Does Anyone Read the Fine Print? - University of Chicago Law School

The University of Chicago
The University of Chicago Law School
Does Anyone Read the Fine Print? Consumer Attention to Standard-Form Contracts
Author(s): Yannis Bakos, Florencia Marotta-Wurgler, and David R. Trossen
Source: The Journal of Legal Studies, Vol. 43, No. 1 (January 2014), pp. 1-35
Published by: The University of Chicago Press for The University of Chicago Law School
Stable URL: .
Accessed: 16/06/2015 15:10
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected]
The University of Chicago Press, The University of Chicago, The University of Chicago Law School are
collaborating with JSTOR to digitize, preserve and extend access to The Journal of Legal Studies.
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
Does Anyone Read the Fine Print?
Consumer Attention to Standard-Form
Yannis Bakos, Florencia Marotta-Wurgler, and David R. Trossen
A cornerstone of the law and economics approach to standard-form contracts is the informedminority hypothesis: in competitive markets, a minority of term-conscious buyers is sufficient
to discipline sellers from using unfavorable boilerplate terms. This argument is often invoked
to limit intervention or regulate consumer transactions, but there has been little empirical
investigation of its validity. We track the Internet browsing behavior of 48,154 monthly visitors
to the Web sites of 90 online software companies to study the extent to which potential
buyers access the end-user license agreement. We find that only one or two of every 1,000
retail software shoppers access the license agreement and that most of those who do access
it read no more than a small portion. Since the cost of comparison shopping online is so low,
the limiting factor in becoming informed thus seems not to be the cost of accessing license
terms but reading and comprehending them.
1 . I N T R O D U CT I O N
Standard-form contracts, often called fine print or boilerplate, are the
most common type of economic contract. They apply to untold billions
of commercial transactions per year. In a typical scenario, a buyer puris Associate Professor of Management at the Leonard N. Stern School of
Business at New York University. FLORENCIA MAROTTA -WURGLER is Professor of Law at
New York University School of Law. DAVID R . TROSSEN is an associate at Knobbe Martens
Intellectual Property Law. This research was supported by grants from the Networks,
Electronic Commerce, and Telecommunications Institute ( and the
Filomen D’Agostino and Max E. Greenberg Research Fund. The authors are grateful for
comments received from Barry Adler, Jennifer Arlen, Oren Bar-Gill, Shmuel Becher, Omri
Ben-Shahar, Kevin Davis, Clayton Gillette, Michael Levine, Ronald Mann, Yoram Margoliath, Jacob Nussim, Jeffrey Wurgler, and participants in the 2008 conference on Legal
Institutions and Entrepreneurship, the 2009 Workshop of Information Systems Economics,
[Journal of Legal Studies, vol. 43 (January 2014)]
䉷 2014 by The University of Chicago. All rights reserved. 0047-2530/2014/4301-0001$10.00
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
chases a good or service and is presented with a preprinted form contract
with terms pertaining to dispute resolution, remedies for product failure,
and warranties, among others, with little opportunity to negotiate the
terms. Examples appear everywhere and can include safety disclaimers
noted on the backs of sporting tickets, warranties packaged with consumer goods, privacy policies and terms of use on Web sites, and photocopying restrictions appearing in the front matter of this journal. Every
reader of this paper has entered into thousands of standard-form contracts, sometimes unknowingly.
Academics, courts, and policy makers have long debated the degree
to which standard-form contracts should be enforced and whether their
content or disclosure should be regulated. All sides in this debate realize
that, in many circumstances, a majority of buyers do not read fine print.
For many buyers, too much time is required to read and give meaningful
assent, and fine print can be too difficult to understand or may seem
unimportant. The central economic question is whether the fact that a
majority of buyers enter standard-form contracts with this imperfect
information results in a market failure: if buyers do not factor contract
terms into their purchase decisions, sellers lack incentives to provide
anything more than the minimally required legal protections.1
Defenders of freedom of contract have generally rejected intervention
by relying on reputational constraints and the informed-minority argument. In this paper, we focus on this latter argument, which derives
from some classic law and economics contributions. The articulation by
Schwartz and Wilde (1979) of the informed-minority argument in this
context is a specific application of work on imperfect information by
the 2009 Harvard-Texas Conference on Commercial Law Realities, the 2009 New York
University colloquium on Law and Economics, the 2009 University of Virginia School of
Law Legal Studies Workshop, the 2009 George Mason University–Microsoft conference
Innovation: Online Markets versus Traditional Markets, the University of Kansas Law
School, the 2009 Hebrew University of Jerusalem conference Contract Law: Interdisciplinary Perspectives, the 2009 Georgetown University Law Center Workshop in Law and
Economics, the 2009 colloquium series at the University of Alabama School of Law, the
4th annual Conference on Empirical Legal Studies, the 2010 Fordham University Law and
Information Society Faculty Workshop series, the 2010 Swiss Federal Institute of Technology (ETH-Zurich) Workshop and Lecture Series in Law and Economics, and the 20th
annual meeting of the American Law and Economics Association. Dennis E. Hermreck,
Daniel Priest, Robert Taylor, Michael Tonkinson, and Joel Willcher provided excellent
research assistance.
1. For a comprehensive review of the factors that might contribute to consumer information problems and subsequent market failures, see Salop (1976) and Beales, Craswell,
and Salop (1981).
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
S TA N D A R D - F O R M C O N T R A C T S /
Spence (1977) in the context of product liability and by Salop and Stiglitz
(1977) in the context of price dispersion and search. Schwartz and Wilde
(1979) argue that sellers will not necessarily offer one-sided terms even
when the majority of buyers do not read standard-form contracts. In
their model, nonreading buyers benefit from an informed minority whose
willingness to pay for the product is sufficiently sensitive to the quality
of the standard terms. When all buyers have the same taste for quality
and sellers are unable to discriminate between reading and nonreading
buyers, sellers will offer the terms preferred by all buyers. This
competitive-markets logic has often been used to resist regulation and
to advocate rules limited to facilitating search by those aspiring to join
the informed minority (see, for instance, Baird 2006; Gillette 2005; Beales, Craswell, and Salop 1981).
Although the informed-minority argument has been influential in the
law and economics literature, it has not been studied much empirically.
While it is widely agreed that standard-form contract readers are in the
minority, the literature offers no evidence of whether this minority remains large enough to plausibly enforce efficient terms, as assumed by
some theorists and courts. In fact, we are not aware of any systematic
studies of the extent to which consumers become informed about contract terms. We present large-sample evidence on the extent to which
buyers actually do read standard-form contracts in a specific setting, and
we identify factors that affect the probability of readership.
In particular, we examine the extent to which potential buyers of
software read end-user license agreements (EULAs). For a sample of
software companies who offer products online, we use potential buyers’
“clickstream” information (that is, the series of uniform resource locator
[URL] information for the page visited while browsing the Web and the
precise timing of such visits) to study their readership of the EULA. We
tracked 48,154 visitors to the Web pages of 90 software companies over
a period of 1 month and recorded their detailed browsing behavior. For
each such user, we observe the exact sequence of Web page addresses
(URLs) accessed in a particular visit and the time spent on each page.
The data also include the demographic characteristics of each user, such
as age, gender, income, and geographical location. Our main finding is
that regardless of how strictly we define a shopper, only one or two in
1,000 shoppers access a product’s EULA for at least 1 second, which
yields an informed minority of .2 percent that is orders of magnitude
smaller than the required informed-minority size in realistic market settings and in theoretical examples suggested in the literature.
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
To put our findings in economic perspective, we perform a simple
calibration exercise. We estimate the fraction of informed shoppers
needed to induce sellers to offer good terms in the software market. We
estimate the marginal cost of providing one pro-buyer term, maintenance
and support, and find that sellers would find it more cost-effective to
lose all informed buyers (that is, under the conservative assumption that
each would decline to buy if the given term is not offered) than to offer
this one term. This conclusion would likely persist for a fraction of
informed buyers that is one or two orders of magnitude higher than .2
We then focus on the factors affecting the probability that a EULA
will be accessed. We find that shoppers are more likely to access EULAs
of smaller companies or companies that offer potentially suspicious
products, such as freeware. The few shoppers who choose to become
informed might be rationally deciding to ignore the EULAs of larger,
more established companies, relying instead on company reputation or
familiarity. We also find that older and higher-income shoppers are more
likely to access EULAs. This may be because these consumers have lower
search and reading costs (for example, because they have a lower opportunity cost for their time, or because they are more educated and
thus find it easier to read contract terms). Thus, a higher fraction of
consumers read EULAs when expected benefits are likely to be higher
or reading costs are likely to be lower, which suggests that the fraction
of informed consumers is limited by the high search and reading costs
of standard-form contracts.
