Mossavar-Rahmani Center for Business & Government
Weil Hall | Harvard Kennedy School | www.hks.harvard.edu/mrcbg
M-RCBG Associate Working Paper Series | No. 97
The views expressed in the M-RCBG Associate Working Paper Series are those of the author(s) and do
not necessarily reflect those of the Mossavar-Rahmani Center for Business & Government or of
Harvard University. The papers in this series have not undergone formal review and approval; they are
presented to elicit feedback and to encourage debate on important public policy challenges. Copyright
belongs to the author(s). Papers may be downloaded for personal use only.
Trust mechanisms and online platforms: A
regulatory response
Mitchell Watt
Hubert Wu
August 2018
Trust mechanisms and online platforms: a regulatory response
Mitchell Watt and Hubert Wu
Mossavar-Rahmani Center for Business and Government Working Paper
A previous version was submitted 27 March 2018 in partial fulfillment of the requirements for the degree
of Master in Public Policy.
Faculty Advisor: Professor Jason Furman
Seminar Leader: Professor John Haigh
This document reflects the views of the author and should not be viewed as representing the views of
any external clients, nor those of Harvard University or any of its faculty.
Acknowledgements
We would like to thank Professor Jason Furman, Professor John Haigh, and Derek Moore,
for their support during this project.
We would also like to acknowledge the following academics and researchers for helpful
discussions and advice on this project including:
Professor Morris Kleiner – Humphrey School of Public Affairs, University of
Minnesota
Professor John Horton – Stern School of Business, New York University
Professor Chiara Farronato – Harvard Business School
Professor Scott Kominers – Harvard Business School
Professor Michael Luca – Harvard Business School
Professor Brad Larsen – Department of Economics, Stanford University
Professor Joseph Aldy – Harvard Kennedy School
Researchers from the European Commission
In preparing this report, we also corresponded with current and former employees of the
following organizations: Uber, Airbnb, Facebook, Slack, Upwork and Yelp. We thank them
for their assistance, and have respected their wishes to maintain anonymous with regard to
their comments.
Contents
Trust mechanisms and online platforms: a regulatory response ............................................................. 1
Acknowledgements ................................................................................................................................ 3
Executive summary ................................................................................................................................ 5
The rise of online platforms .................................................................................................................... 7
Facilitating exchanges on online platforms: the use of trust mechanisms ............................................... 9
Trust mechanisms and the role of regulators ........................................................................................ 11
A new classification strategy for trust mechanisms ............................................................................... 13
Our approach to classification............................................................................................................... 13
Lessons from classification ................................................................................................................... 16
Benefits and costs associated with trust mechanisms .......................................................................... 19
The benefits of trust mechanisms ......................................................................................................... 19
The harms of trust mechanisms ............................................................................................................ 31
Assessing the costs and benefits of trust mechanisms ......................................................................... 37
Recommendations ................................................................................................................................ 38
The broad implications of our findings for our report’s recommendations ............................................. 38
Guidance for regulating trust mechanisms ............................................................................................ 39
Business-facing recommendations ....................................................................................................... 40
Recommendations for government relationships .................................................................................. 41
Bibliography .......................................................................................................................................... 43
Appendix A.1 – Online retail platform profiles ....................................................................................... 50
Appendix A.2 – Short-term accommodation platform profiles ............................................................... 56
Appendix A.3 – Ridesharing platform profiles ....................................................................................... 60
Appendix A.4 – Online freelance labor platform profiles........................................................................ 64
Appendix A.5 – Online advertising platform profiles .............................................................................. 68
Appendix B – Top and bottom professions on O*Net variables ............................................................. 72
Appendix C – Benefits estimation methodology and the full regulatory substitution index ..................... 73
Appendix D – Example database profile ............................................................................................... 79
Appendix E – Design guidelines for businesses ................................................................................... 81
Executive summary
Online platforms are rapidly transforming the
global economy. These businesses, including
six of the world’s ten largest companies, have
evolved new solutions to overcome trust
problems and asymmetries of information
inherent in exchange on the Internet. These
features are known as trust mechanisms.
Trust mechanisms are challenging old
approaches to regulation, with new technology
rendering many existing regulations
unnecessary or unsuitable. Some online
platforms have pursued new forms of self-
regulation; others have taken an adversarial
stance towards government intervention in its
entirety. Meanwhile, standard approaches of
regulators towards trust mechanisms are yet to
be developed.
This report examines some of the benefits and
costs of trust mechanisms in the digital
economy and develops recommendations for
regulators to adapt their approach in response
to these new business models.
