ARE “PAY NOW, TERMS LATER” CONTRACTS WORSE FOR BUYERS? EVIDENCE FROM SOFTWARE LICENSE AGREEMENTS∗

FLORENCIA MAROTTA-WURGLER NYU Law School

Center for Law and Business

August 22, 2005

ABSTRACT

The rise of commerce over the Internet and telephone has led to widespread use of “pay now, terms later” or “rolling” standard form contracts, in which buyers are not able to read the standard terms until after they have purchased the product. While some scholars and judges argue that rolling contracts do not merit special attention, others, including consumer advocates, are concerned that sellers take advantage of delayed disclosure by hiding especially unfavorable terms. Using a large sample of software license agreements, this paper shows that software publishers who use rolling contracts for their online sales do not, in fact, present more one-sided terms than those who make their licenses available prior to purchase. If anything, it is the contracts that require buyers to explicitly agree to terms before completing an online purchase that have the strongest pro-seller bias. Thus, to the extent there are inefficiencies associated with standard form contracts, they are not made worse by delayed disclosure.

∗ Wagner Fellow in Law and Business, NYU Law School and Stern School of Business. Email:

fm275@nyu.edu. Webpage: http://homepages/nyu.edu/~fm275/. I am grateful to Barry Adler, Bill Allen, Jennifer Arlen, Raphael De Coninck, Clayton Gillette, Lewis Kornhauser, Barak Orbach, Roberta Romano, Jeff Wurgler, and especially Peter Siegelman for helpful suggestions, and Leonard Lee, Christine Murphy, and Yuriy Prilutsky for excellent research assistance.

I. INTRODUCTION

The rise of commerce over the Internet and telephone has led to an increase in the use of

a new type of standard form contracts, commonly called “pay now, terms later” or “rolling”

contracts. In such contracts, the buyer orders a good over the phone or the Internet but does not

have an opportunity to review the standard terms governing warranties, limitations on liability,

and other elements of the purchase until later, when the contract arrives bundled together with

the good. A typical example is a license that has been shrink-wrapped inside a software package.

It cannot be read until after purchase, when the box is opened. Airplane tickets and most goods

purchased by phone are other everyday examples.

Opinions about rolling contracts are sharply divided. Some scholars and judges believe

that rolling contracts facilitate transactions and simply reflect the technological evolution of mass

commerce. Most notably, in ProCD v. Zeidenberg1 and Hill v. Gateway 2000,2 Judge

Easterbrook endorsed the use of rolling contracts as long as buyers had a right to review and

reject the standard terms. He pointed out that in many circumstances it would be impractical or

nearly impossible for sellers to communicate their standard terms to buyers prior to purchase.

Clayton Gillette has also pointed out that, as a practical matter, delayed disclosure is unlikely to

make any difference to buyers since they rarely read form contracts.3 Randy Barnett and Robert

Hillman offer complementary arguments to justify treating rolling contracts like any other form

contract.4 Douglas Baird recommends that in the absence of evidence of systematic advantage-

taking, contractual innovations associated with new technologies should be allowed to evolve

naturally.5

Other commentators vehemently disagree with the notion of rolling contracts, arguing

that delayed disclosure runs afoul of the contract-based-theory of mutual assent.6 Stewart

1 86 F.3d 1447 (7th Cir.1996). 2 105 F.3d 1147 (7th Cir. 1997). 3 Clayton P. Gillette, Rolling Contracts as an Agency Problem, Wis. L. Rev. 679 (2004). 4 Randy E. Barnett, Consenting to Form Contracts, 71 Fordham L. Rev, 627 (2002) (noting that the

hostility towards form contracts in general stems from the implicit adoption of promise-based conception of contracts; when this is abandoned, rolling contracts appear to be less problematic); Robert A. Hillman, Rolling Contracts, 71 Fordham L. Rev. 743 (2002) (arguing that Llewellyn’s theory of blanket assent to reasonable terms should govern the enforcement of rolling contracts).

5 See James J. White, Default Rules in Sales and the Myth of Contracting Out, 48 Loyola L. Rev. 53 (2002) at 70, citing letters from Douglas G. Baird [hereinafter Baird, Letter] and from Alan Schwartz [hereinafter Schwartz, Letter] to Lawrence J. Bugge, Chair, UCC Article 2 Drafting Committee (March 9, 1999).

6 See e.g. Stewart Macaulay, Symposium: Freedom of Contract: Solutions in Search for a Problem?, 2004 Wis. L. Rev. 777 [hereinafter Macaulay, Freedom of Contract]; Deborah Post, Dismantling Democracy: Common

2

Macaulay stated that “it hurts to be told that these are contracts” and that “if you are a contract

purist, it is very difficult to offer a convincing argument that these clauses work to create a

contract with the desired effect.”7 Together with consumer advocates, this group of scholars

argues that rolling contracts should not be enforceable because of their potential for abuse. In

particular, by concealing the terms until after the buyer has received the good, sellers may be

inclined to offer less favorable terms. For example, Jean Braucher argues that marketing

products to consumers without effectively communicating standard terms before payment is

unfair and deceptive, as well as a violation of various federal and state acts.8 Moreover, if

standard terms cannot be reviewed until after the product is delivered, comparison shopping

becomes costlier and thus the forces of competition, which might otherwise induce sellers to

internalize buyers’ interests in the setting of standard terms, will be diminished.9 Finally,

commentators have paid special attention to transactions involving that involve the general

public as opposed to large business buyers, since the former may be less able to comparison shop

effectively.10

Neither side of the debate, however, can point to any systematic evidence to support its

position. This paper fills the gap by studying the standard terms offered by software companies

who sell their products online. Several factors make this a particularly appropriate environment

in which to address the question of bias in “hidden” contract terms. First, the End User License

Agreement (EULA) associated with each software product presents a rich standard form

contract. Second, there is dramatic variation in how forthcoming companies are in presenting

Sense and the Contract Jurisprudence of Frank Easterbrook, 16 Touro L. Rev. 1205 (2000); Roger C. Bern, “Terms Later” Contracting: Bad Economics, Bad Morals, and a Bad Idea for a Uniform Law, Judge Easterbrook Notwithstanding, 12 J. L. & Pol’y 641 (2004).

7 Macaulay, Freedom of Contract, id at 807. See also James J. White, Autistic Contracts, 65 Wayne L. Rev. 1693 (2000) (noting that the notion of assent is problematic in the context of rolling contracts).

8 Jean Braucher, Delayed Disclosure in Consumer E-Commerce as an Unfair and Deceptive Practice, 46 Wayne L. Rev., 1805 (2000).

9 Id. See also Randy E. Barnett, The Sound of Silence: Default Rules and Contractual Consent, 78 Va. L. Rev. 821, 866 (1992).

10 See e.g. Andrew Brandt & William Wallace, What Have You Signed Away Today?, PC World Magazine (August 2001) (noting that consumers agree to onerous software EULAs because they rarely read them); Annalee Newitz, Dangerous Terms: A User’s Guide to EULAs, www.eff.org/wp/eula.php (introducing a number of unfair EULA terms and explaining that in many situations consumers are legally bound by them even if they were not able to see the terms until after purchase); Todd Bishop, Should Microsoft Be Liable for Damages? Seattle Post Intelligencer (September 12, 2003) (explaining why consumer advocates argue that software makers should pay for damages from viruses); The Economist, A Lemon Law for Software? (March 14, 2002) (exploring whether software makers should be made more accountable for damage caused by faulty programs by noting that consumers “are on their own, with little in the way of legal redress for any damage caused by faulty software.”)

3

their EULAs on their website to potential buyers. Some software publishers show the terms to

buyers before purchase, while others simply do not (i.e., they use rolling contracts). Third,

software products are homogeneous enough from company to company to allow for meaningful

large-sample comparisons of their EULAs. Thus, a simple way to test which side is right—those

who favor rolling contracts or those who oppose them—is to compare the standard terms offered

by software companies that do and do not utilize rolling contracts.

I conduct my analysis using a unique, hand-collected sample of 515 software EULAs.

This is a large subsample of the EULAs that I analyze in a recent paper.11 The current sample

includes EULAs for one or at most two representative products for each of 468 software

publishers who sell their products through their corporate website. The companies whose EULAs

are included in the sample include virtually all well-known software publishers as well as

hundreds of smaller firms. In addition, each product’s EULA is classified as either “business,” if

the product is directed to business users, and “consumer,” if the product is directed to the general

public or home office users, which allows for separate analyses of each sample.

To determine whether firms that delay disclosure offer more one-sided terms than those

that are forthcoming, I measure the “net buyer friendliness” of each EULA using a simple index.

This index is based on an analysis of 25 common standard contract terms that allocate rights and

risks between buyers and sellers.12 It includes terms pertaining to acceptance of the license

notices, scope of the license, restrictions on transfer, warranties and disclaimers of warranties,

limitations on liability, maintenance and support services, and conflict resolution. I assign a

negative one point score for each term that is more pro-seller relative to the default rules of

Article 2 of the Uniform Commercial Code (UCC), and a positive one point score for each term

that is more pro-buyer relative to the default rules. I give a score of zero if the contract is silent in

regards to the specified term, or if the specified term matches the default rule. The overall pro-

seller or pro-buyer bias of a given EULA is computed as the sum of the values for each of the 25

terms. While crude, this methodology captures the overall bias of the EULA in a way that allows

for an empirical analysis. I also relax the implicit assumption that each of the 25 terms matters

equally to buyers by using seven subindexes that isolate the bias in particular sets of terms.

11 Florencia Marotta-Wurgler, Competition and the Quality of Standard Form Contracts: An Empirical

Analysis of Software License Agreements, NYU Law and Economics Working Paper No. 05-11 (2005). 12 Id.