Our main contribution is the first large-sample study of the extent
to which consumers actually inform themselves of important rights and
obligations before entering a standard-form contract. We find that the
fraction of consumers who read such contracts is so small that it is
unlikely that an informed minority alone is shaping software license
terms. We note, however, that the absence of an informed minority does
not immediately imply that EULA terms will be inefficiently biased in
favor of sellers. Some sellers, at least prominent ones, could be disciplined by other mechanisms, for example, the aforementioned concern
for their reputation if onerous terms might eventually be discovered.
Furthermore, shoppers may assume that no matter what the EULA terms
state, they will be protected by the courts and thus will rationally choose
not to become informed about the EULA terms. In other words, it may
be rational not to read; what our evidence suggests is that EULA terms
in our setting are not being policed by an informed minority of readers.
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
S TA N D A R D - F O R M C O N T R A C T S /
To summarize, while other factors may discipline sellers from offering
adverse EULA terms, we show that the informed-minority hypothesis,
the most widely applied argument for the efficiency of standard-form
contract terms, does not seem compelling here. Our findings call into
question the promise of policies to ameliorate potential market failures
by requiring increased or mandatory disclosure, particularly in a setting
where the informed minority is most likely to arise because of the low
cost of online comparison shopping.2 Future empirical work should focus
on readership in other contexts to establish whether an informed minority might exist, as well as the potential validity of alternative mechanisms to discipline sellers in our context.
Section 2 offers theoretical and empirical background on the
informed-minority hypothesis. Section 3 explains our methodology, Section 4 presents our data, and Section 5 discusses the results. Section 6
2 . F O R M C O N T R A CT S A N D C O M P E T I T I O N F O R I N F O R M ED B U Y E R S:
Despite the transaction-cost-reducing benefits associated with the use of
form contracts, such as reduced drafting and negotiation costs, academics and policy makers have debated their fairness and the desirability of
their enforcement. Concern for consumer welfare has resulted in numerous articles, laws, and initiatives to regulate these contracts. For
example, in addition to existing contract law doctrines to protect buyers
from abusive terms, such as unconscionability and unfair surprise, several state consumer laws prohibit the use of forum selection clauses and
disclaimers of implied warranties in consumer contracts (see, for example, Gatton v. T-Mobile USA, Inc., 152 Cal. App. 4th Supp. 571,
585 [1st Dist. 2007]; Fidelity & Deposit Co. v. Gainesville Iron Works,
Inc., 125 Ga. App. 829 [1972]; Idaho Code sec. 29-110; N.C. Gen. Stat.
sec. 22B-3; Mont. Code 36 sec. 18-1-403). Federal laws such as the
Truth in Lending Act (15 U.S.C. 41) and the Magnuson-Moss Warranty
Act (15 U.S.C. 2301) seek to decrease reading and search costs by requiring standardized disclosure of mandated terms. More recently, as
can be seen in Lemley (2006), there has been heated debate about
whether online contracts, such as terms of use, privacy policies, and
2. For an analysis of whether increased contract disclosure is associated with increased
readership, see Marotta-Wurgler (2012). See also Ben-Shahar and Schneider (2011).
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
software license agreements, should be enforceable or subject to mandatory disclosure rules or provisions.
2.1. The Informed-Minority Hypothesis
The concern that standard-form contracts are likely biased toward drafters is based on the view that many buyers do not read or understand
the terms. Salop and Stiglitz (1977) explore the conditions under which
a market with consumers who are heterogeneous in their willingness and
ability to become informed about product prices might reach a perfectly
competitive price equilibrium and find that even in the presence of many
uninformed consumers, a market can yield a competitive equilibrium if
enough informed consumers shop for the lowest price.
In a widely cited contribution to the theory of standard-form contracts, Schwartz and Wilde (1979) extend this argument to settings in
which consumers vary in their ability to become informed about contract
terms. They show that if a sufficient number of buyers are informed
about the price and contract terms of a given product, sellers who cannot
discriminate between buyer types will offer the product with efficient
terms at a competitive price to all buyers, as the cost to the seller of
losing a critical mass of informed consumers outweighs the benefits of
offering inferior terms to the uninformed inframarginal consumers. Imperfect information alone is thus not sufficient to warrant market intervention. This conclusion has become the cornerstone of the law and
economics view of standard-form contracts. For example, Priest (1981),
Baird (2006), and Hillman (2006a), among others, have repeatedly relied
on the informed-minority argument to support freedom of contract.
Instead of intervention, consumers should be given a meaningful opportunity to become informed about the terms prior to purchase.
Others have expressed doubts about this mechanism. Katz (1990)
posits that not reading may be rational, given the low probability of
adverse events triggering unfavorable clauses. Bar-Gill (2004) and Gabaix and Laibson (2006) question the underlying assumptions of the
informed readers and offer behavioral accounts of a failure to read and
understand terms. Eisenberg (1995) and Goldberg (1997) argue that
sellers might find it more profitable to take advantage of nonreaders
than to cater to readers. Ben-Shahar (2009) argues that because nobody
reads fine print, regardless of its accessibility, rules requiring increased
disclosure are useless, if not dangerous.
The informed-minority argument is reflected in contract doctrines and
current proposals to increase consumer protection in mass-market trans-
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
S TA N D A R D - F O R M C O N T R A C T S /
actions online that stress that consumers must be given a meaningful
opportunity to read.3 For instance, in Principles of the Law of Software
Contracts, the American Law Institute (2010, p. 117) seeks to “promote
reading and the opportunity to read terms” to alleviate market failures.
Its goal is to reduce the cost of accessing the contract to promote a
sufficiently numerous informed minority of buyers.
While this paper limits its scope to assessing whether readership levels
are consistent with the informed-minority hypothesis, we note that other
mechanisms may incentivize sellers to offer terms preferred by buyers
even if none read. Sellers might be constrained by reputation or the
threat of litigation, or they might offer one-sided terms to all consumers
but might relax them to accommodate reasonable complaints (see Gillette 2004; Bebchuk and Posner 2006). In the case of experience goods
or repeat purchases, buyers who do not read terms might ultimately
become familiar with the contents of boilerplate. Our data cannot speak
to the relevance of these mechanisms.
2.2. Prior Evidence
Despite the theoretical prominence of the informed minority, there has
been little empirical investigation of its validity, presumably because
observing readership is difficult. However, there is some related survey
evidence. Hillman (2006b) surveys 92 contracts students and finds that
only 4 percent of those who purchased products online claim to read
standard-form contracts as a general matter. Becher and Unger-Aviram
(2009) survey 147 students and find that 60 percent of respondents claim
that they skim or read parts of a standard-form contract before entering
a transaction. Plaut and Bartlett (2012) survey 182 undergraduates and
find that about 80 percent claim not to read contracts and that much
of the remainder claim to skim them. These surveys are suggestive but
somewhat limited; they are based on self-reported behavior or hypothetical commercial scenarios, and the survey subjects are not representative; for example, in some cases they include law students who will
write boilerplate for a living.
Other studies show that standard-form contract terms are less onesided in favor of sellers than might be possible if buyers were completely
3. A notable case reflecting this view is Specht v. Netscape Communications Corp. (306
F.3d 17 [2d Cir. 2002]), in which the courts refused to enforce an arbitration agreement
made available via a browse wrap (using hyperlink somewhere on the seller’s Web site) a
few screens later because it did not provide sufficient notice. Ben-Shahar (2009) offers a
comprehensive evaluation of this doctrine.
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
uninformed and thus are consistent with the existence of the informed
minority as well as other disciplining mechanisms. Priest (1981) studies
62 product warranties and finds that they are not biased toward sellers
but rather reflect the relative ability of buyers and sellers to prevent and
insure against loss. Marotta-Wurgler (2007, 2008) analyzes the terms
of 647 online EULAs and shows that while almost all of them are more
restrictive than the relevant default rules, they do not all converge to
the legal minimum. In a study of contracting practices by online retailers,
Mann and Siebeneicher (2008) find that few sellers offer excessively onesided contracts.
3 . R ES E A R CH F R A M E W O R K
We explore the presence of an informed minority of buyers by studying
the browsing and shopping behavior of online consumers. In particular,
we tracked the behavior of visitors to the Web sites of 90 software
companies, and we examine the rate at which shoppers choose to become
informed about the EULAs that govern the featured software.
