions, 100s of billions in the US?  Is that all damage?  Lost profits: what monetary effects does it actually have on TM owners?

Counterfeiting scenarios: (1) seller fools the buyer: drugs, aircraft parts.  Mostly we aren’t talking about that, but rather (2) buyer fooling everyone else: handbags, sunglasses, footwear.  Or (3) nobody is fooled. Is that even counterfeiting? Dilution?

Standard story: substitution. Though substitution is highly contested.  Overuse—becomes uncool. Positive stories: Sampling: I try it, I like it, and ultimately I buy the original (as with sampling in music downloading).  Advertising: Might drive word of mouth.  Still might be a case for enforcement—the legitimate seller still needs some way of recovering; illegality can be a form of price discrimination by segmenting the market.

Sprigman: law on books v. law on streets. In criminal counterfeiting, DOJ maintains and most courts accept that once they prove the D’s mark is identical to/substantially indistinguishable from the registered mark, confusion is conclusively presumed.  Civil counterfeiting plaintiffs are now taking the same position.  Real example: if I take a wallet and fake up an expensive LV emblem, and stitch “fake” all over it, so no consumer would be confused, the gov’t says that’s criminal counterfeiting.  The Q of harm from counterfeiting has to be assessed w/in that framework.

McKenna: Other ways in which context matters: many kinds of products can be counterfeited, and some are much more harmful than others. Yi Qian has empirical evidence: similarity of price matters a lot. Where price is quite different, advertising effect tends to swamp substitution effect, but reverses as prices converge.

Joel Waldfogel: Deadweight loss can turn into consumer surplus if I can get a cheap copy of a handbag; no loss unless other people react badly to the fact that I own a copy.  Can look across products: but are there systematic data for fake Gucci purses?  Can look across individuals to see what their stock of fake/real stuff is. Do people who buy fake stuff start buying less real stuff over time?  (Drawing on music downloading studies.)

Stefan Bechtold: can present consumers with well-known brands and knockoffs whose similarity is exogenously varied. Ask about perception of knockoff and brand—determine whether there is some advertising/spillover effects.

Mike Meurer: Child labor/slave labor producing counterfeit goods? 

Sprigman: terrorism, narcotics, organized crime: it’s a problem w/manufactured goods generally!

Rahul Telang: comparative pressure on original seller to innovate to avoid copying? Pressure to lower prices that could be good?

Hemphill: Qian gets into that in her work on footwear in China. Firms most affected by counterfeits respond by going up vertically in quality—fancier, invest in vertical integration.  Natural experiment in that 1995 marked an exogenous fall in TM enforcement as inspectors shifted to tainted food. Also some companies were more in with the Chinese gov’t than others, and could get more help—measured distance from gov’t. At the high end advertising effect dominated over substitution, opposite at the low end. One of the Megaupload papers finds an effect like this: substitution dominates for blockbusters but advertising dominates for smaller movies.

Sprigman: The potential is for lower cost advertising when other people sell copies than when you advertise for yourself. Other people give you credibility by choosing to consume the counterfeit. But the proper function of law here is a relevant consideration. 

Telang: Bill Gates says, if you copy any product, copy mine.  I’ll eventually get you to shift over/buy complement.

Hemphill: generating some illegality may enable your price discrimination.

Orly Lobel: Different things you can do in this realm: Dishonesty of honest people.  Who is willing to do what?  Also can measure willingness to pay: and can measure deliberate ignorance/degree of certainty about whether something is a counterfeit.  Also: secondary markets. If there’s less demand/supply for counterfeits that can affect the size of the secondary market.

Hemphill: WIPO 2010 has a literature review on empirical studies—a few in the area of why people buy.

Glynn Lunney: categories of counterfeits. If the law allowed counterfeiting when no direct purchaser is confused the line might become more blurred.  Could become difficult for buyers to tell the difference.  If substitution and overuse are balanced by sampling/advertising, we should be able to allow the TM owner to enforce and TM owner would enforce efficiently; the question is consumer welfare—deadweight loss, desire for prestige brand.

RT: Dan Ariely on honesty—if you feel like a fake, you behave more dishonestly: another separate effect and it’s not clear how much law affects this but to the extent law affects magnitude it may matter.  Also, Lunney’s theory is unlikely to materialize because TM owners will overenforce: they will often see things in a property frame whether or not allowing some counterfeiting would be beneficial, if they have the right to enforce and are not forced to give up on the idea of control by practical realities.  (See Jessica Silbey’s work.)