4

My main findings concern the relationship between contract bias and pre-purchase

accessibility. I find that companies that provide their EULAs to consumers only after purchase,

i.e., that utilize a rolling contract, do not offer more pro-seller terms than firms that make their

EULAs available pre-purchase. These results should be reassuring to those comfortable with the

logic of the ProCD ruling and the increasing presence of rolling contracts. In fact, contrary to the

concerns expressed by supporters of pre-sale disclosure, I find that those companies that force

consumers to click “I agree” to the terms of a EULA before completing their purchase actually

present somewhat more pro-seller terms than firms that offer rolling contracts. Sellers, it seems,

are not being sneaky by “hiding” one-sided contracts. Instead, sellers whose boilerplate is more

one-sided tend to make their contract harder to challenge by requiring buyers to unequivocally

accept it. These results hold for both consumer- and business- oriented products.

In summary, this paper tests and rejects the notion that “pay now, terms later” contracts

are worse for buyers. At least with respect to software license terms, buyers do not, on average

receive more pro-seller contracts when the terms are disclosed only after purchase.

Commentators and consumer advocates, then, should not be particularly concerned about rolling

contracts. It is important to note, however, that the tests in this paper are not designed to answer

the broader question of whether all EULAs, or standard form contracts in general, contain poor

quality terms according to some absolute standard. Rather, the results suggest that, to the extent

there are inefficiencies associated with standard form contracts, they are not made worse by

delayed disclosure. The results call into question the desirability of recent proposals urging for

mandatory pre-sale disclosure and proposals urging notice that terms will follow in situations

where disclosure is inconvenient or impossible.13

The paper proceeds as follows. Section II describes the debate over the desirability of

rolling contracts. Section III describes the general methodology and the sample of EULAs.

Section IV describes the pre-purchase accessibility of software EULAs. Section V discusses the

pro-seller and pro-buyer bias of EULA terms. Section VI contains the main analysis, examining

the bias in terms as a function of pre-purchase accessibility. Section VII concludes.

13 See Robert A. Hillman, On-line Boilerplate: Would Mandatory Website Disclosure of E-standard Terms Backfire? (forthcoming, Mich. L. Rev., 2006) (questioning the desirability of mandatory disclosure of terms in Internet transactions noting that, despite its advantage of low-cost compliance, mandatory disclosure may not increase readership, and hence will not induce sellers to compete for customers by including reasonable terms, but might make suspect terms more easily enforceable).

5

II. ROLLING CONTRACTS: THE DEBATE

A wide range of modern transactions are governed by standard form contracts.14 Whether

the transaction involves the purchase of a car or a gym membership, the buyer is generally

presented with boilerplate in a take-it-or-leave-it fashion with no opportunity to negotiate over

the terms contained therein. While the buyer might be able to determine the quantity and delivery

method, and might be able to bargain over price, secondary terms such as length of warranties,

forum selection, and remedies for breach of contract, will be part of non-negotiable boilerplate.

Legal academics agree that the use of standardized contracts facilitates mass commerce

by drastically reducing drafting and negotiating costs.15 Moreover, some of these savings can be

passed on in the form of lower prices. The use of boilerplate may also help reduce product

differentiation, thus helping buyers compare prices.16 And while some commentators are

concerned about the potential for seller overreaching in the form of abusive terms, others have

countered that, in addition to the legal protections granted by the UCC and doctrines of

unconscionability and reasonable expectations, competitive market forces ensure that buyers’

interests are taken into account.17

In traditional standard form contract exchanges the buyer is at least able to see the terms

of the contract before deciding whether to purchase the good. Commerce over the phone and

Internet, however, has led to the use of “rolling” or “pay now, terms later” standard forms. In

exchanges involving such contracts the buyer is not able to read the contract until after she has

paid for the good and received delivery. Rolling contracts have sparked heated debate. While

some believe that they are essential to the development of e-commerce and new exchange

mechanisms, others are concerned about their potential for abuse.

A. ProCD v. Zeidenberg and the Defense of Rolling Contracts

The scholarly debate largely originated with the 7th Circuit’s decision in ProCD v.

Zeidenberg.18 Judge Easterbrook, writing for the court, held that a standard form contract that

was shrink-wrapped inside a software box was enforceable under the UCC.

14 See W. David Slawson, Standard Form Contracts and Democratic Control of Lawmaking Power, 84

Harv. L. Rev. 529 (1971). 15See Todd D. Rakoff, Contracts of Adhesion: An Essay in Reconstruction, 96 Harv. L. Rev. 1173, at 1222

(1983); Friedrich Kessler, Contracts of Adhesion- Some Thoughts About Freedom of Contract, 43 Colum. L. Rev. 629 at 631 (1943).

16 Richard Posner, Economic Analysis of Law, 3rd ed. Little, Brown and Company (1986). 17 See Restatement (Second) of Contracts §211 (1979) and U.C.C. §2-302, and letters cited supra note 5. 18 Supra note 1.

6

The plaintiff, ProCD, created a database using information from various telephone

directories which it sold at two different prices. The cheapest version of the database was sold to

the general public and the more expensive version was offered to manufacturers and retailers

who used it to identify potential customers. To discriminate between these buyer types, ProCD

included a clause limiting the use of the database to non-commercial purposes in the license

agreement of the software sold at the lower price. The license was included in the software’s

CD-ROM and in the user’s manual, and also appeared on the computer screen every time the

software was run. But both the CD-ROM and the manual were inside the product’s box, which

could be opened only after purchase, and therefore the license constituted a rolling contract. The

license also stated that the buyer could return the software within 30 days if she did not find the

license agreeable.

Defendant Zeidenberg purchased the low-price version of the software in a retail store

and decided to ignore the restrictions in the license by selling ProCD’s database to other

commercial users at a markup, but less than ProCD’s commercial price. ProCD sued for an

injunction against Zeidenberg. The district court held in favor of the defendant, reasoning that

because at the time of purchase Zeidenberg could not have agreed to terms hidden in a box, those

terms were not part of the contract.

In reversing the district court’s decision, Judge Easterbrook first determined that software

licenses were ordinary contracts and thus governed by the law of contracts and the UCC.19 He

then noted that it would be impractical to both sellers and buyers if the law required sellers to

post the license terms in microscopic print somewhere on the box. He cited UCC section 2-

204(1), which states “[a] contract for the sale of goods may be made in any manner sufficient to

show agreement, including conduct by both parties which recognizes the existence of such a

contract.” He concluded that a valid contract was created when ProCD proposed a contract that

19 Courts generally assume that Article 2 of the UCC governs software licenses. See e.g. Novacore Techs.,

Inc. v. GST Communications Corp., 20 F. Supp. 2d 169, 183 (D. Mass. 1998); VMark Software, Inc. v. EMC Corp., 37 Mass. App. Ct. 610, 611 n.1, 642 N.E.2d 587 (1994); USM Corp. v. Arthur D. Little Sys., Inc., 28 Mass. App. Ct. 108, 119, 546 N.E.2d 888 (1989). But see I. LAN Sys., Inc. v. NetScout Serv. Level Corp., 183 F. Supp. 2d 328, 332 (2002) at 333 (stating that “software licenses exist in a legislative void” and that in the context of a dispute between business parties “Article 2 technically does not, and certainly will not in the future, govern software licenses, but for the time being, the Court will assume that it does”). See also Mark A. Lemley, Intellectual Property and Shrinkwrap Licenses, 68 S. Cal. L. Rev. 1239 (1995); Lorin Brennan, Why Article 2 Cannot Apply to Software Transactions, 38 Duq. L. Rev. 459, 545 (2000).

7

the buyer could accept by using the software after having had an opportunity to read the terms of

the license, as Zeidenberg had done.

While his decision hinged on general notions of contract formation, Judge Easterbrook

remarked on the potential efficiency losses of not enforcing software shrink-wrap licenses. In the

software industry, the majority of sales take place over the phone or the Internet where there is

no box to peruse. With downloadable software, the buyer often purchases a serial number that is

then used to activate the software. Failure to enforce the post-payment terms would subject

manufacturers to broad implied warranty terms and consequential damages. Judge Easterbrook

reasoned that such an arrangement would harm consumers by “driving prices through the ceiling

or return transactions to the horse and buggy age.”20 He concluded by stating “terms of use are

no less part of ‘the product’ than are the size of the database and the speed with which the

software compiles listings. Competition among vendors, not judicial revision of a package’s

contents, is how consumers are protected in a market.”21

In a decision a year later, the 7th Circuit reaffirmed that the holding in ProCD was not

limited to software licenses, but applied to all contracts where payment preceded terms. In Hill v.

Gateway 2000,22 the plaintiffs sued to deny enforcement of an agreement containing an

arbitration clause that was delivered together with a computer that the Hills had purchased from

Gateway by phone. In denying the Hills’ request, Judge Easterbrook noted that Gateway invited

acceptance by the buyers’ retention of the computer beyond the 30-day return period, and that by

doing so, the Hills accepted the contract. As in ProCD, the buyers in this case were given an

opportunity to review the standard terms and to return the product if they found them

disagreeable. In recent years, a number of courts have followed the ProCD logic and enforced

rolling contracts.23

Some scholars, agreeing with Easterbrook’s decisions, also noted that because the

majority of buyers rarely read the terms in form contracts it does not matter whether boilerplate

20 See ProCD, supra note 1, at 1452. 21 Id at 1453. 22 Supra note 2. 23 See e.g., I. Lan Systems, Inc. v. Netscout Serv. Level Corp., 183 F. Supp. 2d 328, 38 (D. Mass. 2002);

Mortenson Co. v. Timberline Software Corp., 998 P.2d 305, 307 (Wash. 2000); Lozano v. AT&T Wireless, 216 F. Supp. 2d 1071, 1073 (C.D. Cal. 2002); Rinaldi v. Iomega Corp. 41 U.C.C. Rep. Serv. 2d (Callaghan) 1143 (Del. Super. 1999); Scott v. Bell Atlantic Corp., 726 N.Y.S.2d 60, 64 (App. Div. 2001); Walton v. Experian, No. 02C5067, 2003 WL 22110788, (N.D. Ill. Sept. 9, 2003); O’Quin v. Verizon Wireless, 256 F. Supp. 2d 512 (M.D. La. 2003).