Online software purchases provide an apt setting in which to look
for the informed minority. First, while nonprice features, such as associated contractual rights and restrictions, are important for all types of
products, they are a particularly significant consideration for information goods such as software, because terms form an integral part of the
way the product is or may be used. Second, some of the terms in EULAs
have been the subject of litigation in the past decade. (See, for instance,
Mortenson Co. v. Timberline Software Corp., 998 P.2d 305 [Wash.
2000]; Davidson & Assoc. v. Internet Gateway, 344 F. Supp. 2d 1164
[D. Mo. 2004]). For instance, as end users increasingly rely on software
to perform a variety of routine tasks and critical functions, damages
from software failure can be significant. Third, shopping for competing
goods and the terms that govern them is cheap and easy online relative
to most commercial settings. Search costs are also low (Bakos 2001).
To the extent that the informed minority exists, this is among the settings
where we are relatively likely to find it, especially given our access to
clickstream data. Finally, several recent debates on legal reform in
standard-form contracts focus on electronic contracts in general and
software contracts in particular. Our study of the informed minority in
online software markets places us at the center of these debates.
To empirically investigate the presence and size of the informed minority, we classify visitors to the Web sites of the companies in our
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
S TA N D A R D - F O R M C O N T R A C T S /
Figure 1. Empirical framework
sample, described below, as potential buyers and those visiting for other
reasons, such as looking for user forums or troubleshooting information.
We denote by s the fraction of potential buyers (shoppers); nonshoppers
make up the remaining fraction 1 ⫺ s. We denote by e1 the fraction of
shoppers and by e2 the fraction of nonshoppers who read the online
EULAs. Finally, we denote by b1 the fraction who purchase the product
(buyers) among shoppers who read the EULA and by b2 the fraction of
buyers among shoppers who do not read the EULA. This framework is
depicted in Figure 1. In this setting, the informed minority corresponds
to the fraction e1 of shoppers who read the online EULA.4
4. It is possible that, for some shoppers, accessing the end-user license agreements (EULA)
will not affect their probability of buying the product. For instance, some shoppers either
may not know what a EULA is or may discover after accessing the EULA that they are
not capable of comprehending its language, or shoppers may access the EULA accidentally
or out of curiosity. To the extent that such accesses of the EULA do not make a shopper
part of the informed minority, e1 will overestimate the fraction of shoppers who are part
of the informed minority. We can explore the significance of this to some degree by studying
the time spent on the EULA page by those who access it. On the other hand, because we
do not consider other ways in which shoppers might become informed about the terms
(for example, by word of mouth or repeat purchases), there is a possibility that e1 will
underestimate the size of the informed minority. We comment on the likely significance of
this effect in Section 5.
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
10 / T H E J O U R N A L O F L E G A L S T U D I E S / V O L U M E 4 3 ( 1 ) / J A N U A R Y 2 0 1 4
Next we estimate the number of visitors in our sample for each of
the six categories in Figure 1. We estimate the number of readers and
nonreaders among visitors classified as buyers, shoppers, and nonshoppers. We use access to a EULA page for more than 1 second to identify
readers. This creates an upward bias of our estimate of the truly informed
readers in that some accesses are accidental, inconsequential to the buying decision, accessed so briefly that little content could have been
grasped or read but not understood. We use initiation of a secure checkout process to identify buyers and other contextual information to distinguish shoppers from nonshoppers.
We can break down readers into se1b1 readers who buy and se1(1 ⫺
b1) readers who do not buy. In addition, s(1 ⫺ e1)b2 buyers are not
readers, and s(1 ⫺ e1)(1 ⫺ b2) shoppers neither read nor buy. A priori,
we expect that few nonshoppers read EULAs, and thus we expect (1 ⫺
s)e2 to be small. Finally, the fraction of nonshoppers who do not read
EULAs is (1 ⫺ s)(1 ⫺ e2), which, as expected and as we confirm, is large.
With these inputs, we can estimate the fraction e1 of shoppers who
constitute the informed minority, by writing this fraction as
e1 p
se1 b1 ⫹ se1(1 ⫺ b1)
se1 b1 ⫹ se1(1 ⫺ b1) ⫹ s(1 ⫺ e1)b2 ⫹ s(1 ⫺ e1)(1 ⫺ b2 )
We then analyze the seller’s choice of which terms to offer, to assess
whether our estimates are plausibly consistent with an informed-minority equilibrium.
4 . D ATA
Our large clickstream data set represents the browsing behavior of
92,411 U.S. households in January 2007. This data set was made available to us by a major online research company, which has recruited a
representative panel of U.S. households that have agreed to install on
their computers a data collection plug-in that records the URL address
of each Web page visited. The data collected include the exact sequence
of Web pages visited and the amount of time spent on each page. In
raw form, this is a data set of significant size.5
The panel of households was selected to be demographically and
geographically balanced and representative of the population of U.S.
5. Information was captured for 6,355,922 user sessions and 461,027,284 corresponding
page views.
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
S TA N D A R D - F O R M C O N T R A C T S /
households with Internet access.6 The information captured in the raw
data for each Web page visited by a panelist is coded with both a user
identifier that anonymously but uniquely identifies each panelist and a
session identifier that delimits each panelist’s Web browsing into separate
sessions. Additional information captured includes the URL of each Web
page visited, the time that each Web page was accessed, the length of
time spent on that page, whether that page was within a secure (that is,
encrypted) connection, the Web server delivering the Web page, and a
unique identifier for the company or division owning that Web server.
The recorded page views compose the bulk of the data, but we were
provided with useful additional files that include nonpersonally identifiable demographic information about the panelists and a corporate hierarchy identifying the parents, if any, of the divisions or companies that
own the Web servers appearing in the data (for example, Office and
Outlook are properly identified as companies or divisions having the
same corporate parent: Microsoft).
4.1. Sample Construction
We consider the central standard-form contract in one important market.
We study user visits to the Web pages of software companies that sell
or distribute their products through their corporate Web sites and that
make their EULAs available on their site for users to peruse at their
option (prior to making any purchase decision). We use the data provider’s classification of markets to identify visits to software companies
only. We subsequently identify in our data two types of software companies that make their products available for online purchase or downloading: retailers and freeware providers. Retailers license their software
for a price through their corporate Web site. Freeware providers offer
their software for free to anyone wishing to download it; examples
include browser toolbars and plug-ins.7 We are interested in observing
6. This data provider’s panel is one of the largest representative media research samples
in existence. During the period the data were collected, the sample of participants was
defined using random-digit-dialing principles: the company selected a random set of phone
numbers from all available residential numbers in the United States and attempted to recruit
each for at most 15 times at different times of the day and on different days. The panel
also included university students and individuals in the workplace. The company updated
its demographic information regularly, had implemented various procedures to keep the
panel updated, and ensured that tracking was unobtrusive to prevent any distortions in
7. We classify a company as a retail company if it offers its core or much of its software
for sale, even if it also offers free software.
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
12 / T H E J O U R N A L O F L E G A L S T U D I E S / V O L U M E 4 3 ( 1 ) / J A N U A R Y 2 0 1 4
users’ propensities to become informed about the terms of these two
types of software.
For the purpose of having a sufficiently homogenous sample of sellers,
we exclude subcategories such as vendors not making their products
available for online purchase or downloading, peer-to-peer software providers, and Web hosting companies. We exclude companies with fewer
than 50 unique visitors who viewed at least two pages during their visit;
our interest is in users with intent or potential intent to purchase (shoppers), and users who view only a single page are less likely to have such
intent. We identified 197 companies that satisfied the above conditions.
For each of these companies, we obtained the URLs of all EULAs
available on the company’s Web site. To find the EULAs, we visited each
company’s Web site and used manual browsing, Google searches within
the Web site, and, if available, searches of the Web site provided by the
company. In addition, we searched all page views in the clickstream data
corresponding to these companies to identify possible EULA pages (for
example, pages whose Web address contained the words “EULA,” “legal,” or “terms”), which we then investigated manually.