Strandburg: might be that counterfeiting doesn’t matter as long as it’s illegal. Companies might want it to be illegal while being satisfied with a certain level.

Hemphill: much variation in reactions to buying counterfeits.  122 attendees at counterfeit purse parties, with no previous exposure to brand. 46% subsequently bought an original from the same line. Some felt very bad about having bought/being complimented.

McKenna: note how much this conversation doesn’t have to do w/TM law.  The trouble is situations we all know there’s no confusion.  It either is or isn’t another problem. Ought to be a question of whether we want copyright or design protection. If the mechanism of harm isn’t confusion, it doesn’t belong in TM.  Sui generis design choices are very different.  If we all recognize that there are problematic forms of counterfeiting, that’s where it’s most likely the case there’s genuine confusion.  So long as TM does its traditional role, no problem. Next we ask whether TM should address another type of product.

Sprigman: maybe we know it’s not confusion, but INTA doesn’t know.  There aren’t studies that fully contextualize shopping experiences. Need to know more about shopping intention. Lack empirical infrastructure to say where confusion really is.  His intuition agrees that in large swathes people know what they’re getting and it’s just reducing deadweight loss, but there’s work to be done.

Strandburg: relates to panel session: confusion is all related to context, which includes price, but no TM study will ever ask whether people think the $20 thing comes from Louis Vuitton: decontextualization of confusion.

Empirical IP Research Conference: trademarks

Plenary Session: Measuring Consumer Confusion in Trademark Infringement

Facilitator: Barton Beebe (NYU)
Lanham Act: confusion is vaguely defined.  Used to include “purchasers” but Congress deleted that phrase.  43(a)’s language is even broader.  Flexible and slippery.

Panelists: Joel Steckel (NYU Stern School of Business)
Topic: Consumer Confusion Surveys Used in Litigation, Commenting on: Robert H. Thornburg, Trademark Surveys: Development of Computer-Based Survey Methods, 4 J. Marshall Rev. Intell. Prop. L. 91 (2005)

Wants to talk about dilution.  Thornburg paper: not a very good paper, but that’s because it does two things—a catalog of different survey types/principles, then talks about difficulty of getting internet surveys into evidence. Much of what he said in 2004 is obsolete in 2014, when internet surveys are by and large admitted with some exceptions.  Drilling down into one aspect: dilution surveys. If confusion is slippery, dilution makes that look like sandpaper. Harm the reputation/impair the distinctiveness of a mark: what does that mean?

Evident from usual surveys that courts accept that courts don’t really know what they’re trying to measure.  Exxon survey: Nike v. Nikepal, where respondents were simply asked what if anything came to your mind when I first said the word Nikepal, 79% said Nike.  Makes you wonder, what did the other 21% say?  That is considered evidence; it’s evidence of association, but doesn’t correspond to impaired distinctiveness or harmed reputation.

Associative network memory model of brands.  Knowledge is a network of bits of info: TM or associated logo, slogans, etc., and associations—anything that comes to mind when you think of a brand name. When the word Coke is presented to customers, they think of “tradition,” “nostalgia,” etc. and Pepsi is more excitement/use/taste.  Dilution offense occurs when a junior brand enters and has another set of brand associations.  Tarnishment if a new association is negative and somehow gets attached to the senior brand: if Victor’s Little Secret causes consumers to associate Victoria’s Secret with pornography.  Test it by seeing if test group associates VS with porn.  Blurring: distinctiveness would be impaired if link between Nike and sports was weakened; reaction time to associating Nike with sports would be a measure of blurring.

Takeaways: dilution has been difficult to measure because it wasn’t defined well.  He argues that his definitions are suitable theoretically based definitions.  Marketer’s perspective: a sensible definition would include the potentially dilutive power of mixing and adding brand associations. Measuring the degree to which brand asociations are held and how easily they are recalled can provide measures consistent w/the language of the law.

Lisa Larrimore Ouellette (Stanford)
Topic: Cognitive/Psychological Approaches to Modeling When and How Consumers Get Confused, Commenting on: Thomas R. Lee et al., An Empirical and Consumer Psychology Analysis of Trademark Distinctiveness, 41 Ariz. St. L.J. 1033 (2009); Thomas R. Lee et al., Trademarks, Consumer Psychology, and the Sophisticated Consumer, 57 Emory L.J. 575 (2008)

There is a fair amount of disagreement about TM’s goals and therefore about what facts matter.  Take courts at their word: consumer mindset/confusion is what we care about.  Multifactor test for likely confusion.  Buyer sophistication: there’s a whole literature about consumer care in buying choices and courts aren’t paying attention to that, which we should bring in. No clear direction for likely confusion; another paper looks at likelihood of bridging the gap.