First, we map the space of trust mechanisms
operating today, and propose a classification
schema focused on the participants,
informational content and function of a trust
mechanism. While online platforms differ in
terms of the transparency of their trust
mechanisms, we show that design choices
depend on the nature of the trust problem faced
in the industry. We also demonstrate that firm-
level differences in trust mechanisms may
impact the competitive dynamics of an industry.
Second, we present evidence on three
categories of benefits of trust mechanisms: a
reduction in the regulatory burden for
businesses, the expansion of markets enabled
by trust mechanisms and an enhanced ability of
governments to target spending using
information gathered by online platforms. The
impact of trust mechanisms on occupational
licensing alone is estimated to be a decreased
regulatory burden of more than $790 million.
Third, we analyze the potential harms to the
market that may be associated with trust
mechanisms, including new forms of
discrimination; the possibility of new market
failures and imperfections; and the possibility of
strategic manipulation with competitive
implications for markets.
Our report concludes with several
recommendations for business regulators
based on our analysis and findings.
Summary of recommendations
Based on our analysis, we believe regulators should focus on reducing the potential for harms
caused by trust mechanisms while maximizing the likelihood of their economic benefits. With this
goal in mind, we recommend that regulators:
1. Investigate the development of an online database of information about the characteristics
and function of trust mechanisms employed by platforms.
2. Require businesses to release publicly information about the characteristics and functions of
trust mechanisms employed on their platform.
3. Issue guidelines to businesses concerning how to minimize potential harms caused by trust
mechanisms in online platforms.
4. Write to state and local authorities about areas in which occupational licensing laws could be
weakened in response to the emergence of trust mechanisms.
5. Investigate areas where regulators’ activities could be better targeted using data from trust
mechanisms.
The rise of online platforms
Online platforms – businesses that create
value by facilitating exchanges between two or
more interdependent groups – are rapidly
remaking swathes of the US and global
economy. At the time of writing, six of the
world’s ten largest companies (by market
capitalization) possess online platforms as their
dominant operating model, or a significant
portion of their activities. For some, digital
platforms and the outcomes they produce are
“little short of miraculous”,
1
“the maws into
which traditional companies are now
disappearing”,
2
and a case in point of Karl
Marx’s 1859 observation about how technology
shapes economic institutions.
3
Our operating
definition of a platform is contained in Box 1.
Platforms are not a recent invention. For
example, physical marketplaces, newspapers,
stock exchanges, auction houses, and credit
cards are all platforms which have existed for
many years, and sometimes even centuries.
Even certain digital platforms have existed for
some time, dating back approximately to the
1
Parker, Geoffrey G., Marshall W. Van Alstyne, and
Sangeet Paul Choudary. Platform Revolution: How
Networked Markets Are Transforming the Economy
and How to Make Them Work for You. WW Norton &
Company, 2016, p. 5.
2
Manville, Brook. “Are Platform Businesses Eating
the World?,” Forbes, 14 February, 2016,
https://www.forbes.com/sites/brookmanville/2016/02/
14/are-platform-businesses-eating-the-world/.
(Accessed 14 December 2017).
3
Weyl, E.G. and Alexander White. “Let the Right
'One' Win: Policy Lessons from the New Economics
of Platforms.” Coase-Sandor Working Paper Series
in Law and Economics No. 709, 2014, p. 1.
creation of the World Wide Web. For example,
Amazon Marketplace and eBay were each
founded nearly two decades ago and
today retain much of their original operation
models.
Despite this, online platforms are rapidly
becoming increasingly influential in the US and
global economy. This can be seen in recent
changes to the distribution of the world’s largest
companies, as in Figure 1. This scale extends
beyond these firms’ economic size. For
instance, Facebook and Google jointly
accounted for 99% of all new digital advertising
and approximately two-thirds of US digital ad
investment in 2017.
4
,
5
They are also among the
world’s most visited websites: of the world’s 50
most popular websites, all were either online
platforms or had platform elements.
6
4
PwC and Interactive Advertising Bureau. IAB
internet advertising revenue report: 2016 full year
results. 2017.
5
“Google and Facebook Tighten Grip on US Digital
Ad Market: Duopoly to grab more than 60% of the
2017 digital ad spend.” eMarketer, 21 September,
2017, https://www.emarketer.com/Article/Google-
Facebook-Tighten-Grip-on-US-Digital-Ad-
Market/1016494. (Accessed 21 March 2018).
6
Alexa Internet. “Alexa Top 500 Global Sites.”
https://www.alexa.com/topsites, 2018, (Accessed 4
January 2018).