8

is made available before or after purchase.24 Failure to read should not be a source of concern,

however, since these scholars argue that sellers in well-functioning markets will compete for the

business of the marginal consumer. As long as a minority of informed buyers demands contracts

reflecting the general buyer population preferences and sellers are unable discriminate between

buyer types, sellers will offer the preferred terms in order to obtain the business of the informed

types.25 Rolling contracts need not interfere with this process, these scholars argue, since even if

sellers do not make the contract available before purchase, concerned buyers are still able to shop

for terms. For instance, prior to purchase, concerned buyers can contact the seller and request the

standard terms; or, after purchase, they can return the product if they find the terms disagreeable.

Clayton Gillette has argued that the interests of non-reading buyers’ might be internalized by

representatives such as the market, courts, and regulators. Assent by representation, he argues,

can satisfy the same objectives as personal assent.26

Those who object to recent proposals to regulate rolling contract terms offer

complementary arguments. Douglas Baird argues that several laws and regulations, such as the

Magnuson-Moss Warranty Act, offer enough protection against unfair terms.27 Alternatively,

sellers’ reputations and the forces of competitive markets will also ensure that buyers’ interests

are taken into account, even if buyers are presented with take-it-or-leave-it terms that cannot be

bargained over. In consequence, any effort to prescribe the terms in standard form contracts will

make buyers worse off because sellers will be unable to compete on the basis of the terms they

24 See Gillette, supra note 3; Avery Katz, Your Terms or Mine? The Duty to Read the Fine Print in

Contracts, 21 Rand J. of Econ. 518, 533 (1990) (arguing that non-drafting parties will generally find it too costly to read pro-seller standard terms, will assume that they have been included, and discount the price they are willing to pay accordingly); Melvin Eisenberg, Text Anxiety, 59 S. Cal. L. Rev. 305 (1985) (reasoning that it is rational for consumers to refuse to read the text in standard form contracts); Robert A. Hillman, On-Line Consumer Standard-Form Contracting Practices: A Survey and Discussion of Legal Implications, In: Is Consumer Protection an Anachronism in the Information Economy? Ashgate Press (2005) (noting that a survey of consumer behavior on the Internet reveals that consumers rarely read standard terms online). See also supra note 4.

25 See Alan Schwartz & Louis Wilde, Intervening in Markets on the Basis of Imperfect Information: A Legal and Economic Analysis, 127 U. Pa. L. Rev. 630 (1979); A. Schwartz & L. Wilde, Imperfect Information in Markets: The Examples of Warranties and Security Interests, 69 Va. L. Rev. 1387 (1983); George L. Priest, A Theory of the Consumer Product Warranty, 90 Yale L. J. 1297 (1981); A. Michael Spence, Monopoly, Quality, and Regulation, Bell Journal of Economics 6 (1975); But see Victor P. Goldberg, The ‘Battle of the Forms’: Fairness, Efficiency, and the Best-Shot Rule, 76 Or. L. Rev. 155, 165 (1997) (noting that in insufficiently competitive markets sellers may include restrictive terms in boilerplate, even if doing so might mean losing some reading buyers, since more may be gained by offering restrictive terms to non-readers than by losing the business of the readers).

26 See supra note 3 at 683. 27 See Baird, Letter supra note 5.

9

offer. Overall, Baird suggests that additional regulation in this area should be undertaken only

with caution, as it may hinder useful innovation.28

In a related vein, Alan Schwartz reasons that in situations where sellers would find it

infeasible to include terms pre-purchase, a rule requiring sellers to notify buyers before purchase

that additional terms are to follow is problematic.29 Buyers will not receive any new information

if sellers are required to disclose that additional terms will follow, Schwartz argues, because they

already know that there will be additional terms. Requiring sellers to note that additional terms

will follow just provides an opportunity for buyers to sue over whether the notice is adequate.

Litigation over this issue may override buyers’ real concern, which is that the product did not

perform as expected, but the buyer’s claims are either precluded or too idiosyncratic to sway a

jury. Schwartz concludes that a rule requiring pre-purchase notice of additional terms may just

create an opportunity for wasteful litigation.

Finally, a rule requiring pre-sale disclosure by Internet retailers might seem less

problematic, as sellers can easily post their standard terms on their website. Such a rule might

seem likely to facilitate shopping, increase competition, and encourage sellers to offer reasonable

terms. Robert Hillman, however, argues that such a rule may backfire. If buyers do not read

standard terms at all, mandatory disclosure will not increase readership and term shopping. It

might, however, allow suspect terms to be more easily enforced because the contracting process

appears more legitimate.30

B. Concerns About Rolling Contracts

The decisions in ProCD and Hill have left others deeply worried, and courts in a handful

of jurisdictions still refuse to enforce terms that have not been explicitly agreed to.31 Most

fundamentally, rolling contracts threaten to eliminate the contract-based-theory of mutual

28 Id. 29 See Schwartz, Letter supra note 5. 30 See Hillman supra note 13. 31 See e.g. Step-Saver Data Systems, Inc. v. Wyse Technology, 939 F.2d 91, 102 (3d Cir. 1991). Arizona

Retail Sys. Inc. v. Software Link 831 F. Supp. 759 (D. Ariz. 1993); U.S. Surgical Corp. v. Orris, Inc. 5 F. Supp. 2d 1201 (D. Kan. 1998); Klocek v. Gateway, 104 F.Supp.2d 1332 (U.S.D.C. Kan., 2000) (holding that material terms in standard forms are not binding unless expressly accepted by the buyer). But see Davidson & Assocs. v. Internet Gateway, 334 F. Supp. 2d 1164 (D. Mo. 2004) at 4 (distinguishing the facts from Klocek by stating that transaction involved the sale of services and that “Mortgage Plus had to affirmatively click the ‘Yes’ button in assenting to the Software Licensing Agreement as a prerequisite to installing the DocMagic software”).

10

assent.32 In particular, some courts, as well as a group of scholars and consumer advocates, fear

the potential for abuse presented by rolling contracts. By concealing the standard terms until

after the buyer has already purchased the good, sellers might systematically take advantage of

buyers and hide unfavorable terms.

For example, in explaining a seller’s preference for post-payment disclosure, Stewart

Macaulay writes that “[m]anufacturers of consumer goods seek to gain a wide freedom from

contract by packing inside the box contract clauses that attempt to repeal various laws that

businesses dislike.”33 Jean Braucher argues that marketing products to consumers without

effectively communicating standard terms before payment is both unfair and deceptive, as well

as a violation of various federal and state acts.34 She notes that post-payment disclosure of terms

may also suppress benefitial competition on terms. For instance, software companies that offer

no warranty, she claims, have incentives not to disclose that fact in advance, because disclosure

might scare away some customers. Finally, she points out that disclosure of the terms prior to

purchase is particularly cheap and easy where companies sell their products online such as in the

case of software, as it requires only posting the standard terms on the website.

Behavioral and psychological factors have also been emphasized by those who see rolling

contracts as a source of market failure. Russell Korobkin, for example, suggests that consumers

tend to focus on the salient or attention-grabbing features of a product, such as price, quantity,

and warranty terms.35 Consequently, sellers will compete only on those salient attributes while

providing low-quality non-salient terms. But not disclosing terms until after purchase is another

way to render terms less salient, increasing the potential for abuse.36

32 See supra note 6. See also Richard Epstein, Response to Letter of Thomas E. Workman, Life Insurance Counsel addressed to the Members of the House of Delegates of the American Bar Association, January 29, 2003 (citing Workman’s initial remark that “we find confusing […] the apparent view that delayed boilerplate language, not available to customers when they make their software acquisition decisions, represents freedom of contract to anyone but the drafter-namely, the software manufacturer.”); Melissa Robertson, Is Assent Still a Prerequisite for Contract Formation in Today’s E-Conomy? 78 Wash. L. Rev. 265 (2003).

33 See Macaulay, Freedom of Contract, supra note 6 at 807. See also Listserve Comment by Stewart Macaulay, University of Wisconsin School of Law Common Sense and Contracts Symposium: The Gateway Thread - AALS Contracts Listserv, 16 Touro L. Rev. 1147 (2000) at 1148-1149 (stating that “Judge Easterbrook and the judges who have followed his opinion tell us that misrepresentation is the oil that lubricates capitalism . . . [i]t is okay for Gateway to hide what it is doing”).

34 See Braucher supra note 8. 35 See e.g., Russell Korobkin, Bounded Rationality, Standard Form Contracts and Unconscionability, 70 U.

Chi. L. Rev. 1203, 1217 (2003); Rakoff, supra note 15. 36 J. Edward Russo, The Value of Unit Price Information, 14 J. Marketing Rsrch. 193, 194 (1977) (noting

that consumers make more accurate decisions when information is displayed in a convenient way).

11

III. METHODOLOGY AND SAMPLE

A. A Test Based on EULAs of Software Sold over the Internet

The previous section illustrates that commentators have sharply different views about

whether rolling contracts are good, bad, or irrelevant. Ultimately, however, whether rolling

contracts are more one-sided than regular standard terms is a fairly straightforward empirical

question. This paper attempts to answer this question by exploiting the unique nature of software

products sold online. In particular, using a large sample of End-User License Agreements

(EULAs) of software products that are sold over the Internet, I examine whether terms offered

after purchase are, or are not, more pro-seller than those available pre-purchase.

For several reasons, this is a particularly nice setting in which to test whether rolling

contracts are unusually exploitative. First, as we will see, software companies vary greatly in the

way they present their EULAs to potential buyers. Some present their terms prominently on their

website prior to purchase. Others use rolling contracts, disclosing terms only after purchase of

the software. Still others make their terms available prior to purchase, but effectively hide the

contract deep in the innards of a complex website.

Second, although rolling contracts date back many years, several recent and important

cases, as mentioned above, stem from transactions involving new technologies such as computer

software and hardware.37 They are particularly pervasive in e-commerce transactions, which

were estimated at $69.2 billion for 2004 and are growing in importance each year.38 An analysis

of contracts for software sold online thus places us at the core of the debate.