Sophisticated consumer paper: courts have ID’d many factors as relevant to this: low price of goods; purchase complexity, frequency of purchase; education, age, gender, and income of buyers; professional buyers or hobbyists. But little attention to consumer psych literature.  That literature discusses consumer motivation as well as consumer ability to exert effort to make distinctions; they need both.  Motivation can depend on the realm and on the person (some people have greater needs for cognition than others).  Longer time to make decision without distraction can make difference.  Courts make generalizations about low price, but that’s not the relevant question.  Low financial risk may be high physical or social risk, creating more motivation.

Consumers who saw Mercedes Benz computers were more likely to think that Cadillac would bridge the gap to consumers too.  Those test subjects with more experience buying consumers or more education were more likely to be confused—cuts against judicial doctrine.

Secondary meaning survey: created fictitious marks with no secondary meaning and mimicked typical trademark use on products—Chocolate Abundance, Party Hat, Fudge Covered Cookies etc. for chocolate covered cookies.  If you believe the Abercrombie spectrum means something you’d expect some linear progression.  What they found was that consumers were unlikely to see generic terms as indicating source, but for the rest it was all the same.  And even generic term, over ¼ of consumers saw it as indicating a brand name.  Presentation matters: large font/ordinary TM place makes TM perception more likely, small font/placement makes TM perception less likely.
boxes of chocolate cookies with different words in large oval on front of package

Abercrombie may serve other interests, like protecting competitors—can still be reasons to encourage parties to select more arbitrary/fanciful terms over descriptive ones.  Still, it’s not good for telling us consumer mindset, and we need other measures of distinctiveness and strength.  Inherent distinctiveness is a very weak proxy for what consumers are looking at. 

Could test other factors like similarity of marks and proximity of goods in similar ways.  Courts’ general assumptions may be unreliable.

Mark McKenna (Notre Dame)
Topic: Qualitative Studies of Consumers Becoming Confused During Shopping, Commenting on: G. Miaoulis & N. D’Amato, Consumer Confusion and Trademark Infringement, J. Marketing 48-55 (April 1978); Vincent-Wayne Mitchell & Vassilios Papavassiliou, Exploring Consumer Confusion in the Watch Market, 15 Marketing Intelligence & Planning 164-172 (1997)

There aren’t very many good studies about confusion.  One reason: marketing folks aren’t interested in confusion as defined in TM law. They’re intensely interested in how marketplace practices affect a brand, but there’s a very important distinction between brands and marks; brands are much broader concepts including associated meaning which can be affected by a much wider range of practices that might or might not involve TMs/confusion.  Marketers are much more focused on the question of harm.  Those might be in cases where consumers know accurately that a new product comes from the existing source (might be thought to simulate 100% confusion)—is there any harm to the brand under those circumstances?  TM law accepts claims of harm much more readily than empirical evidence justifies.  Studies are also focused on harm to the brand, not harm to consumers except maybe incidentally. So you get statements like the M&D’A study—confusion in many circumstances won’t harm consumers because consumers don’t care. Thus these studies press normatively to include in the definition of confusion a broader range of effects than TM academics care; trying to shape legal doctrine to test the things they care about.

First study: attempts to measure extent to which consumers exposed to new product that shares features—here, packaging of mints, features of the brand but not the TM—will generalize characteristics from the known product to the new product.  Authors count stimulus generalization as confusion! They want this to count even if normal purchasing conditions make other characteristics distinguish the products, which they call “intellectually discriminating” between the products.  “Confusion” despite the absence of source confusion on the theory that this generalization harms the brand by making it less “differentiated.” And indeed they so find—people think the new product will be refreshing and minty or will taste good. That, they think, is terrible!  Consumers porting info from Tic-Tac to other mints makes Tic-Tac no longer as distinctive.
tic-tacs, Dynamints, and Mighty Mints

Conventional TM perspective: completely irrelevant to doctrinal convention. TM is concerned with one kind of generalization: source information—who is responsible b/c of the presence of mark. Non-source info shouldn’t matter; and porting info for reasons unrelated to protectable to mark should also be irrelevant. Normatively this is correct; ability to generalize is often unqualifiedly good for consumers and for competition even if bad for particular brand owner. Effective communication of “my product is minty and tastes good too” is called competition. TM’s job is not to prevent people from selling products with similar desirable characteristics. That’s why we approve comparative advertising.  How we describe a new restaurant: almost always in relation to something you already know.