Key takeaways
Online platforms are rapidly remaking the global economy, with six of the ten largest
companies in the world now platform businesses.
Trust mechanisms are the tools that power online platforms by enabling trust and allowing
transactions to occur.
These trust mechanisms present a suite of new challenges for regulators.
Box 1: What are platforms and online platforms?
A platform
is a business model that creates value by facilitating exchanges between two or more
interdependent groups. Platforms are also known as two-sided markets (for two groups) or multi-
sided markets (for more than two groups).
An online platform is a platform that substantively utilizes information technology (such as Internet
connectivity) as well as non-physical environments like websites and mobile applications, in order to
operate.
Platforms are distinct from networks
, which enable connections between like groups. An example of
the difference between the two is Microsoft Instant Messenger (an online communication network
that connects like groups) and Facebook (a platform which in addition to being a communication
network also connects individuals with advertisers).
Figure 1: Top five publicly traded companies in the world by market capitalization
Facilitating exchanges on online platforms: the use of trust mechanisms
All transactions require a minimum level of trust
between participants in order to occur. This is
because any exchange requires a credible
before the fact commitment that no parties will
renege on their side of the agreement after the
fact. Without this, transactions may not occur
even if they would benefit both parties.
7
Box 2
illustrates this in a simple economic game.
Levels of trust among Americans have
fluctuated over time. Most recently, in 2016,
some 31% of Americans in the General Social
Survey said that ‘most people can be trusted’.
8
As Figure 2 demonstrates, this represents a
significant decline in general trust since the
7
See Greif, Avner. "The fundamental problem of
exchange: a research agenda in historical
institutional analysis." European Review of
Economic History 4, no. 3 (2000): 251-284.; and
Akerlof, George A. "The market for “lemons”: Quality
uncertainty and the market mechanism."
In Uncertainty in Economics, pp. 235-251. 1978.
8
Smith, Tom W, Peter Marsden, Michael Hout, and
Jibum Kim. “General Social Surveys.” National
Opinions Research Center. 1972-2016. Data
1970s, with an uptick in
recent years that some have attributed to the
rise of the ‘sharing economy’ and its associated
increase in trust-dependent exchanges with
relative strangers.
9
Online platforms face (at least) three additional
trust challenges that distinguish them from other
kinds of firms:
1. Parties in online environments are often
anonymous to each other and decoupled
from their offline identities.
10
accessed from the GSS Data Explorer website at
gssdataexplorer.norc.org.
9
Tanz, Jason. “How Airbnb and Lyft finally got
Americans to trust each other.” Wired, 23 April,
2014, https://www.wired.com/2014/04/trust-in-the-
share-economy/. (Accessed 19 March 2018).
10
Or, as The New Yorker put it in its famous 1993
cartoon by Peter Steiner: “On the Internet, nobody
knows you’re a dog” (Steiner Peter. ‘”On the
Internet, nobody knows you’re a dog”’ [Cartoon]. The
New Yorker, 5 July, 1993).
Figure 2: Level of trust among Americans, 1972-2016
2. Physical interactions that traditionally
occur in offline environments are often
impossible in an online environment.
3. Online platforms possess the ability to
collect and use a large amount of data
about participants and their activities.
As a result of factors including these, platforms
have developed novel and diverse ways to
facilitate exchange between their
participants.
11
,
12
These solutions form a rich
11
Martens, Bertin. “An Economic Policy Perspective
on Online Platforms.” Institute for Prospective
Technological Studies, Digital Economy Working
Paper 2016/05. 2016.
world of ratings systems, user-generated
reviews, profiles, public transaction
histories, centralized guarantees, and many
other means to overcome to overcome both the
information asymmetries that exist between
participants on a platform.
In this report, we call these trust mechanisms,
with our operational definition discussed in Box
3 below.
12
Marina Lao et al. “The ‘Sharing’ Economy: Issues
Facing Platforms, Participants & Regulators” A
Federal Trade Commission Staff Report. 2016, p. 9,
35.
Box 2: Why trust is needed for transactions to occur
To see why transactions between two parties may not occur on their own, consider the following game,
known as the game of trust.
Player 1 starts by choosing whether to enter into an exchange with Player 2. If Player 1 chooses to
do so, Player 2 then chooses to either fulfil her contractual obligations or to renege on the agreement.
Entering into an exchange yields a payoff of
that is divided such that both players better off than if
they do not choose to exchange.
However, Player 2 can gain more than
by reneging on the agreement. In this occurs, Player 1
receives a payoff
and is worse off than had the exchange not been initiated in the first place.
Anticipating this, Player 1 will not choose to enter into an exchange with Player 2 to begin with.