Some EULAs were presented as browse wraps—that is, they were
posted as a hyperlink somewhere on the seller’s Web site. We included
all these companies because we can easily measure whether users voluntarily clicked on the EULA hyperlink.8 A minority of companies presented their EULAs as click wraps. This mode of presentation requires
consumers to click on an “I agree” button acknowledging the EULA
terms before they can purchase a product. There are two types of clickwrap sites. One type presents EULA terms via a hyperlink adjacent to
the “I agree” button and thus requires an additional click to access the
EULA. In this case, while all buyers are forced to acknowledge the
EULAs, we can measure what fraction takes the extra step and actually
clicks on the contract link, which is a necessary step in becoming informed about the terms. The other type presents the terms in a scrollable
8. A possible concern is that these contracts are not prominent enough to be binding.
As noted in Section 2.1, courts have been reluctant to enforce browse wraps. However,
this mode of contract presentation is not too problematic in the online software market
because most sellers also present the EULA prominently after purchase, at the time of
installation of the software. This mode of contract presentation, or “pay now, terms later”
contracting, has been held to be valid by most courts. Even though the contract is available
after purchase, consumers who aspire to become members of the informed minority would
have an incentive to check contract terms before buying because it is less costly to comparison shop this way than to purchase and return software just to see the terms of the
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
S TA N D A R D - F O R M C O N T R A C T S /
text box above the “I agree” button; we removed from our data set
companies that use this type of click wrap because we do not have a
way to measure whether buyers read the terms. While scrolling through
and reading EULA terms would likely increase the total time spent on
the corresponding checkout pages, which we can observe, there are several other actions that users typically are required to take on the same
pages. As a result, measures based on the total time spent on the checkout
page where the terms are presented were too noisy to be useful for this
type of company. Finally, we removed from our data set companies that
did not make their EULAs available online.
After excluding companies for which we do not have enough data
or that are otherwise inappropriate for our tests, we arrived at a final
sample of 78 retail and 12 freeware companies. We have no reason to
believe that our basic results or conclusions would change significantly
were we to increase the number of companies in the sample or the time
window during which panelists were followed. In addition, the size of
our sample is probably more usefully characterized in terms of the tens
of thousands of company visits that we track, described below, because
each of these represents an opportunity to access a EULA and is thus
the essential unit of observation.
4.2. Company and Product Characteristics
All else being equal, consumers may feel less need to scrutinize the terms
in EULAs from companies that are large or old because they assume
that such companies are more trustworthy and fair. To test this hypothesis, we obtain information about each company’s annual revenue,
year of incorporation, and public or private status. These data were
obtained from Yahoo! Finance, Hoover’s,9 or direct communication.
Table 1 reports summary statistics for the company characteristics
of the two types of companies analyzed. For each company, we note the
number of products that it offers (counting each distinctly named product as a separate product), which allows us to calculate the average
revenue per product.
We also collect several product characteristics, and we record one
flagship product per company. Many small- and medium-size companies
market one main product, in which case we select that product as the
flagship. For larger companies, we select the product accounting for the
9. Proprietary business information was purchased from Hoover’s, Inc. (http://www
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
Note. The sample consists of 78 retail companies and 12 freeware companies.
Revenue ($ millions)
Average revenue per product ($ millions)
Age (years)
Public company
Revenue ($ millions)
Number of products offered
Average revenue per product ($ millions)
Age (years)
Public company
Consumer product
Number of products offered
Price ($)
Median price ($)
Trial, featured product
Trial, most products
Consumer product
Table 1. Company and Product Characteristics
S TA N D A R D - F O R M C O N T R A C T S /
largest fraction of sales or, when this information is not available, the
product most prominently featured on the Web site, as these might be
the products about which most consumers care.10 Consumers might be
less inclined to read the EULAs of the most popular (or flagship) products, as they are likely to have established stronger reputations. Still, the
choice of flagship product should not affect our overall findings, as the
readership and visit metrics that we report are for all products, not just
flagship products.
It is possible that users are more inclined to become informed about
the EULA terms of more costly products, so we record the price of the
flagship product as well as the median price of all products available on
the Web site for that firm. We record whether the product is a single or
multiuse license, because multiseat licenses are likely to have higher
prices, and whether the product is offered to developers. We note
whether the company offers a trial version of the flagship product, and
also of the majority of its products, because that may also affect users’
propensity to read terms.11 We also note whether the product is oriented
toward business users or the general public. Finally, we classify each
product into one of 150 software product categories (for example, antivirus or word processing), on the basis of the characterizations of
software at Amazon.
4.3. Contract Characteristics
We want to measure the fraction of shoppers who become informed
about EULA terms. We thus collected all the EULA URLs available on
a company’s Web site. As noted above, many firms sell only one product,
and thus they make available online only the EULA that governs the
use of that product. Other firms sell many products that are all governed
by a single EULA posted on their Web site, and others post different
EULAs for different products. Finally, some firms post the EULAs for
all their current and past versions of all their products. We found 240
unique URLs corresponding to EULAs for our sample companies.
10. We used a flagship product to collect product-related statistics, as our data set did
not allow us to identify the actual product considered by most shoppers, and we were not
able to obtain detailed per-product sales data that we could use to weigh the products
offered by different sellers.
11. Trial versions are generally offered with limited functionalities over a limited period.
Marotta-Wurgler (2007) finds that the majority of trial licenses are noticeably different
(for example, the trial license reads “trial license” and is generally shorter than the product
license), such that a user would not consider them substitutes.
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
16 / T H E J O U R N A L O F L E G A L S T U D I E S / V O L U M E 4 3 ( 1 ) / J A N U A R Y 2 0 1 4
4.4. Defining Shoppers and Shopping Visits
Among panelists in our data who visit a given company’s Web site, we
need to define shoppers (namely, visitors with some potential to purchase), since a potentially large fraction of visitors may be browsing
without any intent to purchase. We define a user visit as all page views
(URL accesses) from a company’s Web site during a single user session.
The first definition of user visit identifies shoppers by examining the
intensity of a visit to a company Web site. A user with intent to purchase
is likely to view several pages on the retail side of the company’s Web
site. We follow Moe and Fader (2004) and Catledge and Pitkow (1995)
and define our broadest definition of a shopping visit as one with at
least two page views on a company’s Web site. A second, more restrictive
definition includes all visits by users who accessed at least five pages on
a given company’s Web site. Bucklin and Sismeiro (2003) find that this
is progressively more likely to exclude casual browsers.
At the other extreme, a visitor who has selected a product and initiated a checkout or payment process has demonstrated intent to purchase. Thus, we use the initiation of the checkout process as the strictest
criterion to identify visits with intent to purchase. We identify such events
by identifying for the 90 companies in our sample the Web page addresses that would be accessed only during the checkout and payment
process and by subsequently recognizing visits that access such pages.
While knowing that a user started a checkout or payment process provides no guarantee that the transaction was completed, it indicates a
high likelihood that a transaction was at least contemplated. This definition of shopping visit is likely to be overly restrictive, as it excludes
visits that do not result in the initiation of a checkout process.12
To summarize, the three measures described above establish the shopping intent of a session with increasing strictness. As our definitions of
a shopping visit become stricter, we expect that estimates of the informed
minority become more conservative, and the actual number is likely to lie
somewhere between the three estimates that our methodology provides.
4.5. Defining Shopping Visits: Single Sessions versus
Monthly Aggregates
To define a shopping visit, we adopt the two approaches that are standard in the literature using clickstream data. The first approach, used
12. Given the low conversion (of visitors to buyers) rates in e-commerce, such visits
likely represent the majority of shopping visits.
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
S TA N D A R D - F O R M C O N T R A C T S /
by our data provider and the industry in general, defines user sessions
as periods of Web browsing activity separated by at least 30 minutes of
inactivity. Under this definition, as summarized by Moe and Fader
(2004), a user can have multiple visits to a given company in a day, a
week, or a month.
The second approach recognizes that a user’s shopping activity on a
given company’s Web site can span several days or even weeks given
the low cost of access. Johnson et al. (2004) find that repeated visits to
a company’s site within a month typically correspond to the same shopping cycle. We thus aggregate visits to a unique company’s site in a given
month and present these aggregated sessions as an alternative measure
of company visits with intent to purchase.
4.6. Demographic and Geographic Data
We use personal information about our panelists to identify characteristics of shoppers and shopping households that are associated with
becoming informed about standard terms. Our data set includes the age
and sex of the head of the household, household income, household size,
and whether there are children present in the household.
In Table 2, we report summary statistics. For the sample of 48,154
visitors who accessed a minimum of two pages on at least one company’s
Web site in the sample during a single uninterrupted session, the average
age is 46.22 years, and the reported age range is 18–99 years. Average
income for heads of households is $60,502 (standard deviation,
$39,704). Income (and perhaps age) is top coded; median income
($37,500) better describes the sample. About half of the heads of households are male. The average number of household members is 2.78.