Modern TM law becomes more intelligible if you read it through the lens of these studies. Doctrines that most befuddle academics are b/c courts squeeze into TM the broader concerns in these studies—dilution, initial interest confusion, post sale confusion.

Mitchell & Papavassiliou study: followed customers around watch store and asked them what aspects made the experience more challenging and what they did to overcome those challenges.  Consumers say: fragmented nature of market; newness of tech in watches; hidden nature of watch movement; role fashion; too many brands/shops; low frequency of purchase; purchase of watches as gifts. The word for these effects isn’t confusion, but contextual factors that interfere with decisions in shopping/raise search costs, but not confusion. But maybe that means confusion isn’t well defined. 

Very few of these interferences are things TM has anything to say about. Maybe that’s a good reason to avoid thinking of TM as a means to combat search costs but rather a particular sort of interference.

Info overload/overchoice.  Consumers can struggle to choose when there’s too much info. Study tries to figure out consumer strategies—clarifying goals, narrowing choice, seeking additional info. This 1998 study could usefully be updated for new contexts like the internet.  More interesting: what relationship TM has to the problem of overchoice.  Paper simply asserts that counterfeiting is a source of confusion and law can fight that, but the study is from a store apparently lacking in counterfeits. Paper says subbranding is contributor to confusion, undertaken by brand owners themselves as a way of differentiating products: abundance of functionally similar products differentiated on a fashion platform. We think of TM as reducing search costs by making search less time consuming, but TM also affects differentiation in the market, and has a lot to say how close products may be and how they must be differentiated—product configuration like color of band.

To what extent is TM contributing to and reducing search costs, and how do those net out? If TM incentivizes lots of dimensional differentiation, that can increase search costs.

Studies differ but have common pressure to define confusion more broadly. Search costs rhetoric has made it easier to redefine confusion in this way.  As Mitchell study illuminates, there are lots of other kinds of search costs, and it’s too easy to call those confusion.  Academics: if you accept that TM doctrine has responded to these papers’ concerns even if the doctrine isn’t well suited, then we have to do more than point out the irrelevance of these types of confusion. Vacuum about types of confusion that ought to be relevant; studies focus on irrelevant question. Where empirical evidence is, doctrine moves. To resist, need to do some good empirical work about when consumers are confused in the way we want to mean it.

I’ve been asked to speak about the big data approach to consumer confusion. One promise, likely illusory, of big data is that we need not care any more about causality; we can merely find correlations, however unexpected, and react to them.  But trademark infringement, and even dilution, requires a causal narrative as currently understood: this use created this mental state in the consumers who saw it.  As such, big data may not be our savior.  Instead, large datasets may be able to change the theory of what trademark infringement (and even dilution) is, the same way that previous advances in marketing and consumer psychology led some to reconceptualize what trademark infringement and dilution are, but that will be a normative choice even more than an empirical one.

We’ve seen one American case really take a stab at using “big data,” in the form of Google ad clickthrough rates.  That’s 1-800 Contacts, Inc. v., Inc., decided by the 10th Circuit in 2013.  The court says that the theory of IIC is that consumers seeking 1-800-Contacts (which we know, the court says, because they searched for the term) clicked on a ad while believing it was a 1-800 site and, though no longer confused when they arrived at a site, were nonetheless diverted.  We have no idea how many were confused when they clicked and how many were not confused but rather were seeking a possible alternative to 1-800, but the court says that we do know the upper bound of the former number: the total number of clickthroughs, which was a tiny fraction of the impressions.  For ads without the 1-800 Contacts trademark in the ad copy, got a 1.5% clickthrough rate.  This was too tiny to count as likely confusion even if all clickthroughs were the result of IIC, so couldn’t be liable for infringement based on those ads.  But ad clickthrough rates are always very low, so even a straight-up counterfeiter as in Rosetta Stone v. Google would seem to escape liability for infringement if you measure confusion in this manner.

This brings me to an argument made in a 2000 article by Alex Simonson, Survey Design in False Advertising Cases: he argues that we need to pay attention to attention versus comprehension.  We might have a survey where a small percentage of people notice a particular product feature, say a picture of a black rooster and the label Gallo Nero on the neck seal of a bottle of wine, but those who do almost all make a connection to E&J Gallo wines.  Or our survey might show that many people notice the neck seal, but only a few make the connection to E&J Gallo.  Simonson argues that those are different results even if they produce the same net number of so called confused consumers: for legal purposes, we ought to be concerned more about comprehension than attention.  Attention, after all, varies in the real world, and so artificial attempts to measure it may not be very good—but if we look at the percentage of people within the set who noticed a feature or symbol who were confused by that feature or symbol, we have a result that may be more generalizable as well as more important. 