Third, the product itself, non-customized or “pre-packaged” software, is homogeneous

enough to make for a meaningful comparison of the EULAs from different companies. Even if

the products themselves vary in their purpose and function (for instance, anti-virus software

versus a program for estimating contractors’ bids), program failure is typically caused by generic

factors such as bugs, buyer misuse, or compatibility problems involving the buyer’s hardware or

other software. Consequently, sellers of many different types of software tend to protect

themselves from certain liabilities in similar ways. However, to the extent that different types of

37 Carnival Cruise Lines v. Shute, 499 U.S. 585 (1991) (holding that a forum selection clause in the back of

a ticket which the plaintiffs received post-purchase was an enforceable contract). See also Gillette, supra note 3, for a discussion of how many rolling contract cases involve new technologies.

38 See U.S. Census Bureau, Quarterly Retail E-Commerce Sales, 4th Quarter of 2004. http://www.census.gov/mrts/www/data/pdf/04Q4.pdf.

12

software are not directly comparable, we can still use within-category variation (i.e., include

category dummy variables in the regressions) to test whether rolling EULAs are more biased.

B. Sample Collection

The sample analyzed in this paper is a large subsample of that recently analyzed in

Marotta-Wurgler.39 In that paper, I present a comprehensive analysis of EULA terms and

examine the role of market competition in shaping these terms. The starting point for obtaining

the sample of EULAs used here was the Software Industry Directory 2005 CD-ROM, a

comprehensive list of 7,700 software development and publishing companies. The companies in

the Directory are mostly U.S.-based and range from $50,000 in sales to giants such as Microsoft

and IBM. All market software of one type or another to U.S. buyers. For each company in the

Directory, I manually determined whether the company sold its software online through its

corporate website, i.e. where a buyer could select a product and click all the way through to

payment and checkout.

Limiting the sample to companies that sell their products directly through their website,

although they may utilize other distribution channels, enables me to clearly document how firms

differ in the way they present EULAs to buyers before purchase. However, since virtually all

prepackaged software companies sell their products, at least in part, through their website, this

sample selection procedure should be interpreted mainly as a way of weeding out thousands of

custom software publishers (e.g., systems integrators) and resellers who do not actually

manufacture the software sold through their website. Custom developers do not offer non-

negotiable standard terms. And resellers do not offer any additional data, since the EULA

attaches to the product and not the particular vendor. Finally, many other companies listed in the

Directory are actually divisions of bigger companies, further reducing the number of

independent firms (and EULAs) to consider.

For each company in the remaining sample, I chose one representative product that could

be purchased online. I selected the company’s flagship software package, when it was apparent;

otherwise, I chose a product at random. To test whether companies impose poorer standard terms

on unsophisticated buyers, I also recorded whether the product appeared to be targeted to a

39 Marotta-Wurgler supra note 11.

13

consumer or business user.40 A few dozen companies offer a “business” and “consumer” version

of the same product. For these companies, I collected EULAs for both versions of the product.

The next step was to obtain the EULA for each of the selected products. Some companies

make their licenses easily available on their website, while others require some (and perhaps

considerable) trial-and-error searching. If the EULA was available anywhere on the site, I

collected it. Otherwise, I emailed the company asking if the EULA was available on their

website (to double-check that it was not) and if not, to please email or fax it to me. Most

companies complied, although a small fraction did not answer repeated requests, and thus could

not be included in the sample. To reduce selection bias further, I purchased a handful of products

to obtain the EULA.

This process led to a sample of 515 EULAs. Since, as mentioned above, I collected two

EULAs for 47 companies, the sample covers 468 distinct companies. The sample includes

EULAs from almost all well-known software publishers and hundreds of smaller ones.

I also collected several other product and company characteristics for use as control

variables. I recorded the selected product’s price, recognizing that all else being equal, good

terms for consumers are more costly to provide. Finally, many companies offer a free version of

their products for buyers to try for a limited time period. In some cases, these trial versions

contain the EULA of the retail version of the product, although generally they do not. Still, I

noted whether a trial version of each product is available, since this enables buyers to verify

whether the product suits their needs, is compatible with their system, and so forth.

With a few exceptions, most licenses in the sample are of unlimited duration and single-

user. Because a multi-user license would tend to have a higher price, I recorded whether the

license is single or multi-user. Another important license feature is whether the license is

“developer” or “standard.” Developer licenses allow the buyer to use the software to develop

derivative products, and hence are common for products that aid programmers in creating new

software. Since some of the rights granted under the developer EULA differ from standard

EULAs, I recorded this aspect of the license as well.

Other company-level characteristics in the Directory dataset (or, where missing, gathered

from other sources, such as Hoover’s Online, or direct correspondence with the company)

40 For example, a product entitled Cyber Sentinel 3.0 Home Edition, designed to prevent children from

accessing adult sites from a home computer, is categorized as “consumer” software, while software products targeted to large firms, such as Client Management Services v1.30, is categorized as “business” software.

14

include annual revenues, whether a company is publicly traded or private, the year of

incorporation, and number of employees.

Table 1 provides summary statistics for the company and product characteristics. Panel A

shows that average sales are $371 million, but the standard deviation of this figure is $4.27

billion, indicating that the average is driven by very large companies. The median sales are $2

million. Publicly-traded companies make up 15% of the sample. Panel B shows that a little under

half of the products in the sample, 41%, are oriented towards consumers (or home businesses)

rather than proper businesses. About 70% of the companies offer a trial version of their software.

Last, I classified each product into one of 146 software product categories, from antivirus

to voice recognition software. I borrow these categories from Amazon.com, the largest Internet

retailer, which uses them to organize the tens of thousands of software products that it markets.41

IV. PRE-PURCHASE ACCESSIBILITY OF EULAS

A. Measuring Accessibility

A key variable in this study is the accessibility of the selected product’s EULA on the

company’s website. The most basic aspect of accessibility is whether the EULA is findable at all

on the website, or instead is made available only post-purchase, packaged together with the

product. To capture this distinction, I label as “Findable” those EULAs that can be found

somewhere on the sellers’ website. EULAs that are only made available until after purchase are

not findable.42 By definition, rolling contracts are not findable.

Table 2 gives summary statistics on this dimension of EULA accessibility. About 48% of

sellers make their EULA available in their website, whereas 52% provide the EULA only until

after purchase. A very small percentage of sellers offers its software with no EULA at all, and

since it is unclear whether to classify these as either Findable or not, I exclude these from the

subsequent analysis.43

41 See Marotta-Wurgler, supra note 11, for a breakdown of licenses by Amazon.com classification. 42 Many sellers ship the software and the EULA to a physical address provided by the buyer. Sometimes,

however, buyers can download the software directly from the seller’s site. In these cases, if the buyer is unable to access the EULA until after she has paid for the software, the EULA will still be coded as “not findable,” since it is not accessible until after purchase, which is the critical distinction.

43 Arguably, products sold without a EULA are among the most buyer friendly, as they are sold outright and not licensed and include the implied warranties of the UCC. I thank Lewis Kornhauser for this point. I exclude these cases from my analysis of the effects of contract location on contract bias because the location of a contract that does not exist is hard to define. In any case, there are only four companies that use no EULA whatsoever and that would otherwise be in my sample. Not surprisingly, all are quite small.

15

Given that the EULA is findable somewhere on the website, I also measured whether it

was relatively more or less accessible. To do so, I started by determining the most obvious or

most natural navigational path through the website that a consumer would use to purchase the

product. This is usually quite straightforward, since commercial websites are designed to get

buyers to purchase as easily as possible, starting from the homepage. I then counted the

“distance,” in terms of the minimum number of clicks, between the most obvious click path to

purchase and the actual location of the EULA. EULAs that were a greater number of clicks away

from the most natural path of purchase were deemed less accessible, albeit still findable. A buyer

belonging to (or at least aspiring to) the “informed minority” would find it easier to comparison

shop among firms with low distance scores.

For example, consider a buyer who wishes to purchase Internet Security 7.0 from

McAffee’s website. After selecting the product and proceeding to check out, he will discover that

before he is allowed to enter his credit card information, he must agree to the product’s EULA by

clicking on “I agree” below a scroll box that contains the standard terms. Because the EULA is

directly on the most natural click path for purchasers and requires no extra clicks to find, it has a

distance score of zero. As Table 2 shows, 4.7% of sellers, or about 10% of those that make the

EULA available on the website site, score a zero, indicating the maximum possible degree of

pre-purchase accessibility.

Symantec, on the other hand, presents a license for its Norton Antivirus 2005 product that

is a minimum of two clicks away from the most obvious path of purchase. A link at the bottom

of the Symantec homepage, entitled “license agreements,” provides links to the EULAs of

Symantec products. Thus it takes a buyer one click from the main page (which is always used as

the starting point of the most natural path of purchase) to access the list of EULAs, and a second

click to see the EULA of the desired product, for a total distance of two clicks.

As mentioned above, McAfee forces the buyer to acknowledge that he agrees to a license

before he is allowed to pay. The buyer may or may not read the terms in the screen, but by

clicking “I agree” he knows that the software is governed by a license and that he has agreed to

its terms. The same is true for companies that force buyers to check “I agree” in a box before

entering their credit card information, although they may not show the actual EULA without one

more click on the provided link. In these peculiar in-between cases, I record the distance as 0.5,

because, although the buyer must actively acknowledge the existence of a contract, he must also

16

click once to see its terms. Table 2 shows that 1% of all sellers, or 2% of sellers whose EULAs

were Findable, display their license to buyers in this manner. To more clearly capture the

distinction between EULAs where the buyer is forced to acknowledge a EULA from those where

it is not explicitly presented prior to purchase (and can only be found through a determined

search), I label EULAs with distance scores of 0 and 0.5 as “Forced.” About 5.7% of sample

firms, or 11.8% of those that make their EULAs findable, fall into this group.