There are children in 41 percent of these households. Table 2 also shows
summary statistics for the sample of visitors who accessed a minimum
of five page visits on at least one company’s Web site and for unique
visitors who selected a product for purchase and began the checkout
5 . R ES U LT S
Our analysis here is based on shopping visits to company Web sites in
which the user accessed a EULA. We identify these visits by matching
the URLs corresponding to all the EULAs we collected to the clickstream
of URLs accessed by users during their company visits. We compute
descriptive statistics about company visits and EULA accesses under
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
User accessed at least two pages (N p 48,154):
Age (years)
Gender (1 p male)
Income ($)
Household size
Children (1 p yes)
User accessed at least five pages (N p 33,655):
Age (years)
Gender (1 p male)
Income ($)
Household size
Children (1 p yes)
User initiated checkout (N p 2,831):
Age (years)
Gender (1 p male)
Income ($)
Household size
Children (1 p yes)
Table 2. Characteristics of Users Visiting at Least One Sample Company’s Web Site for Three Definitions of a Visit
S TA N D A R D - F O R M C O N T R A C T S /
alternative definitions of a visit with intent to purchase. Finally, we
present regressions to study the determinants of the (as it turns out, low)
probability that a EULA will be accessed.
5.1. Company Visits and End-User License Agreement Accesses
We want to measure the fraction of buyers who seek to become informed
about EULA terms in deciding whether to purchase, and as noted we
define the sample to include only those company Web sites where EULA
access is possible but optional. Table 3 and Table 4 summarize the characteristics of visits to such companies, measured either as uninterrupted
sessions (Table 3) or as visits by unique users, aggregating all the monthly
sessions by individual users (Table 4). In each case, the data are presented
for each definition of a shopping visit to a company’s site. We separate
visits according to the type of company visited, noting that only retailer
visits include secure checkout page views; there is no need for a secure
checkout process for a free product. In addition to company visits, the
tables show information on page views and visits with EULA access.
The number of pages viewed before the first EULA access and the length
of time spent viewing EULAs give us some indication of shoppers’ level
of care or intent in accessing EULA pages. For simplicity, we report the
definition of a company visit of intermediate strictness (five page views)
for the uninterrupted sessions in Table 3, but the reader can explore
alternative definitions in Tables 3 and 4 and see that our results are
robust to alternative definitions of shopper and shopping visit.
When a visit is defined as requiring five or more pages accessed at
the company visited, there were 72,282 visits during uninterrupted sessions to software retailers and 13,715 to freeware companies. The median number of pages viewed in a given visit to a retailer was 10 pages,
and the median visit length was 183 seconds (3.05 minutes). Distributions of page views and visit durations are skewed. End-use license agreements were accessed 57 times among software retailers (.08 percent),
and 30 visits were made among freeware companies (.22 percent). The
median number of pages seen before accessing a EULA was eight for
retailers and four for freeware providers. These numbers are already
telling, but another consideration is whether shoppers who access the
EULA actually read it. For users in this group, the average length of
time spent on the EULA page was 62.7 seconds, and the median time
was 32 seconds. (Note that we are defining “access” as a EULA visit of
at least 1 second, for the purposes of obtaining a conservatively high
number of EULA accesses.) Half of the accesses to EULAs lasted less
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
Note. Values in parentheses are standard deviations.
User accessed at least one secure
checkout page:
User accessed at least five pages:
User accessed at least two
28,663 13.4
72,282 19.8
13,715 25
Mean Median
Visit Definition
Page Clicks per
Company Visit
568.9 207.5
303.4 101
436.8 183
Company Visit
Pages Viewed
EULA Accesses
before EULA EULA Access
! 30
EULA Visits
Median Visits
Mean Median Mean Median Seconds Minutes
Table 3. Company and End-User License Agreement (EULA) Visits: Uninterrupted Sessions
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
Note. Values in parentheses are standard deviations.
User accessed at least one secure
checkout page:
Mean Median
User accessed at least five pages:
User accessed at least two
Visit Definition
Page Clicks per
Company Visit
Company Visit
160.4 106
Pages Viewed
EULA Accesses
before EULA EULA Access
! 30
EULA Visits
Median Visits
Mean Median Mean Median Seconds Minutes
Table 4. Company and End-User License Agreement (EULA) Visits: Monthly Aggregates of Uninterrupted Sessions
22 / T H E J O U R N A L O F L E G A L S T U D I E S / V O L U M E 4 3 ( 1 ) / J A N U A R Y 2 0 1 4
than 30 seconds, and 90 percent spent less than 2 minutes on the contract.
To give these numbers some context, the average number of words
in EULAs for retail products in the sample (unreported) is 2,277, with
a median of 2,187 words and a standard deviation of 1,148. The time
spent on the EULAs relative to their length indicates that readers rarely
read terms in their entirety, especially as they are generally written in
complex legalese. Since consumers are unlikely to be aware of the default
rules, even if EULAs do spell out some terms in clear language, they
may still be misunderstood. Bailey and Bailey (1999) find that the average (nonlegalese) reading rate of American adults is 250–300 words
per minute, so a complete read of the typical EULA would require 8–
10 minutes, not 1 minute. In other words, even the small number of
EULAs accessed in our sample is still likely to be an overestimatation—probably a substantial one—of the number of effectively informed
readers. On the other hand, the small number of people who read EULAs
may not be representative of the average reader and may have developed
the ability to quickly skim the essential information.
Since our results could be biased if companies with relatively few
visits in our sample are systematically different in terms of the probability
of having visitors access their EULAs (for example, because visitors are
less likely to be ex ante familiar with the terms offered by such companies), we recalculated the frequency of EULA accesses using frequency
weights to adjust our data by the inverse of the total number of visits
at each seller. This resulted in lower rates of EULA access across all
definitions and thus addressed concerns that the lower frequency of visits
to certain types of firms in our sample could be biasing our observed
access rates downward.
Aggregating all monthly sessions of an individual user into a monthly
visit (shown in Table 4) leads to similar results. The overall results,
however, indicate that the impressions from Table 3 are robust to the
precise definition of company visits. Ultimately, the highest fraction of
readers among retail shoppers, across all shopper and session definitions,
is .65 percent, or about six readers per every 1,000 shoppers.
5.2. Interpreting the Results: Can This Be an Informed-Minority
Coming back to the empirical framework of Figure 1, we see that visitors
to the Web sites of the companies in our sample can be classified as
potential buyers or users visiting for other reasons. We measure the total
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
S TA N D A R D - F O R M C O N T R A C T S /
number of page views during each visit, whether a EULA was accessed,
and whether a secure checkout session was initiated. These data, reported for individual sessions in Table 3 and for monthly visitors in
Table 4, allow us to estimate the number of readers, buyers, and shoppers
by using access to a EULA page as a proxy for reading, initiation of the
checkout process as a proxy for buying, visits with five or more page
views as a proxy for identifying shoppers, and visits with between two
and five page views as a proxy for identifying nonshoppers. On the basis
of the data in Table 4, we estimate the number of monthly visitors in
our sample for each of the six categories shown in Figure 1.
We find that se1b1, or nine, readers who buy and se1(1 ⫺ b1), or 47,
readers who do not buy. There are s(1 ⫺ e1)b2, or 2,982, buyers who
are not readers and s(1 ⫺ e1)(1 ⫺ b2), or 40,670, shoppers who neither
read nor buy. Few nonshoppers would be expected to read EULAs, so
it is not surprising that the value of (1 ⫺ s)e2 is small; in our sample, it
equals five (of 25,664 visits). Finally, the large majority of nonshoppers
do not read EULAs: (1 ⫺ s)(1 ⫺ e2), or 25,661, on the basis of the above
proxies. We thus arrive at an estimate for e1, the fraction of shoppers
who constitute the informed minority of [se1b1 ⫹ se1(1 ⫺ b1)]/[se1b1 ⫹
se1(1 ⫺ b1) ⫹ s(1 ⫺ e1)b2 ⫹ s(1 ⫺ e1)(1 ⫺ b2)] p 56/43,708 p .13
It is possible that considering all visitors with five or more page views
as shoppers overestimates the number of shoppers. An alternative estimate could be obtained by assuming that, among actual shoppers, the
conversion ratio to initiate a checkout session among nonreaders is the
same as that for readers, 9/56 p 16.1 percent (which is higher than the
purchase conversion ratios of 2–5 percent cited in the marketing literature but reasonable if not all checkout sessions that we capture result
in actual purchases). In that case, the informed-minority fraction of all
shoppers would be the same as the fraction of buyers, namely, 9/2,991
p .30 percent.13,14
13. Most models of the informed minority predict that the conversion ratio for nonreaders would be the same as or higher than the conversion ratio for readers, as the latter
is less likely to purchase the product if they are not satisfied with the terms of the EULA.
An upper bound on the size of the informed minority can be obtained if we assume that
b2 p 100 percent, that is, that 100 percent of nonreaders proceed to purchase the product.
In that case, the informed minority would be 56/(2,991 ⫹ 47) p 1.84 percent of the total
number of shoppers.