I think this point applies beyond surveys: it’s a truism that most advertising is completely useless—and now we have eye tracking studies showing that most people literally do not see most ads, to which one English court has referred in its reasoning, and MRIs showing that a lot of the time our brains don’t even react to ads, not even bothering to process them.  In that sense, there is no such thing as an ad that is likely to confuse, because it’s not likely to reach us in the first place.  But the people who don’t notice the ad are arguably irrelevant to any confusion inquiry.  It’s only those who notice who might be confused. So, while I hate initial interest confusion and think is right on the merits, I don’t think the analysis is right.  All we know is the number of people who clicked on the link—but we don’t know how big a percentage that is of the number who noticed the link.  It might be that almost everybody who looked at that link clicked on it, possibly because they were confused, and that scenario ought to concern us if it occurs, especially to the extent that there are no countervailing benefits from having the ad—and it’s hard to explain how the ad might benefit consumers who don’t notice it at all, though one might possibly construct an argument about chilling effects on truthful advertisers.

Also, we need a concept of relevant confusion, or materiality.  Something that people mostly don’t pay attention to because they don’t care about it shouldn’t be deemed confusing.  The broader takeaway is that we still need a theory to explain what we should care about and why; big data do not remove the need for big ideas.

That leads me into the papers on which the organizers specifically asked me to comment: Stefan Bechtold & Catherine Tucker, Trademarks, Triggers and Online Search, J. Empirical Leg. Stud. (forthcoming), and Lisa Larrimore Ouellette’s The Google Shortcut to Trademark Law, 102 Calif. L. Rev. 351 (2014).

Bechtold and Tucker used a large dataset to explore the effects of Google’s policy changes allowing more competitor keyword purchases of trademarks in Germany and France.  Basically, they found little net change in the number of consumers who ended up on the trademark owner’s website, but a change in composition.  They divided searches into navigational searches, where the consumer is searching for the keyword because she is directly interested in using the search engine as a shortcut to find a specific webpage such as the trademark owner’s website, and non-navigational searches where the consumer is doing something else.  They estimated which were which by looking at how long the search was—so iPhone would be navigational but “how to restart my iPhone would be non-navigational”—and some other contextual factors. 

After Google’s liberalization, navigational searches became less likely to lead to the trademark owner’s website, while non-navigational searches were more likely to do so.  Percentagewise, they classified 20% of searches as purely navigational, while 80% were non-navigational. The policy change was associated in a 9.2 % decrease in consumers visiting the trademark owners’ websites when they used a search phrase that exactly matched the trademark. But consumers who were searching using the trademark alongside other words were more likely to reach the trademark owners’ websites in 14.7% of  cases. 

What can we glean from this?  The authors rightly recognize the limits of their results.  While they suggest that a navigational searcher’s search process might be “impeded” by unauthorized use of a mark, “as her attention is drawn to many third-party websites in which the searcher is not interested,” we don’t know whether she perceives any impediment.  A searcher committed to finding the trademark owner’s own site can usually determine without clicking which site is which, especially with the prominent brands the authors tested.  David Franklyn and David Hyman have shown that consumers are often confused about whether a search result is organic or paid, but rarely confused about the underlying source of those ads. 

What the authors can test—and thus what their data might push trademark theory to care about—is whether trademark owners suffer any loss of consumer visits from the policy change. They conclude that their findings do provide an upper bound to potential negative effects of the policy change on trademark owners: they only suffer from negative consequences within the subclass of navigational users, only 20% of searches. In European law, arguably these negative effects have something to do with the “investment function” and the “advertisement function” of trademarks, though I have to admit I don’t really know what those are other than ways of stating ownership claims regardless of any effect on consumers.  Even if the consumer began by wanting to visit the trademark owner’s site, we can’t say without knowing more that she’s worse off under Google’s liberalized policy.  Maybe the new choices made her rethink her initial desires.  If the ads weren’t confusing, we need some other reason to say that’s wrongful, and even the European approach doesn’t explain why diversion is a wrong to the consumer.

I’m intrigued by some subsidiary findings for the light they can throw on what we don’t know about consumers: First, “searches on Google appear to be consistently associated with fewer visits to the trademark owner’s site and more searches and activity before a visit to the trademark owner’s site even before the policy change” compared to Bing or Yahoo!  Second, “relative to France, searches originating in Germany are less likely to lead to a trademark owner’s site and also … [German] searchers are more likely to engage in multiple searches prior to a visit to a trademark owner’s site.”  National origin and Google brand loyalty appear to be independent effects on the extent to which users are precommitted to trademarks, which ought to shake our confidence that we know what matters to consumers, much less why.