Some products’ EULAs are a minimum of six clicks away from the most natural path of

purchase. Given the labyrinthine structure of some corporate websites, in which each page links

to many others, such licenses can be extremely difficult to find.

B. Determinants of Pre-Purchase Accessibility

To see whether certain products’ EULAs are more likely to be made available before

purchase, Table 3 shows regressions where the dependent variable is a measure of EULA

location and the independent variables include product and firm characteristics. The first two

models study whether or not the EULA is findable anywhere on the website. Firm size, as

measured by revenue, is the only firm characteristic that is significantly positively related to

whether a EULA is findable. Specifically, the coefficient on the natural log of revenue implies

that a one unit increase in this variable increases the probability that the EULA is findable by

5%. A plausible interpretation is that larger firms are more likely to have in-house counsel who

may advise making the EULA available before purchase to increase its enforceability.

Developer licenses are also more likely to be findable, perhaps because license terms are

more important for users of such products. Many of the benefits that these users obtain from the

product hinge on how restrictive is the license. Younger software publishers are also more likely

to make their EULAs available on the website. One possibility is that buyers feel less compelled

to thoroughly examine the terms of a familiar seller whose reputation is well-known. A simpler

explanation is that older firms are less likely to be exclusively software companies. They may

sell many products online that are completely unrelated to software, thus reducing the likelihood

that the structure of their website will specifically cater to the software buyer. The inclusion of

product category dummies (indicating different Amazon.com “markets”) does not change these

results, indicating that they hold within categories as well.44

44 That is, the coefficient on log revenue, for example, reflects the comparison of Symantec and smaller

anti-virus companies, not Symantec and small companies that sell graphics or word processing software.

17

The second pair of models further confirms this notion. Larger firms are also more likely

to force buyers to agree to the terms of the EULA before allowing them to purchase. This again

may reflect the advice of sophisticated in-house counsel. On the other hand, controlling for size,

public firms are slightly less likely to force consumers to acknowledge a EULA. These results

are less robust to the inclusion of market fixed effects.

In the last pair of models, the dependent variable is the distance score, given that the

EULA is findable. These estimates indicate that (given the EULA is findable) it requires more

clicks to find EULAs when the seller is larger or younger. The explanation most likely involves

the fact that larger companies and, controlling for size, younger companies tend to have more

detailed websites—the former because the company simply has more products to describe and to

sell the latter because their webpages tend to be more sophisticated.

Interestingly, none of the regressions in Table 3 indicates any relationship between the

EULA location and whether the product is consumer or business-oriented. If it were the case that

sellers took advantage of naïve consumers by hiding their exploitative terms until after purchase,

we might expect to see a significant negative relationship between consumer-oriented products

and findability. In fact, rolling contracts are used with about equal frequency for both business-

and consumer-oriented software products.45

V. EULA Bias

A. Measuring Overall Bias

A software program’s EULA delineates the rights and obligations of the buyer and the

seller. EULAs of pre-packaged software tend to address similar issues, such as restrictions on use

and transfer, warranty disclaimers, and limitations on liability, regardless of whether they apply

to an anti-virus or a graphics program. As discussed in more detail in Marotta-Wurgler, software

companies generally license rather than sell their software, and other than the additional post-

transfer restrictions sellers can impose through licensing rather than selling, the legal

implications regarding enforcement of the EULA and its terms are the same as those regarding

45 The specifications in Table 3 are all linear probability models. I also tried using logit specifications for

the first four models and obtained identical results. In the case of the last two specifications, I redefined the variable “Distance” as ln((Distance + 0.01)/(6.01-Distance)) to obtain a more bell-shaped dependent variable. This again led to similar results as presented in Table 3, with the exception that ln revenue is positively significant and ln age is negatively significant in the fifth specification.

18

regular standard forms.46 Many courts have held that the sale or license of software should be

interpreted as the sale of a good within the meaning of the UCC.47 Consequently, when faced

with a dispute over the validity of a software EULA or a particular term contained therein, courts

have relied on Article 2 of the UCC to determine its enforceability.

To measure the bias of a given EULA, I use the “EULA bias index” methodology from

Marotta-Wurgler. This is a simple index that takes account of 25 common standard contract

terms that allocate rights and risks between buyers and sellers. It is defined as follows. For each

of 25 terms, a negative one point score is assigned if the term is more pro-seller than the default

rules of Article 2 of the UCC, a positive one point score is assigned if the term is more pro-buyer

relative to those rules, and a zero score is assigned if the contract is silent in regards to the

specified term, or if the specified term matches the default. The scores for each term are then

summed up to construct an overall measure of “net buyer friendliness” for that particular EULA.

For example, a provision found occasionally in EULAs entitles buyers to receive

software updates and upgrades for a specified period after purchase. Since there is no default rule

in Article 2 mandating such an entitlement and since, other things being equal, a buyer would

clearly prefer a EULA that entitles her to receive updates than one without such an entitlement,

the presence of this provision is awarded a positive one point score in the overall index.

Table 4 lists all 25 terms, details how each is scored, and provides the mean and standard

deviation for each term’s score. To identify important terms, I rely on the discussion in the

practitioner-oriented manual by Michael Overly and James P. Kalyvas, as well as the textbook

on electronic commerce by Ronald J. Mann and Jane K. Winn.48 I concentrate on terms that

generally govern buyers’ “normal use” of the software, excludes terms dictating sellers’ use of

buyers’ personal information, and terms describing copyright or patent rights (as they belong

exclusively to the realm of federal intellectual property law), and terms that would be deemed

unenforceable by any court, such as disclaimers of good faith.

The 25 provisions that comprise the overall index fall into seven categories. The first

category, Acceptance of License, includes a term designed to alert the buyer of her options

should she find the license or product disagreeable. The second category, Scope of License,

46 Marotta-Wurgler supra note 11. 47 See e.g. Advent Sys. Ltd. v. Unisys Corp., 925 F.2d 670, 676 (3d Cir. 1991); Step-Saver Data Systems v.

Wyse, 939 F.2d 91 (3d Cir. 1991); Downriver Internists v. Harris Corp., 929 F.2d 1147, 1150 (6th Cir. 1991). 48 Michael Overly and James P. Kalyvas, Software Agreements Line by Line, Aspatore, Inc. (2004);

Ronald J. Mann and Jane K. Winn, Electronic Commerce, Aspen Publishers (2002).

19

contains four terms restricting the buyer’s use of the software. The third, Transfer of License,

includes two terms that limit the buyer’s ability to sell or transfer the software. The fourth

category is Warranties and Warranty Disclaimers and is comprised of seven terms that delineate

the degree and type of warranty protection offered to buyers. For example, it notes whether

sellers offer any express or limited warranties, and whether they disclaim the implied warranties

of merchantability and fitness for a particular purpose under the UCC.49 The fifth category,

Limitations of Liabilities, contains seven terms specifying the extent of the seller’s liabilities for

different types of buyer loss arising out of use of the software, as well as terms allocating risks of

loss. It also covers the buyers’ available remedies, if any, for such losses. The sixth category,

Maintenance and Support, takes into account whether the base price of the software includes

these services.50 The last category, Conflict Resolution, includes three terms that restrict a

buyer’s choices regarding her decision of where to sue and under what law (forum selection and

choice of law clauses), whether to have a jury trial (arbitration clauses), and how legal fees are to

be allocated as a result of a legal dispute. I again refer the reader to Marotta-Wurgler for

additional details on the construction of the bias index.

The advantage of this overall index is that it is transparent and allows for empirical

comparisons of EULAs. However, because each of the 25 terms is given the same weight, a

built-in assumption is that each terms matters equally to buyers. I relax this assumption by

measuring the relationship between contract location and the bias of particular sets of terms,

considered separately, rather than the overall bias.

B. EULA Bias Summary Statistics

As constructed, the maximum possible overall bias score in the overall bias index is 7,

corresponding to a very buyer-friendly EULA relative to the UCC default rules. The minimum

49 The index does not consider whether the warranties comply with the Magnuson-Moss Warranty Act for

various reasons. 15 U.S.C. §§2301-2312 (1976). First, the act regulates the disclosure of consumer warranties and not whether warranties should be given or not. Second and most importantly, it is still not settled whether Magnuson-Moss applies software products. The act applies to “any tangible personal property which is distributed in commerce and which is normally used for personal, family, or household purposes.” To this date and after heated discussion, there is still no consensus on whether software should be interpreted as “tangible personal” property. See e.g., Symposium, Warranty Protection for High-Tech Products and Services, Federal Trade Commission (2000), at http://www.ftc.gov/bcp/workshops/warranty/.

50 Some sellers offer a menu of maintenance and support services for an additional price or under a separate contract, while others just include basic maintenance and technical support in the products’ base price. To be able to make meaningful comparisons among firms, I counted whether a company includes a base level of maintenance and support (over 30 days) in its base price, and then used the product’s price as a control variable instead of recording whether a firm offers support under separate contract.

20

attainable score is -20, indicating a contract that disclaims all remedies and greatly restricts

buyers’ use of the software. It is important to note that the overall index score for each EULA

should not be interpreted as measuring whether the contract is biased for or against the seller in

an absolute sense, but rather as a measure of bias relative to the default rules of Article 2 of the

UCC. It would be more difficult and problematic to construct an absolute measure of bias, as one

would have to consider elements unique to specific software product categories. Indeed, one

reason the range of possible scores is not centered at zero is that the default rules of the UCC

tend to benefit buyers.

Table 4 shows that the average overall bias index score in the sample is -5.63, meaning

that, on average, contracts are considerably more pro-seller than the default rules of the UCC.

The minimum overall score in our sample is -15 and the maximum is 2. Thus, there is a very

wide range in how buyer-friendly are software license agreements. These summary statistics are

quite close to those reported in Marotta-Wurgler, which is not surprising since the current sample

contains 515 out of the 647 EULAs that I study there.