14. If we assume that real shoppers will purchase from some merchant (while they may
visit many) and that among these shoppers the ones who constitute the informed minority
are equally likely to access a EULA at the sites of any of the merchants they visit, then
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
24 / T H E J O U R N A L O F L E G A L S T U D I E S / V O L U M E 4 3 ( 1 ) / J A N U A R Y 2 0 1 4
The bottom line is that the fraction of visitors who access EULAs is
very small, on the order of .1 percent. While a number of alternative
estimates can be calculated, these estimates point to that fraction being
well under 1 percent. Assuming that no other disciplining mechanisms
are at play, is it conceivable that such a small informed minority could
protect all buyers and discipline sellers into providing efficient contract
terms, thus preventing a market failure? The literature offers few meaningful suggestions as to how large the informed minority needs to be,
and these are typically provided in the context of illustrative examples.
Schwartz and Wilde (1979) offer an example in which the informed
minority needs to be 20–30 percent to be effective. Our estimates here
are imperfect, but they are two orders of magnitude smaller.15
Theoretically, the size of the informed minority required to induce
sellers to provide good terms depends on the trade-off between the gross
profit from selling to informed buyers (determined from the marginal
cost of the product) and the cost of providing better contract terms.
Consider a seller who may offer standard contract terms that are more
or less favorable to the buyers, which we will call good and bad terms,
respectively. The fraction r of buyers who become informed about the
terms reflects the cost of finding and reading the standard-form contract
and the expected benefit from doing so and is determined by the characteristics of the setting (for example, buyer search strategies as in
Schwartz and Wilde [1979]). In our sample, this corresponds to the
fraction e1 of shoppers who are in the informed minority. Buyers value
the rights and restrictions incorporated in the standard-form contract
(for example, warranty terms, the ability to transfer the product, and
so on), and thus good terms are valued more than bad terms. But good
terms are naturally more expensive for the seller to provide than bad
the behavior of visitors in our sample who initiate a checkout session is representative of
shoppers as a whole, and their likelihood of accessing EULAs (30 percent) provides an
estimate of the size of the informed minority among these most determined shoppers.
15. The estimates presented above are based on monthly visits as reported in Table 4.
They are conservative in the sense that using visits defined as individual sessions would
result in lower estimates for the size of the small minority. Using session data from Table
3 would result in seven readers who buy, 50 readers who do not buy, 4,859 buyers who
are not readers, 67,366 shoppers who neither read nor buy, and 59,447 nonshoppers who
include only six readers. The fraction e1 of shoppers in the informed minority would be
57/72,282 p .079 percent. The fraction of readers who initiate checkout sessions would
be 7/57 p 12.3 percent, and assuming the same conversion ratio for nonreaders would
give 7/4,866 p .14 percent as the informed minority. A conversion ratio of 100 percent
for nonreaders would give an upper bound for the informed minority of 57/(4,866 ⫹ 50)
p 1.16 percent.
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
S TA N D A R D - F O R M C O N T R A C T S /
terms, which results in corresponding product costs of cg and cb , with
cg 1 cb ≥ 0. Amending our earlier notation, we see that informed buyers
purchase with probability b1 if the terms are good and b3 if the terms
are bad (b1 1 b3), and uninformed buyers still purchase with probability
b2. The exact values of b1, b2, and b3 are determined by the characteristics
of the setting, but it is natural to consider b1 1 b2 1 b3. The seller offers
good terms if the expected payoff from doing so is higher than under
bad terms:
[rb1 ⫹ (1 ⫺ r)b2 ](p ⫺ cg) ≥ [rb3 ⫹ (1 ⫺ r)b2 ](p ⫺ c b).
Equivalently, the fraction of readers required to induce offering good
terms16 is
r ≥ (cg ⫺ c b)
Z[( )
b ⫺ b3
(c ⫺ c b) ⫹ 1
(p ⫺ c b) .
b2 g
This fraction becomes smaller as the incremental cost of providing good
terms decreases and as the probability that shoppers who become informed will drop out if they see bad terms increases.
Given certain values for these unknown parameters, any fraction of
informed shoppers could support an informed-minority equilibrium.
However, the market for software maintenance and support (M&S) can
be used to derive very rough estimates of the likely range of one of these
parameters, the marginal cost of good terms, and to put our observed
fraction of readers into perspective. Maintenance and support is a key
term in software EULAs,17 and thus the cost of supplying M&S should
be an order-of-magnitude approximation of the cost of offering good
EULA terms.
To estimate the cost of M&S terms, we obtained product price and
annual M&S price for 520 software products from the 42 software
companies in the sample of Marotta-Wurgler (2007) that provided M&S
separately on a periodic basis (that is, did not charge per incident). On
average, M&S is priced at 26 percent of the product price (exclusive of
the M&S). The median is 20 percent, and the standard deviation is 22
percent. Since there was high intercompany correlation, we focused on
16. An outcome in which the seller offers bad terms is inefficient if, assuming buyer
valuations Vg for good terms and Vb for bad terms, Vg ⫺ Vb 1 cg ⫺ cb, as it corresponds
to an inefficient provision of terms because the buyers value good terms above the seller’s
cost of providing them.
17. Marotta-Wurgler (2007) identifies and measures 23 important and common terms
that allocate rights and risks between buyers and sellers of software, and maintenance and
support (M&S) is one of these terms.
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
26 / T H E J O U R N A L O F L E G A L S T U D I E S / V O L U M E 4 3 ( 1 ) / J A N U A R Y 2 0 1 4
company means. The distribution for the 40 companies remaining after
dropping two obvious outliers has a mean of .29, a median of .24, and
a standard deviation of .16.
Thus, a year of M&S for software is on average priced at 25–30
percent of the product price.18 Since M&S costs are primarily variable
(labor) costs, if the market for M&S was perfectly competitive, this
would provide some indication of the marginal cost of M&S and thus
a floor on the marginal cost of pro-consumer EULA terms. There are
several reasons why 25–30 percent of the product price may be too high
an estimate: (1) consumers may be more likely to purchase M&S from
the seller of the software, and thus software companies may price as a
two-part tariff, with a lower price for the up-front purchase (the software
product) and a higher price for the subsequent purchase (M&S), (2)
consumers who purchase M&S are likely to have higher M&S costs due
to adverse selection and/or moral hazard, and (3) software companies
may have substantial market power in providing M&S because of barriers to entry for competitors that are not as familiar with their product
or consumers’ propensity to purchase M&S from the seller of the original
software. All of these factors would result in an M&S-to-product price
ratio that is higher than the cost of providing M&S. On the other hand,
M&S is only one of 23 key EULA terms, which include several other
types of warranties and permissions to copy or distribute the software
that can impose opportunity costs. Furthermore, M&S pricing is similar
in enterprise software markets, where significant competition exists from
third-party M&S providers and purchase of M&S contracts is almost
On balance, it is reasonable to assume that the cost of this level of
M&S is around 20 percent of the product price p, or .2p. Dividing the
numerator and denominator of the fraction of informed buyers necessary
to induce the seller to offer good terms by p, we get
( )Z[( )
cg ⫺ c b
( )]
b1 cg ⫺ c b b1 ⫺ b3
1⫺ b .
If (b1 ⫺ b3)/b2 ≤ 1—which would be the case if readers purchase with
at least the same probability as nonreaders when they discover good
terms (bad terms reduce or eliminate this probability)—b1 ≥ b2 and thus
1 ⫺ b1/b2 ≤ 0, and cb ≥ 0 and thus 1 ⫺ cb /p ≤ 1, we get r ≥ (cg ⫺ cb)/
18. One year was the most common duration—as well as the median duration—of free
M&S for the companies that provided such a period of free M&S.
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
S TA N D A R D - F O R M C O N T R A C T S /
p, or r ≥ .2. In other words, if the cost of providing good terms is .2p
and the three reasonable assumptions in the previous sentence hold, as
is the case in our data, then one would require r ≥ .2 to support an
informed-minority equilibrium. This is 200 times larger than the value
of r ≈ .001 (or .1 percent) that we observe in the data. Alternatively,
our data suggest that, for the informed buyers to induce the seller to
offer good terms, their incremental cost would have to be almost negligible at less than .1 percent of the selling price.19 In either case, our
data are unlikely to be consistent with an informed-minority equilibrium.
A similar and perhaps simpler approach to whether we might be
observing an informed-minority equilibrium is to look at the seller calculus for visitors who initiate a checkout session. In an informedminority equilibrium, sellers would be offering good terms, and according to Table 4, EULAs are accessed in six of 3,534 visits with checkout
sessions. Since initiating a checkout session is a requirement to complete
a purchase, if sellers were to offer bad terms, they might lose up to 100
percent of these readers. Thus, if sellers are trading off the net revenue
from the sales to the six readers against the cost of providing good terms
to the 3,528 nonreaders, as would be the case if we were observing an
informed-minority equilibrium, sellers would keep offering good terms
if the cost of doing so was less than .17 percent of the selling price.