More generally: the paper provides important information about consumer behavior, but not about what consumers are thinking—we can still only infer why they typed what they did.  Theory cannot be abandoned.  Moreover, the specific results cast into doubt the free riding model where it’s wrongful for a non-trademark owner to change consumer behavior regardless of what the consumer thinks—it seems that unauthorized uses may not in fact work harm in the aggregate.  However, in a big data world where we don’t have any interest in knowing why behavior changed, only that it did, a court might attempt to fine-tune the rule, and not allow keyword purchases on the naked mark or so called navigational search, in order to make things even better for trademark owners.  This is an example of a new legal theory that can emerge from new methods of measurement; it wasn’t suggested by previous doctrine.  Whether that is a good or a bad idea depends not on consumer behavior but on one’s theory of trademark rights.

Lisa Oullette’s paper on using search engines to make trademark judgments primarily addresses a more basic question: how do we know whether a claimant has a mark at all, or a strong mark.  She argues that if a mark is strong—either inherently distinctive or commercially strong—then many top search results for that mark will relate to the source it identifies, so you can use search engines to determine distinctiveness. Relatedly, she argues, the extent of results overlap between searches for two different marks can also be relevant for assessing the likelihood of confusion of those marks.  For marks that don’t point uniquely to the claimant except within a particular set of goods and services she suggests using searches that add more information, like Mission Burritos rather than Mission, which might be deemed non-navigational in the Bechtold/Tucker framework. 

Some thoughts:

(1) This theory may have the most potential implications for registration system: registration occurs in the abstract, not a full marketplace inquiry, and at least for word marks does not concern itself with the current visual appearance of the mark.  The paper’s argument would imply that we should take search engine results much more seriously in the registration context to determine what a word means. 

(2) However, we still need a theory of trademark meaning: example from the paper’s treatment of MICRO-THIN for condoms: all the results for “micro thin condoms” referred to the relevant plaintiff’s mark, but as the paper points out, this could mean strength for condoms or it could just mean that the plaintiff is currently the only user of a not particularly distinctive term within that category—I think this is a more significant weakness than the paper acknowledges, because by entering “micro thin condoms” as the search term you have already taken the position of a consumer using the term to locate something, while what we should want to know is “would a consumer use this term to identify source”?  The paper’s conclusion: “when consumers search for MICROTHIN condoms, they are not simply looking for condoms that are ‘extremely thin’—they are generally looking for Kimono MicroThin condoms.”  But that “when” contains the assumption that drives the result.  We don’t know from these search results if the term is distinctive in the sense of serving as an identifier of source—instead the paper’s inquiry is whether the term is relatively unique within its field, and that really changes the basis for trademark protection into a less consumer-focused rationale and far more producer-oriented.  The traditional question of source significance asks “if” consumers search for micro-thin condoms as an identifier of source, not what happens “when” they enter the search terms.  Questions I would be much more interested in would be “how often do consumers search for Kimono condoms or Kimono micro-thin condoms?” and “how often do consumers use micro-thin to modify other brand names in their searches?”  The Bechtold/Tucker type of data are far more probative of that than the search outputs.

(3) IP law’s largely textual focus, which makes it much more confident handling words than other symbols.  Already reflected in the doctrine: Abercrombie spectrum for word marks might be empirically flawed, but it seems much more manageable than alternatives proposed to identify inherent distinctiveness in symbols or trade dress: Seabrook test for trade dress, for example, is just another way of saying, four times, is this distinctive?  Unfortunately, just as with current doctrine, image search is far behind word marks; the Google shortcut might be most productive with respect to the least troublesome marks.  And when there are both verbal and visual components, there are additional complications: there are word marks that are stronger when coupled with design features: a video “tube” site imitating YouTube’s red and white ovals.

(4) Concerns about using searches to assess likely confusion.  To take another example from the paper, TELMEX currently produces first page results all related to one company, and the paper suggests that therefore a new mark AUDITORIO TELMEX might be confusing.  But that seems to overweight the senior user’s interest: consider the old Gruner + Jahr v. Meredith case, PARENTS versus PARENT’S DIGEST.  When I type in “parents magazine” I get only results for the former on the first page, but when I type in “parent’s digest” I get no results for PARENTS magazine—one might conclude that PARENTS is a rather weak mark since it doesn’t survive even minimal alteration. 