VI. ARE “PAY NOW, TERMS LATER” EULAS WORSE FOR BUYERS?

A. Overall Bias and Accessibility

We are now ready to address the central debate involving rolling contracts, namely,

whether hidden terms are more one-sided than terms made available before purchase. Table 5

presents the average overall bias of EULAs by the degree of their accessibility. It also compares

the mean bias of EULAs of a given accessibility to the mean bias of rolling EULAs (abbreviated

PNTL in the table, for Pay Now, Terms Later).

The results show that those EULAs made available before purchase are actually

significantly less buyer-friendly than rolling EULAs. In particular, findable EULAs are less

buyer-friendly than rolling EULAs by 0.62 points, a statistically significant difference. Findable

EULAs can be broken down further, into those that are Forced and those that are not. While

Findable EULAs of both types are significantly more pro-seller than rolling contracts, it is

particularly surprising that the Forced EULAs—those that the buyer must explicitly

acknowledge before buying the product—are the most one-sided contracts of all. This is in

striking contrast to the view that sellers hide more one-sided contracts.

21

Table 6 analyzes these effects using regression analysis. The dependent variable is the

overall bias index. The first column shows the most basic test, where the dummy Findable is the

only independent variable. This simply replicates the result in the first row of Table 5. The

second column uses Forced as the only independent variable. The third column shows that

among the subset of EULAs that are not Forced but still Findable, there is no relationship

between bias and how distant (i.e., how many clicks away) the EULA is from the most obvious

path of purchase. This is another strike against the view that sellers hide particularly one-sided

contracts. The fourth column separates the effects of Forced and Findable by including both in

the regression. It shows that a Forced EULA scores an additional 0.48 points lower, on average,

than one that is Findable but not Forced, but the difference between those two groups is not

significant. These results could also be anticipated from Table 5.

Figure 1 illustrates these results graphically, displaying how the average overall bias of

EULAs varies with accessibility. As suggested by the tables, the figure shows that Forced

EULAs are more pro-seller than all other types of EULAs, while rolling contracts are in fact the

least biased of any category.

Of course, we cannot immediately conclude that a EULA that is two-thirds of a point or

one point more pro-seller would be perceived by a buyer as worse in any meaningful sense.

However, for my purpose, this is a secondary issue. The key point is that rolling contracts are

clearly not more pro-seller than those made available before purchase. Put simply, the evidence

rejects the main arguments presented against rolling contracts.

B. The Effect of Buyer Type, and Seller and Product Characteristics

The remaining specifications of Table 6 add control variables and split the sample into

consumer- and business-oriented products. The effects of Forced and Findable remain generally

similar when controlling for product category fixed effects, company size and age, product price,

and the consumer product dummy, but with the inclusion of these controls Forced and Findable

usually are not both significant. Consistent with the results in Marotta-Wurgler, larger companies

and younger companies tend to present more pro-seller terms.51 The size relationship may have

to do with the fact that larger companies are more likely to obtain legal advice from counsel who

may take more precautions to limit exposure to liability. The age relationship may involve seller

reputation. Older companies may create a more homogenous customer base over time and thus

51 Id.

22

be better able to provide insurance for certain types losses, as opposed to companies with a very

diverse customer base. However, this is speculation.

Critics of rolling contracts have been particularly concerned that sellers will take

advantage of unsophisticated non-business-type buyers who are believed to lack the resources,

knowledge, and ability to comparison shop for terms. Whereas Table 3 showed that sellers do

not hide EULAs of products directed to the general public more than they do their business-

oriented products’ EULAs, the coefficient on Consumer in Table 6 tests whether non-business-

oriented products have EULAs that are more pro-seller. Again consistent with the results in the

Marotta-Wurgler analysis, there is no consistent relationship between whether the software is

directed to members of the general population and the overall bias, thus lending no support to the

hypothesis that sellers take special advantage of non-business buyers.

I also consider a related hypothesis. If non-business buyers are less sophisticated and

resourceful than business buyers, perhaps sellers will only try to hide one-sided terms for

consumer-oriented products. The last two specifications test this by examining whether EULA

location affects overall bias for a given buyer type. The results show that rolling contracts are not

worse for either type of product; rather, Forced and/or Findable contracts are worse for both

business- and consumer-oriented products.

C. EULA Bias by Subindex and Accessibility

As noted earlier, a drawback of the overall index is that it implicitly assumes that all

provisions matter equally to buyers. However, suppose that buyers care only about certain

provisions, and that sellers hide contracts that are worse on those provisions but not worse

overall. For instance, Korobkin suggests that buyers value only salient terms of standard form

contracts, such as warranties.52 If that is the case, then we should ask whether sellers who use

rolling contracts offer worse warranties, rather than worse terms overall.

This section addresses this question by studying the relationship between EULA location

and subindex bias scores. I create seven subindexes that capture the net bias of each group of

terms detailed in Table 4. For example, the transfer of license ubindex is comprised of two terms:

“are there restrictions on transfer?” and “can the licensee transfer the software to an end user

who accepts the license terms without the licensor’s prior permission?” I then repeat the previous

regression analysis but use the bias of each subindex as the dependent variable.

52 See Korobkin, supra note 35.

23

The results in Table 7 show that hidden EULAs are not worse on any of the main groups

of terms that comprise such contracts. In other words, hidden contracts are not worse in any

particular dimension. In fact, consistent with earlier results, the only significant pattern is that

sellers who make their EULAs more accessible actually offer worse terms in various respects.

This is the true for transfer of license, limitations on liability, and conflict resolution provisions.

Given that conflict resolution provisions such as forum selection and arbitration are often

mentioned as the most regular form of seller “abuse,” it is noteworthy that software publishers do

not rely on delayed disclosure as a way of imposing such terms.53

Finally, in results that are omitted to save space, I repeat the regressions in Table 7 but

restrict the sample to consumer-oriented products only to investigate whether sellers take

advantage of consumers by hiding particular one-sided terms. I find that with the exception of

limitations on liability, unsophisticated consumers do not receive worse terms than business

buyers. And, they tend to receive more pro-buyer terms with respect to transfer of license and

conflict resolution.

VII. SUMMARY AND IMPLICATIONS

“Pay now, terms later” or “rolling” contracts are increasingly common with the

acceleration of commerce over the Internet and telephone. The goal of this article is to address

the debate about whether rolling contracts impose harsher terms on buyers from an empirical

perspective. I collect and analyze the terms of several hundred end-user license agreements for

software packages sold online. The wide variation in the degree of pre-purchase accessibility of

these EULAs enables me to give a fairly convincing answer to whether rolling contracts do, or

do not, offer worse terms to buyers.

I find that the terms included in rolling contracts are not systematically more pro-seller

than those included in contracts disclosed before purchase. In fact, contrary to fears often

associated with rolling contracts, I find that contracts displayed pre-purchase are actually

somewhat more pro-seller than rolling contracts. I find similar results after controlling for

product price, seller size, consumer- versus business-product orientation, and other effects.

53 See Paul D. Carrington, Regulating Dispute Resolution Provisions in Adhesion Contracts, 35 Harv. J. on

Legis. 225 (1998); Paul H. Haagen, New Wineskins for New Wine: The Need to Encourage Fairness in Mandatory Arbitration, 40 Ariz. L. Rev. 1039 (1998); Jean R. Sternlight, As Mandatory Binding Arbitration Meets the Class Action, Will the Class Action Survive?, 42 Wm. & Mary L. Rev. 1 (2000).

24

It is important to note that my tests are not designed to address whether sellers offer poor

quality terms in an absolute sense, but rather they are focused on the comparative analysis

between terms that are made available pre-purchase and those that are not. Thus, the appropriate

conclusion is that to the extent that there are inefficiencies associated with standard-form

contracts, they are not made worse by delayed disclosure. This conclusion is encouraging to

supporters of Judge Easterbrook’s rationale for enforcing rolling contracts. It also implies that

regulation directed at eliminating or reforming rolling contracts—such as the recent proposals to

revise section 2-207 of the UCC to require sellers to follow specific disclosure practices—is not

warranted, nor is the strong opposition to the Uniform Computer Information Transactions Act

on the grounds that it allows for delayed disclosure.54

54 See note 5, supra.

25

BIBLIOGRAPHY

Barnett, Randy E. “The Sound of Silence: Default Rules and Contractual Consent.” Virginia Law

Review 78 (1992): 821-66.

Barnett, Randy E. “Consenting to Form Contracts.” Fordham Law Review 71 (2002): 627-45.

Bern, Roger C. “‘Terms Later’ Contracting: Bad Economics, Bad Morals, and a Bad Idea for a

Uniform Law, Judge Easterbrook Notwithstanding.” Journal of Law and Policy 12

(2004): 641-796.

Bishop, Todd. “Should Microsoft Be Liable for Damages?” Seattle Post Intelligencer. September

12, 2003.

Brandt, Andrew. “Click With Caution: User Licenses Get Tough.” PC World.com. April 9, 2001.

Brandt, Andrew and Wallace, William. “What Have You Signed Away Today?” PC World

Magazine. August, 2001.

Braucher, Jean. “Delayed Disclosure in Consumer E-Commerce as an Unfair and Deceptive

Practice.” Wayne Law Review 46 (2000): 1806-67.

Brennan, Lorin. “Why Article 2 Cannot Apply to Software Transactions.” Duquesne Law Review

38 (2000): 459-589.

Carrington, Paul D. “Regulating Dispute Resolution Provisions in Adhesion Contracts.” Harvard

Journal on Legislation 35 (1998): 225-32.

Eisenberg, Melvin A. “Text Anxiety.” Southern California Law Review 59 (1985): 305-311.

Epstein, Richard. Response to Letter of Thomas E. Workman, Life Insurance Counsel addressed

to the Members of the House of Delegates of the American Bar Association, January 29,

2003.

Federal Trade Commission, “Warranty Protection for High-Tech Products and Services.”

Symposium transcript: http://www.ftc.gov/bcp/workshops/warranty/. (2000).

Gillette, Clayton P. “Rolling Contracts as an Agency Problem.” Wisconsin Law Review 2004

(2004): 679-722.