This argument relies on fewer assumptions than that developed
above, and it once again leads to the same conclusion, namely, that our
data seem inconsistent with an informed-minority equilibrium. Hence,
if the informed minority were the only disciplining mechanism, the natural implication would be that sellers are offering the lowest quality
M&S terms. As they do not appear to take such advantage, it must be
that other mechanisms are at play and that (perhaps) in the absence of
such mechanisms, more buyers would read.
5.3. Robustness Checks: Becoming Informed without Reading?
Consumers can become informed in a number of ways. The Internet
contains several consumer product review sites, blog posts with comments about product quality, and specialized news outlets that might
19. Of course, one could argue that offering a pro-buyer dispute resolution clause might
cost the seller .1 percent of the price. Although it is theoretically possible, we find it unlikely
that sellers will change this clause to in response to the threat of losing one in 1,000 buyers.
More important, the terms that matter the most to consumers (such as M&S and warranties)
are among the costliest ones.
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
28 / T H E J O U R N A L O F L E G A L S T U D I E S / V O L U M E 4 3 ( 1 ) / J A N U A R Y 2 0 1 4
convey the content of standard-form contract terms to a few more shoppers.
From the Alexa Web site,20 we obtained a list of 25 of the most
trafficked sites likely to have information about EULA terms, product
quality information, and sellers’ practices. We then measured the rate
at which shoppers in our sample accessed these sites. Very few sites focus
exclusively on software and EULA terms. Others that occasionally discuss EULA terms are technology-related news sites, such as Wired and
Ars Technica; sites that offer general consumer protection news and rants
about abusive practices by sellers, such as Consumerist; and sites with
general technology news, such as PC Magazine. Finally, there are general
consumer information sites that do not focus on software but contain
software product reviews, such as Consumer Reports.
We reviewed the pages accessed by shoppers visiting each URL associated with these 25 sites to make sure that the pages accessed contain
information about EULAs. Of the 131,729 sessions with at least two
pages accessed, only three shoppers accessed pages with EULA information in consumer review sites.21 At the monthly level of aggregation,
11,657 (16.8 percent) visits accessed at least one of the 25 consumer
sites, but none of that activity was related to EULAs. Of this group,
only 69 shoppers accessed pages with particular software product reviews or pages containing information such as reviews of tax software
in that month. A total of 84 shoppers accessed news or general information pages about software, such as how to obtain certain free plugins. The remaining visits were to pages unrelated to software or EULA
terms, so it does not appear that consumers are becoming informed by
these alternative means.
It is also possible that the quality of terms is reflected in the seller’s
price, and thus shoppers do not need to become informed about these
terms. As is the tenet of the informed-minority hypothesis, however, for
terms to be reflected in prices, a sufficient number of consumers need
to be informed about terms when they comparison shop (unless all sellers
offer the worst enforceable terms, and consumers know this and therefore have no incentive to become informed about terms). MarottaWurgler (2007) finds that most sellers offer terms worse than those
provided by default rules but not so bad that no information can be
20. The top Web sites were obtained from Alexa, Top Sites (
21. Two shoppers downloaded the EULAlyzer software from Javacool Software to obtain
a review of a particular EULA. The third accessed a reference to a mock EULA.
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
S TA N D A R D - F O R M C O N T R A C T S /
gained by becoming informed about them. Our results are consistent
with this, although we are not able to draw conclusions about the economic efficiency of this outcome.
Marotta-Wurgler (2007) is particularly relevant to this paper, as it
looks specifically at the relationship between EULA term bias and price
and finds almost no detectable relationship between them. It is possible
that this is because terms for prepackaged software products are a relatively small component of price, and it is hard to measure the termprice relationship without controlling for the many other product attributes that affect price. However, this finding is also consistent with
our results, as sellers would have little incentive to adjust their prices to
reflect the quality of EULAs if consumers fail to become informed and
thus compare terms. In such a situation, we would indeed expect that
consumers can infer very little about EULA terms just by looking at
In addition, although consumers might be expected to have stronger
incentives to become informed for higher priced products, the effect of
price on the propensity to read terms is not statistically significant. However, we refrain from interpreting this evidence too strongly, because our
investigation of why consumers do or do not read is limited compared
with what we can explore directly, namely, the prediction that they read
at all.
Another possibility is that the degree of disclosure of the EULA terms
(for example, the prominence with which they are displayed) signals
their quality. For instance, an equilibrium may arise in which good terms
are prominently displayed while poor terms are made harder to find or
inaccessible. Our results show that such an equilibrium, if it exists in
our setting, is not maintained based on an informed minority. Furthermore, Marotta-Wurgler (2009) finds that prominently disclosed clickwrap contracts were roughly as one-sided as non-click-wrap contracts
that required more effort to discover. This suggests that firms are not
using the form of disclosure to signal the quality of terms. They do not
seem to use click wraps to increase awareness of good terms, and they
do not seem to use browse wraps to hide bad ones.
5.4. Determinants of End-User License Agreement Visits
Here we briefly examine those characteristics of the company, product,
user, and Web site that distinguish the readers (or, more precisely, the
clickers) from the nonreaders. Table 5 reports logit regressions in which
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
At Least
Five Pages
At Least One Secure
Checkout Page
At Least
Two Pages
Note. The dependent variable is EULA access. Standard errors, clustered by visitor, are in parentheses.
Significant at the 10 percent level.
*Significant at the 5 percent level.
**Significant at the 1 percent level.
ln Age
ln Income
Pages viewed
Public company
ln Revenue per product
ln Median price
At Least
Two Pages
Uninterrupted Sessions
Table 5. Logit Regressions of Determinants of Online Visits to End-User License Agreements (EULAs)
At Least
Five Pages
At Least One Secure
Checkout Page
Monthly Aggregates of
Uninterrupted Sessions
S TA N D A R D - F O R M C O N T R A C T S /
the dependent variable is a dummy for whether a EULA was accessed
during a particular company visit.
Factors that have a positive, albeit very small, effect on propensity
to read are whether the product offered is freeware (since consumers
might think there is a catch) and the number of pages visited in a session
(longer visits might indicate a more serious intent to shop). A factor that
has a negative effect on readership is the natural log of a company’s
revenues (as a proxy for size) divided by the number of products. A
reason behind this is that consumers may be less likely to access EULAs
for products they know about and thus may trust, relying instead on
product familiarity. We use average product revenues as a proxy of
consumer trust.22 Still, the main result here is that the most important
term in the regression is the constant term: EULAs are rarely read by
5.5. Why Do So Few Consumers Read?
The small fraction of consumers accessing EULAs suggests a high cost
of finding the EULA and reading its terms. If the primary cost lies in
locating the contract, then mandating disclosure should increase the
fraction of informed consumers. However, Marotta-Wurgler (2012) analyzes this same data set and finds that increased contract disclosure is
not associated with increased readership. This suggests that the primary
cost facing consumers is in reading and comprehending contract terms;
it is possible that consumers do not access these terms even if they can
do so with only the click of the mouse, as they expect reading EULA
terms to be prohibitively costly. Measures that reduce the cost of comprehending the contract terms are likely to be more successful in increasing the fraction of informed consumers. Thus, a regulatory approach
focusing on shortening and simplifying online contracts, standardizing their
terms, and providing a standardized summary is more likely to increase
readership than an approach focusing solely on disclosure. Regulations
that mandate the disclosure of basic credit terms in a standardized manner
and large fonts, such as the Schumer box in the United States and the
22. In unreported regressions, we examined whether shoppers are less likely to read the
EULAs of products that are more likely to involve repeat purchases (such as upgrades)
because they may be familiar with the terms from previous use. Other products, such as
test preparation software, are less likely to be purchased repeatedly. The relationship between a dummy indicating whether a product is likely to be purchased repeatedly and the
users’ propensity to access EULAs is generally positive and, under certain definitions of
shoppers, is statistically significant.
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
32 / T H E J O U R N A L O F L E G A L S T U D I E S / V O L U M E 4 3 ( 1 ) / J A N U A R Y 2 0 1 4
summary box in the United Kingdom, can reduce the cost of reading and
comprehending contract terms. Simple and plain language requirements
can be seen in the same light.