The case law suggests that we should be interested in what consumers think when they see the junior use, and thus search on the junior mark is more probative than search on the senior mark. When I type in “Auditorio Telmex” I get no results relating to Telmex, the senior user.  The results from the previous paper might also be useful here, given that consumers often refine their searches when they started with too abbreviated a term to get useful results.  So, for example, if we found a certain number of people trying Telmex first and then Auditorio Telmex, that could be evidence not of confusion but of accurate understanding that these were two different entities and that the initial shorthand just wasn’t enough to identify the actual target.  Note that we could call that dilution, or we could say that such results show that Auditorio Telmex wouldn’t dilute the naked Telmex mark at all, but we would need to be clear about our definition of dilution to make that call.

[did not get to say this para.] Relatedly: as price discrimination grows, different searchers may get very different results depending on their previous purchases or demographic profiles—which relates more closely to many of trademark’s concerns than current geographic personalization: Northeastern study showed this is already happening with segregation based on devices on travel sites.  Might support the paper’s theory insofar as Google starts taking better account of differences in consumers, so price points and consumer sophistication would actually divide search results.  On the other hand, the paper’s argument about using Google to divine relatedness of goods or likelihood of expansion does not seem persuasive: the paper says “if a non-expanded mark is sufficiently strong that consumers might anticipate such expansion, searches with keywords for those fields would likely still have pages related to the mark,” but given search engines’ attempts to provide presently relevant results I can’t see why that would be true—at the very least we’d need a lot more information about who writes these pages saying “I can’t wait for the McDonald’s nav rattan korma.”

These details reinforce my concern that we should be very careful about when we change doctrine in response to what we think we can measure.  We might not even notice that we’re changing the doctrine (which might, for example, narrow the protection for visual marks if Google became a preferred source of evidence—I might be happy with that outcome even if dubious about the mechanism).  We have a historical example of this evolution with judicial treatment of survey evidence and the introduction of a requirement of control groups, which had the practical effect of changing the percentage of consumers exposed to the allegedly infringing use that courts recognized as sufficient to show likely confusion.  That’s probably the happiest story of doctrinal evolution, but the decision to shape the doctrine may be more normative than empirical, and that has to be kept in mind in all these discussions.

Beebe: two cultures—marketers on one side and lawyers on the other. Talking to marketers isn’t the same as talking to economists—who we talk to affects the law just as talking to economists affects patent law.  Marketers say: confusion as to source isn’t interesting—let’s figure out how to measure dilution.  Goes to slipperiness of models and constant push of law away from confusion as to source and towards confusion as to something else.  Marketers: it’s not consumers search costs, but TM search costs: TM law is currently dedicated to minimizing the costs for TMs looking for consumers. Subjects are TMs/brands and objects are consumers; lawyers don’t yet admit that.

Sprigman: for Steckel: cognitive delay as dilution? When I get older, I get cognitively slower, have I diluted the mark?  What’s the harm?

Steckel: when people get older: all associations are delayed, that’s what a control group is for.  There is research showing that even for the most important of purchases—home, car, etc., that at any moment only 3-4 attributes are operative. People have limited info processing capabilities and as such you want to put your best pitch up front.  New Coke debacle: New Coke tasted better, but people didn’t buy the product for its taste.

Sprigman: but there’s no falsity injected into associations through “dilution.”  What you’ve posited, is substantive weakening of meaning, can see effect of that, but not new associations.  Tarnishment story makes more sense, but is a huge First Amendment problem since in every other context you’re allowed to say “Victoria’s Secret sucks,” even as a competitor.

Steckel: you don’t buy that measuring reaction times captures notion of impairing distinctiveness.

Sprigman: need relevance in retail environment.

Steckel: by delaying reaction times, it bumps something from short term memory to long term memory, taking longer to evoke in world of limited attention.

McKenna: this is two cultures: lawyer wants to know where the associations are, with source 1 or source 2.  Steckel is asking what associations will arise—that’s a mismatch.  Substantive meaning of mark v. whether it will take longer to know whether I’m dealing with New Coke or Old Coke.

Sprigman: whose views count for the law?  Typically we don’t impinge freedom of speech without harm.

RT: I wrote a whole article about this.  Don’t agree with the characterization that delays correspond to shift from short term memory to long term memory.  160 milliseconds is not a long enough delay.  This is an example of what happens to empirical work laundered by lawyers: it gets ramped up as more significant than it is in the translation; we regularly mistake statistical significance for practical significance.