Gleick, James. “Click OK to Agree.” New York Times Magazine. May 10, 1998.

Goldberg, Victor P. “The ‘Battle of the Forms’: Fairness, Efficiency, and the Best-Shot Rule.”

Oregon Law Review 76 (1997): 155-71.

26

Haagen, Paul H. “New Wineskins for New Wine: The Need to Encourage Fairness in Mandatory

Arbitration.” Arizona Law Review 40 (1998): 1039-68.

Hillman, Robert A. “Rolling Contracts.” Fordham Law Review 71 (2002): 743-59.

Hillman, Robert A. “On-Line Consumer Standard-Form Contracting Practices: A Survey and

Discussion of Legal Implications.” In Is Consumer Protection an Anachronism in the

Information Economy? Ashgate Press, 2005.

Hillman, Robert A. “On-line Boilerplate: Would Mandatory Website Disclosure of E-standard

Terms Backfire?” forthcoming, Michigan Law Review (2006).

Katz, Avery. “Your Terms of Mine? The Duty to Read the Fine Print in Contracts.” RAND

Journal of Economics 21 (1990): 518-33.

Korobkin, Russell. “Bounded Rationality, Standard Form Contracts and Unconscionability.”

University of Chicago Law Review 70 (2003): 1203-1295.

Lemley, Mark A. “Intellectual Property and Shrinkwrap Licenses.” Southern California Law

Review 68 (1995): 1239-1294.

Lemley, Mark A. “Beyond Preemption: The Law and Policy of Intellectual Property Licensing.”

California Law Review 87 (1999): 111-72.

A Lemon Law for Software? The Economist, March 14, 2002.

Macaulay, Stewart. Listserve Comment, “Common Sense and Contracts Symposium: The

Gateway Thread - AALS Contracts Listserv.” Touro Law Review 16 (2000): 1147-1204.

Macaulay, Stewart. “Freedom of Contract: Solutions in Search for a Problem?” Wisconsin Law

Review 2004 (2004): 777-820.

Mann, Ronald J. and Winn, Jane K. Electronic Commerce. Gaithersburg, MD: Aspen Publishers,

2002.

Marotta-Wurgler, Florencia. “Competition and the Quality of Standard Form Contracts: An

Empirical Analysis of Software License Agreements.” NYU Law and Economics

Working Paper No. 05-11 (2005).

Newitz, Annalee. “Dangerous Terms: A User’s Guide to EULAs.” www.eff.org/wp/eula.php.

Overly, Michael and Kalyvas, James P. Software Agreements Line by Line. Boston: Aspatore,

Inc., 2004.

Posner, Richard. Economic Analysis of Law, 3rd ed. Boston: Little, Brown and Company, 1986.

27

Post, Deborah. “Dismantling Democracy: Common Sense and the Contract Jurisprudence of

Frank Easterbrook.” Touro Law Review 16 (2000): 1205-39.

Priest, George L. “A Theory of the Consumer Product Warranty.” Yale Law Journal 90 (1981):

1297-1352.

Rakoff, Todd D. “Contracts of Adhesion: An Essay of Reconstruction.” Harvard Law Review 96

(1983): 1173-1284.

Robertson, Melissa. “Is Assent Still a Prerequisite for Contract Formation in Today’s E-

Conomy?” Washington Law Review 78 (2003): 265-78.

Russo, J. Edward. “The Value of Unit Price Information.” Journal of Marketing Research 14

(1977): 193-201.

Schwartz, Alan and Wilde, Louis. “Intervening in Markets on the Basis of Imperfect

Information: A Legal and Economic Analysis.” University of Pennsylvania Law Review

127 (1979): 630-82.

Alan Schwartz & Louis Wilde, Imperfect Information in Markets: The Examples of Warranties

and Security Interests, Virginia Law Review 69 (1983):1387-467.

Slawson, W. David. “Standard Form Contracts and Democratic Control of Lawmaking Power.”

Harvard Law Review 84 (1971): 529-66.

Spence, A. Michael. “Monopoly, Quality, and Regulation.” The Bell Journal of Economics 6

(1975): 417-29.

Sternlight, Jean R. “As Mandatory Binding Arbitration Meets the Class Action, Will the Class

Action Survive?” William and Mary Law Review 42 (2000): 1-126.

U.S. Census Bureau. “Quarterly Retail E-Commerce Sales, 4th Quarter of 2004.”

http://www.census.gov/mrts/www/data/pdf/04Q4.pdf.

White, James J. “Autistic Contracts.” Wayne Law Review 65 (2000): 1693-1731.

White, James J. “Default Rules in Sales and the Myth of Contracting Out.” Loyola Law Review

48 (2002): 53-85.

28

TABLE 1

SOFTWARE COMPANY AND PRODUCT SUMMARY STATISTICS

NOTE.—Means, standard deviations, and other summary statistics are based on 515 end-user license agreements (EULAs) from 468 software companies. Total revenue and years since incorporation were obtained primarily from Hoover’s company directory and the 2005 Software Industry Directory database. Age refers to the number of years since incorporation, measured as of the year 2005. The maximum age is set to 25, as operations before 1980 were unlikely to emphasize software publishing. Price is the price of the product for which the EULA is obtained and is from the company’s website. Consumer Product is a dummy variable for whether the target user of the product is the general public (or small business users) rather than large business users. Trial Exists is a dummy variable for whether the seller offers a free trial version of the product.

Mean (s.d.)

Minimum

Median

Maximum

Panel A. Company Characteristics Revenue ($)

371,000,000

(4,270,000,000)

1,000

2,000,000

87,500,000,000

Age (Years)

15

(6.36)

2

15

25

Public Company

.15

(.36)

0

0

1

Panel B. Product Characteristics Consumer Product

.41

(.49)

0

0

1

Price ($)

899

(2,710)

9.95

299

40,000

Trial Exists

.70

(.46)

0

1

1

TABLE 2

PRE-PURCHASE ACCESSIBILITY OF EULAS

NOTE.—Where companies locate EULAs on their websites. The analysis is based on 515 EULAs. For each EULA, I count the distance in number of clicks the contract is away from the most natural clickstream to purchase the product, starting from the company’s homepage. A distance of 0 indicates that the EULA is displayed directly to the buyer. A distance of 0.5 indicates that the buyer is required to acknowledge the EULA, but the contract itself is not displayed without another click. Distances of 0 and 0.5 are described as “Forced” because the buyer is forced to either see the EULA or acknowledge its existence. Distances between 0 and 6+ are described as “Findable” as the buyer is able to find the EULA on the website. Pay Now, Terms Later EULAs are not findable by definition and were obtained through email and telephone request.

Accessibility Description

Distance in Clicks From Buyer’s Clickstream

Number of EULAs (%)

0

24 (4.7)

“Forced”

0.5

5 (1)

1

79 (15.3)

2

62 (12)

“Findable”

3

44 (8.5)

4

12 (2.3)

5

5 (1)

6+

15 (3)

“Pay Now, Terms Later” (PNTL)

N.A.

269 (52.2)

30

31

TABLE 3

REGRESSIONS: DETERMINANTS OF EULA ACCESSIBILITY

NOTE.—Regression results. In the first two models, the dependent variable is Findable, a dummy variable indicating that the contract is available on the company’s website. In the middle two models, the dependent variable is Forced, a dummy variable indicating that consumers are required to agree with the EULA before purchasing. In the last two models, the dependent variable is Distance, which measures, for contracts that are Findable, the number of clicks the EULA is away from the purchasing consumer’s natural click path. The independent variables are a dummy indicating a consumer-oriented product, the natural log of the price of the product, whether there is a trial version of the software available on the website, dummies for multi-user and developer licenses (the default category is single-user license), the size of the company as proxied by the natural log of revenue, and the natural log of the age of the company since incorporating as of 2005. Even-numbered models include market fixed effects based on Amazon.com software classifications. Standard errors are in parentheses. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.

Dep. Var.: Findable Dep. Var.: Forced Dep. Var.: Distance (1) (2) (3) (4) (5) (6) Consumer

.01 (.06)

.02 (.06)

-.04 (.03)

-.03 (.03)

.26 (.26)

-.02 (.27)

Ln Price

-.00 (.02)

.01 (.02)

.00 (.01)

.00 (.01)

-.00 (.08)

-.01 (.10)

Trial

.01 (.05)

-.05 (.06)

-.01 (.02)

-.00 (.03)

.02 (.22)

.13 (.24)

Multi-User

.08 (.08)

.04 (.09)

-.03 (.04)

-.08 (.04)

-.10 (.33)

-.46 (.34)

Developer

.19** (.08)

.19* (.10)

-.05 (.04)

-.01 (.05)

.11 (.31)

.18 (.38)

Ln Revenue

.05*** (.01)

.04*** (.01)

.01** (.01)

.01 (.01)

.08 (.05)

.13*** (.05)

Ln Age

-.11** (.04)

-.12** (.05)

.01 (.02)

.01 (.02)

-.21 (.19)

-.39* (.21)

Public

.00 (.08)

.04 (.08)

-.07* (.04)

-.05 (.04)

.35 (.31)

.03 (.35)

Fixed Effects . Market . Market . Market N

515 515 515 515 246 246

R2

.07 .26 .02 .24 .05 .51

TABLE 4

EULA TERMS AND BIAS: METHODOLOGY NOTE.—The table describes the terms recorded for the EULAs in the sample and how each term is scored for purposes of measuring the overall buyer (licensee) vs. seller (licensor) bias of the contract. Negative scores capture pro-seller terms and positive scores capture pro-buyer terms. Zero scores capture neutral terms or (in case the term is not discussed in the particular contract) terms that would correspond to the default rule.

Term

Bias Measurement

Mean (s.d.)

Acceptance of License

Does license alert consumer that product can be returned if she declines terms?

1=yes 0=no

.50 (.50)

Scope of License

Does definition of "licensed software" include regular updates such as enhancements, versions, releases, etc.?