Consumer access to the terms of standard-form contracts has been at
the center of a legal and policy debate, and a major question has been
whether disclosure of terms in standard-form contracts that govern consumer transactions should be regulated. A related debate has focused
on the enforceability of terms and possible need to regulate disclosure
for software in general and software purchased online in particular. A
central issue in these debates is the validity of the informed-minority
hypothesis: the view that comparison shoppers for standard terms help
sustain efficient terms in equilibrium. In this paper, we investigate the
extent to which consumers actually access the terms of certain standardform online contracts. Our clickstream data allow us to measure the
informed minority with reasonable precision for the first time.
We find that very few consumers choose to become informed about
standard-form online contracts. In particular, we estimate that the fraction of retail software shoppers who access EULAs is between .05 percent
and .22 percent, and most of the few shoppers who do access EULAs
do not spend enough time doing so to have digested more than a fraction
of their content. We also document that shoppers rarely access substitute
sources of information, such as sites containing consumer product reviews or relevant news, to learn about EULA terms. Even under generous
assumptions, it is difficult to envision the probability that EULAs are
read (and understood) growing even to 1 percent. Our estimates of the
size of the informed minority in this market are one or two orders of
magnitude smaller than examples offered in the literature for the size
required to sustain an informed-minority equilibrium, and this is further
confirmed by simple theoretical calculations.
It is possible that buyers in the market for online software believe
that other factors will discipline sellers in terms of the EULA terms they
offer and thus rationally decide not to become informed and establish
an informed minority in this market. Our data do not speak to whether
it is rational or irrational to read or whether or not terms are efficient.
Still, the market that we study offers a very favorable setting to look
for an informed minority of buyers because of the low costs of accessing
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
S TA N D A R D - F O R M C O N T R A C T S /
contracts and shopping between firms and the presence of new and
unknown firms that have yet to establish reputations.
While our results apply directly only to one context, the fact that
online comparison shopping is so cheap and easy raises questions about
whether informed-minority mechanisms could be consequential in other
contexts. Furthermore, if that is the case, an approach similar to our
own should result in readership levels that are substantially higher than
what we observed. Assessing the presence of other factors that discipline
sellers in the market for online software as well as whether readership
levels are adequate to support the informed-minority mechanism in other
markets are promising directions for further research.
The low readership of contract terms, even for those EULAs for which
the terms are prominently accessible and consumers are required to acknowledge reading and agreeing to the terms before purchasing a product, suggests that it is the cost of reading and assessing these terms,
rather than the cost of finding them, that discourages consumers. Hence,
regulation that aspires to promote the emergence of an informed fraction
of consumers solely through additional disclosure may be too optimistic.
If the goal is achievable, it will require making contract terms easier to
read, understand, and compare.
American Law Institute. 2010. Principles of the Law of Software Contracts.
Philadelphia: American Law Institute.
Bailey, R. W., and L. M. Bailey. 1999. Reading Speeds Using RSVP. User Interface
Update. February.
Baird, Douglas. 2006. The Boilerplate Puzzle. Michigan Law Review 104:933–
Bakos, Yannis. 2001. The Emerging Landscape for Retail E-Commerce. Journal
of Economic Perspectives 15(1):69–80.
Bar-Gill, Oren. 2004. Seduction by Plastic. Northwestern University Law Review
Beales, Howard, Richard Craswell, and Steven C. Salop. 1981. The Efficient
Regulation of Consumer Information. Journal of Law and Economics 21:
Bebchuk, Lucian A., and Richard A. Posner. 2006. One-Sided Contracts in Competitive Consumer Markets. Michigan Law Review 104:827–36.
Becher, Shmuel I., and Esther Unger-Aviram. 2010. The Law of Standard Form
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
34 / T H E J O U R N A L O F L E G A L S T U D I E S / V O L U M E 4 3 ( 1 ) / J A N U A R Y 2 0 1 4
Contracts: Misguided Intuitions and Suggestions for Reconstruction. DePaul
Business and Commercial Law Journal 8:199–227.
Ben-Shahar, Omri. 2009. The Myth of the Opportunity to Read in Contract
Law. European Review of Contract Law 5:1–28.
Ben-Shahar, Omri, and Carl Schneider. 2011. The Failure of Mandated Disclosures. University of Pennsylvania Law Review 159:647–749.
Bucklin, Randolph E., and Catarina Sismeiro. 2003. A Model of Web Site Browsing Behavior Estimated on Clickstream Data. Journal of Marketing Research
Catledge, Lara D., and James E. Pitkow. 1995. Characterizing Browsing Behaviors on the World Wide Web. Computer Networks and ISDN Systems 27:
Eisenberg, Melvin A. 1995. The Limits of Cognition and the Limits of Contract.
Stanford Law Review 47:240–41.
Gabaix, Xavier, and David Laibson. 2006. Shrouded Attributes, Consumer Myopia, and Information Suppression in Competitive Markets. Quarterly Journal of Economics 121:505–40.
Gillette, Clayton P. 2004. Rolling Contracts as an Agency Problem. Wisconsin
Law Review, pp. 679–722.
———. 2005. Pre-approved Contracts for Internet Commerce. Houston Law
Review 42:975–1013.
Goldberg, Victor P. 1997. The “Battle of the Forms”: Fairness, Efficiency, and
the Best-Shot Rule. Oregon Law Review 76:155–65.
Hillman, Robert. 2006a. On-line Boilerplate: Would Mandatory Website Disclosure of E-Standard Terms Backfire? Michigan Law Review 104:837–56.
——. 2006b. Online Consumer Standard Form Contracting Practices: A Survey
and Discussion of Legal Implications. Pp. 283–312 in Is Consumer Protection
an Anachronism in the Information Economy? edited by Jane K. Winn. Burlington, Vt.: Ashgate Publishing.
Hillman, Robert, and Jeff Rachlinski. 2002. Standard Form Contracting in the
Electronic Age. NYU Law Review 77:429–95.
Johnson, Eric J., Wendy W. Moe, Peter S. Fader, Steven Bellman, and Gerald L.
Lohse. 2004. On the Depth and Dynamics of Online Search Behavior. Management Science 50:299–308.
Katz, Avery. 1990. Your Terms or Mine? The Duty to Read Fine Print in Contracts. RAND Journal of Economics 21:518–37.
King, Gary, and Langche Zeng. 2001. Logistic Regression in Rare Events Data.
Political Analysis 9:137–63.
Lal, Rajiv, and Carmen Matutes. 1994. Retail Pricing and Advertising Strategies.
Journal of Business 67:345–70.
Lemley, Mark. 2006. Terms of Use. Minnesota Law Review 91:459–83.
Mandell, Lewis. 1971. Consumer Perception of Incurred Interest Rates: An Em-
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions
S TA N D A R D - F O R M C O N T R A C T S /
pirical Test of the Efficacy of the Truth in Lending Law. Journal of Finance
Mann, Ronald, and Travis Siebeneicher. 2008. Just One Click: The Reality of
Internet Contracting. Columbia Law Review 108:984–1012.
Marotta-Wurgler, Florencia. 2007. What’s in a Standard Form Contract? An
Empirical Analysis of Software License Agreements. Journal of Empirical
Legal Studies 4:677–713.
———. 2008. Competition and the Quality of Standard Form Contracts: The
Case of Software License Agreements. Journal of Empirical Legal Studies 5:
———. 2009. Are “Pay Now, Terms Later” Contracts Worse for Buyers? Evidence from Software License Agreements. Journal of Legal Studies 38:309–
———. 2012. Does Contract Disclosure Matter? Journal of Institutional and
Theoretical Economics 168:94–119.
Moe, Wendy W., and Peter S. Fader. 2004. Dynamic Conversion Behavior at eCommerce Sites. Management Science 50:326–35.
Plaut, Victoria C., and Robert P. Bartlett. 2012. Blind Consent? A Social Psychological Investigation of the Non-readership of Contracts. Law and Human
Behavior 36:293–311.
Priest, George. 1981. A Theory of the Consumer Product Warranty. Yale Law
Journal 90:1297–1352.
Salop, Steven. 1976. Information and Monopolistic Competition. American Economic Review 66:240–45.
Salop, Steven, and Joseph Stiglitz. 1977. Bargains and Ripoffs: A Model of
Monopolistically Competitive Price Dispersion. Review of Economic Studies
Schwartz, Alan, and Louis Wilde. 1979. Intervening in Markets on the Basis of
Imperfect Information: A Legal and Economic Analysis. University of Pennsylvania Law Review 127:630–82.
Spence, Michael. 1977. Consumer Misperceptions, Product Failure and Producer
Liability. Review of Economic Studies 44:561–72.
This content downloaded from on Tue, 16 Jun 2015 15:10:39 PM
All use subject to JSTOR Terms and Conditions