Bechtold: Use of big data and limitations.  One response: need more data.  In some areas, this is true—initial interest confusion; we tried to test this and could’ve tested this w/right data.  In general, we are only slowly seeing usable data.  Fully agree that this is not a substitute for a theory of TM that is justified.  Studies can help us find new theories/new effects.  Spillover effects from unauthorized uses can benefit brand owner; no one has clear theoretical view of how this works psychologically or what TM lawyers should do w/ it.

Jeremy Sheff: should ask what questions empirical methods should be used to answer and that depends on our theory of TM law. Consumer psychology: what are we trying to measure in the minds of consumers?  Marketing departments have been looking at these questions though not attuned to doctrine. Competition: if we think there are empirical questions, we could turn to competition/antitrust law.  Look at effects on entry, prices, output, elasticity of demand. These aren’t being investigated right now. That could help make a case for a particular theory or explore our theoretical presumptions.

Orly Lobel: useful to think about this panel v. patents panel.  Much has to do with whether we’re having the same conversations: does the theory map onto the work? Confusion seems like an instrumental, narrow goal—could it be consumer welfare, new entry, etc.?  Current work seems defeatist, focused on what we have now on the books.  This panel spoke to adjudicators: how to apply/interpret doctrinal mandate; previous panel was willing to talk about what the law should be.  Maybe we just don’t know how to aggregate the effects of TMs.  Or maybe we’ve seen more action on patent reform and thus had more demand for studies of bigger Qs.

McKenna: patent traditionally wanted to promote more output. TM traditionally just wanted to prohibit particular practices.  Yet TM is used as a substitute for or complement to patent law; can be measured in terms of same effects.  Antitrust people: don’t think TM has any effect on competition/any market power for brands. That’s bound up in history, but we need more study.

Chris Buccafusco: TM doesn’t start w/empirical assumptions about brains and decisions and then try to figure out harms; it starts w/assumptions about what bads are and tries to get evidence from social scientists.  Mental map of associations is starting point of marketers, but not TM.  Might be any number of conceptions of how consumers use brands; could be less cognitivist models.  Might produce different sets of harms.  His skepticism: TM owners want to find all the harms and use all possible theories of decisionmaking as long as they specify potentially compensable harms.

Gregory Mandel: move to the ought is useful; so is how the law is working on the ground, as the TM papers were focused on.

Scott Hemphill: clickstream data: you can see sometimes that people stayed only for 5-7 seconds. That’s a signal they didn’t get what they wanted. Could also get that from sequence of queries. Might be a way to detect brief confusion (though, RT notes, also self-refutes theory of harm to TM owners).

Irina Manta: not sure that association w/porn is the only way or even the most direct way to get at tarnishment. If you believe in the model, Victor’s Little Secret creates negative association, but people don’t remember why they have a negative association when they see VS again, it’s just a generalized dis-ease.  Quality perception reduced; that should be actionable every time there’s no First Amendment protection, according to TM owners.  (Of course, under Central Hudson, “every time there’s no First Amendment protection” means “every time there is a factually false statement about the TM owner in commercial speech,” which tarnishment is not.)

Steckel: hard to expose respondents to an experience that mimicks what happens in the market; simply asking them about associations w/porn may not do it.

Michael Meurer: there have been some market value studies about value of TMs. Not sure whether there are studies on market effects of TM litigation, but there could be.

McKenna: there are European economists who try to study relations to innovation, but there are many problems. First, they just measure TM applications which has correlation/causation problems.

Katherine Strandburg: Papers related to Abercrombie emphasized the importance of context in what consumers see.  Google genericity case: court distinguished verb use from TM use.  People use google generically as a verb but not as a source.  Can we split the baby here, giving people more protection for the mark in a certain context, but less in other contexts where the use isn’t the same?  Also, tension in Oullette paper: if you do a search and the first page is all about a particular TM, then she argues it’s distinctiveness. But wouldn’t that also suggest a lack of confusion?  That seems perfectly sensible—a more distinctive mark might be less confusable.

Barton Beebe: TM says that stronger mark = greater scope of protection/more likely confusion is. If we abandoned that principle, the edifice would crumble.  Some European courts have suggested that strength decreases likely confusion, but not American courts. (RT: I would say that individual US courts have occasionally reinvented this theory in parody/First Amendment cases, but haven’t made it coherent.)

McKenna: not about empirical reality, b/c empirical evidence supports Strandburg.

Brett Frischmann: Interesting empirical questions would challenge assumptions/theory of TM.  Assumptions about stable/fixed preferences of consumers; examine way TMs enable producers to shape consumer preferences. What are the public harms from TM? Deadweight loss—how to measure in TM law? Chilling effects, speech values.