1=yes 0=no mention -1=no

.09 (.49)

Are there restrictions on use?

0=no or no mention -1=yes (e.g., for business-oriented products, "for business purposes" or "internal purposes only" language; for consumer-oriented products, restrictions on commercial use)

-.19 (.39)

Can licensee alter/modify the program?

0=yes or no mention -1=no

-.63 (.48)

Can licensee create derivative works?

0=largely unrestricted or no mention -1=strict prohibition, derivative works owned by licensor, or need permission of licensor

-.36 (.48)

Transfer of License

Are there limitations on transfer?

0=no or no mention -1=some or full restrictions (licensee cannot assign, transfer, lease, sublicense, distribute, etc.; or, needs written consent of licensor)

-.95 (.22)

Can licensee transfer the software to an end user who accepts the license terms without licensor’s prior permission?

0=yes or no mention -1=no

-.48 (.50)

Warranties and Disclaimers of Warranties

Are there express warranties? 1=yes 0=no

.05 (.22)

Is there a limited warranty stating that software is free from defects in materials and workmanship or that the software will work according manual specifications in force for 31 days or more?

1=yes (recorded as the number of days of coverage) 0=no

.31 (.46)

Is there a limited warranty stating that the media of software distribution and documentation are free from defects in force for 31 days or more?

1=yes (recorded as the number of days of coverage) 0=no

.30 (.46)

Do the limited warranties apply only to the original purchaser?

0=no or no mention -1=yes

-.04 (.19)

Is the disclaimer in caps, bold, or otherwise conspicuously presented?

0=yes or no disclaimers appear -1=no

-.23 (.42)

Disclaims IWM and IWFPP or contains "AS IS" language?

0=no -1=yes

-.91 (.29)

Disclaims warranty that software will not infringe on third parties’ intellectual property rights?

0=no -1=yes

-.37 (.48)

Limitations on Liability

Who bears the risk of loss?

0=licensor, for losses caused by factors under licensor’s control, or no mention -1=licensee

-.15 (.36)

Who bears the performance risk? 0=licensor (for causes under licensor's control), or no mention, or licensee (for uses expressly forbidden by licensor) -1=licensee (language “licensee assumes responsibility of choice of product and functions,” etc)

-.27 (.44)

Disclaims consequential, incidental, and special damages? 0=no or no mention -1=yes

-.90 (.30)

Are damages waived under all theories of liability (contract, tort, strict liability)?

0=no or no mention -1=yes

-.32 (.47)

Disclaims liability for foreseeable losses of licensee? 0=no or no mention -1=yes

-.72 (.45)

What is the limitation on damages?

0=no mention or cap on damages greater than purchase price -1=cap on damages less than or equal to purchase price

-.58 (.49)

Is there an indemnification clause?

1=indemnification by licensor 0=no, no mention, or two-way indemnification -1=indemnification by licensee

-.12 (.43)

Maintenance and Support

Does base price include M&S for 31 days or more?

1=yes 0=no or no mention

.66 (.47)

Conflict Resolution

Forum specified? 0=court, choice of licensee, or no mention -1=specific court or mandatory arbitration

-.31 (.46)

Law specified?

0=same as forum or no mention -1=yes and different from forum

-.00 (.04)

Who pays licensor’s attorney fees?

0=paid by losing party or no mention -1=paid by licensee

-.01 (.10)

Overall Bias Index

(sum of all terms above)

-5.63 (3.04)

33

TABLE 5

EULA ACCESSIBILITY AND BIAS NOTE.—Overall index bias by pre-purchase accessibility. EULAs are classified as Findable, meaning that the contract is available somewhere on the company’s website, or Pay Now, Terms Later (PNTL). Findable contracts are either Forced, meaning that consumers are required to agree before purchasing, or not Forced, meaning that the contract is one or more clicks away from the purchasing consumer’s natural click path. The last column reports the difference between the mean overall bias of EULAs of a given accessibility and PNTL EULAs. ***, **, and * indicate that the means are significantly different at the 1%, 5%, and 10% level, respectively.

EULA Accessibility

Number of EULAs (%)

Mean Overall Bias (s.d.)

Difference vs. Mean Bias of PNTL

Findable

246

(47.7)

-5.96 (3.08)

-0.62**

Forced 29 (5.7)

-6.38 (3.08)

-1.05*

0 clicks 24 (4.7)

-6.08 (3.13)

-0.75

0.5 clicks 5 (1)

-7.8 (2.59)

-2.47*

Not Forced

217 (42.1)

-5.90 (3.08)

-0.57**

1 click 79 (15.3)

-5.59 (3.18)

-0.26

2 clicks 62 (12)

-6.39 (3.23)

-1.06**

3 clicks 44 (8.5)

-5.64 (3.04)

-0.31

4 clicks 12 (2.3)

-5.58 (1.78)

-0.25

5 clicks 5 (1)

-5.56 (3.43)

-0.27

6+ clicks 15 (3)

-6.6 (2.85)

-1.27

PNTL 269 (52.2)

-5.33 (2.90)

.

35

TABLE 6

REGRESSIONS: EULA BIAS, ACCESSIBILITY, AND COMPANY AND PRODUCT CHARACTERISTICS

NOTE.—The dependent variable is the overall bias index of the EULA. Higher values indicate pro-buyer bias; lower values indicate pro-seller bias. The independent variables include measures of pre-purchase accessibility (Findable, Forced, or Distance), product characteristics (a dummy for consumer-oriented products, the log of the product’s price, a dummy for whether a trial version of the product is offered, a dummy for EULAs allowing for multiple users, and a dummy for developer licenses), company characteristics (natural log of revenue, natural log of years since incorporation, and a dummy for publicly traded), and software market dummies based on Amazon.com classifications. The first seven specifications include all licenses with available data, and the last two specifications are restricted to licenses of business-oriented software and consumer-oriented software. Standard errors are in parentheses. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.

Dependent variable: Overall Bias Index All All All All All All All Business Consumer (1) (2) (3) (4) (5) (6) (7) (8) (9) Findable -.62**

(.27) -.57**

(.28) -.26

(.27) -.23 (.29)

.18 (.39)

-1.37** (.54)

Forced -.79 (.58)

-.48 (.60)

-.74 (.57)

-1.34**

(.62) -1.50**

(.75) -1.03

(1.31) Distance -.05

(.09)

Consumer .27

(.33) .25

(.33) .08

(.36)

Ln Price .11 (.11)

.11 (.11)

.06 (.13)

.05 (.17)

.15 (.28)

Trial .51* (.28)

.51* (.28)

.47 (.31)

.66 (.43)

.39 (.55)

Multi-User -.56 (.48)

-.56 (.45)

-.40 (.51)

-.00 (.56)

-1.39 (2.11)

Developer .63 (.46)

.64 (.46)

.93 (.58)

.30 (.66)

2.69 (2.02)

Ln Revenue -.14**

(.07) -.12* (.07)

-.11 (.08)

-.17 (.11)

-.03 (.13)

Ln Age 1.95*** (.25)

1.94*** (.26)

1.54***

(.28) 1.88***

(.39) .88*

(.51) Public -.83*

(.44) -.88** (.44)

-1.04**

(.48) -1.43**

(.70) -.03

(.77) Fixed Effects . . . . . Market Market Market

N 515 515 246 515 515 515 515 304 211

R2 .01 .00 .00 .01 .13 .14 .37 .44 .52

TABLE 7

REGRESSIONS: EULA SUBINDEX BIAS, ACCESSIBILITY, AND COMPANY AND PRODUCT CHARACTERISTICS NOTE.—The dependent variables are the seven subindex bias scores: Acceptance of License, Scope of License, Transfer of License, Warranties and Disclaimers of Warranties, Limitations on Liability, Maintenance and Support, and Conflict Resolution. Higher values indicate pro-buyer bias; lower values indicate pro-seller bias. The independent variables include measures of pre-purchase accessibility (Findable or Forced). Even-numbered models also include controls for product characteristics (a dummy for consumer-oriented products, the log of the product’s price, a dummy for whether a trial version of the product is offered, a dummy for EULAs allowing for multiple users, and a dummy for developer licenses), company characteristics (natural log of revenue, natural log of years since incorporation, and a dummy for publicly traded), and software market dummies based on Amazon.com classifications. Standard errors are in parentheses. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.

Dep. Var.: Acceptance of

License

Dep. Var.: Scope of License

Dep. Var.: Transfer of

License

Dep. Var.: Warranties and Disclaimers of

Dep. Var.: Limitations on

Liability

Dep. Var.: Maintenance and Support

Dep. Var.: Conflict

Resolution (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) Findable

-.00 -.02 (.05) (.05)

.05 (.10)

.10 (.11)

.01 (.06)

.05 (.06)

-.06 (.10)

-.11 (.10)

-.13 (.13)

-.15 (.14)

.00 (.04)

.01 (.05)

-.13*** (.04)

-.13*** (.05)

Forced -.10 -.06 (.10) (.11)

-.25 (.21)

-.34 (.23)

-.25** (.12)

-.35*** (.13)

.06 (.20)

.02 (.22)

-.22 (.28)

-.56** (.31)

.15 (.09)

.16 (.10)

-.14 (.09)

-.21** (.10)

Co. and Product Controls

No Yes No Yes No Yes No Yes No Yes No Yes No Yes

Fixed Effects

. Market . Market . Market . Market . Market . Market . Market

N

515 515 515 515 515 515 515 515 515 515 515 515 515 515

R2

.06 .26 .05 .27 .05 .25 .09 .29 .09 .32 .04 .26 .13 .34

FIGURE 1. EULA Bias by Pre-Purchase Accessibility

-10 -8 -6 -4 -2 0 2

Findable

Forced

Not Forced

PNTL

0

0.5

1

2

3

4

5

6+

PNTL

Pro-Seller Default Pro-Buyer

Acc

essi

bilit

y (C

ateg

ory)

Acc

essi

bilit

y (#

of c

licks

)

Overall Bias Index