Trust mechanisms and the role of regulators
Enabling trust and quality in traditional
marketplaces has been one of the core goals of
trade regulators around the world. Trust
mechanisms seek to achieve a similar goal but
in a private setting. In this sense, trust
mechanisms can be seen as a challenge to the
core function of regulators, and has led the US
Federal Trade Commissioner (FTC) Maureen
Ohlhausen to ask “Can the trust mechanisms
built into some of these new business models
replace regulation? If so, where?” as one of the
five most important questions in the sharing
economy today.
There is also uncertainty about the appropriate
balance of regulation for online platforms, when
compared with more traditional firms. Online
platforms often compete with traditional
suppliers of goods and services, but regulation
appropriate to these firms may not work in the
online context. On the other hand, many
traditional firms also believe that online
platforms should be subject to the same or
similar level or regulation to ensure a level
playing field.
Exploring these roles for regulators is a key
goal of this report. Investigation of these
problems is also more than a theoretical
exercise. For some time already, regulatory
challenges relating to trust mechanisms and
Box 3: What is a trust mechanism?
A trust mechanism is a tool used by an online platform to overcome information asymmetries
between market participants to facilitate transactions. Trust mechanisms can take many forms,
including:
Online platform
Key participants
Trust mechanism
1
eBay
Buyers and sellers
‘Feedback Score’
cumulative rating score
Verified Rights Owner
(VeRO) program
Uber
Drivers and passengers
Two-way rating system
(star rating + comments
on other factors)
Yelp
Restaurants and diners
Star rating system
User reviews
Recommendation
software (an automated
“review filter”)
Facebook
Users and advertisers
User profiles
Airbnb
Renters and landlords
Star ratings on multiple
factors
Written reviews
Medium
Readings and writers
‘Clapping’ piece
approval system
online platforms have come into direct contact
with the regulators. For example:
In December 2017, Yelp’s Vice President of
Global Public Policy contacted Federal
Trade Commission’s Acting Chair Maureen
Ohlhausen asserting that Google violated a
commitment made to the FTC in 2013 to not
scrape content (including pictures and
reviews) from certain websites.
13
The rollout of Uber and AirBnb in hundreds
of international cities has posed challenges
for regulators of the passenger transport
and accommodation industries in the US
and other countries.
In May 2017, the FTC settled a complaint
with two online trampoline sellers that
deceived consumers by directing them to
review websites that falsely claimed to be
independent but were in fact wholly owned
by the sellers.
14
13
Nicas, Jack. “Google Rival Yelp Claims Search
Giant Broke Promise Made to Regulators.” The Wall
Street Journal, 11 September, 2017,
https://www.wsj.com/articles/google-rival-yelp-
claims-search-giant-broke-promise-made-to-
regulators-1505167498. (Accessed 9 January 2018).
14
Federal Trade Commission. “FTC Stops False
Advertising, Phony Reviews by Online Trampoline
Sellers,” https://www.ftc.gov/news-events/press-
releases/2017/05/ftc-stops-false-advertising-phony-
reviews-online-trampoline (Accessed 8 January
2018).
A new classification strategy for trust mechanisms
The growing ubiquity of online platforms and the
diverse types of the trust problems they face
has brought about the development of many
different types of trust mechanisms in the
market. There have been few attempts to
catalog, in a structured way, the many design
features of trust mechanisms.
15
However, we
argue that in order to appreciate the many
implications of trust mechanisms for regulation,
a better understanding of the universe of trust
mechanisms is vital.
Our approach to classification
Two key methodologies influenced our
approach to a classification scheme for trust
mechanisms: a theoretical analysis based on
mechanism design and field research.
Mechanism design approach
Mechanism design is a field of economics,
inspired by engineering, that studies the design
of protocols in systems to incentivize rational
agents to act in a desired way. Informally, a
mechanism design problem in economics is
15
An early example is Dellarocas, Chrysanthos.
"The digitization of word of mouth: Promise and
challenges of online feedback mechanisms."
Management science 49, no. 10 (2003): 1407-1424.,
but development in online markets has rendered this
analysis insufficient.
to find the “rules of the game”
16
(i.e. possible
actions of players and a way to aggregate these
actions) to bring about some outcome in a
situation where agents have private information.
Trust mechanisms can be thought of as a
solution to the mechanism design problem for
online platforms, in that they constitute a set of
possible actions by platform users and a way
for the platform to aggregate these messages
and use them to bring about exchange. The
trust mechanism is illustrated in the classic
Mount-Reiter diagram
17
from mechanism design
theory in Figure 3 below.
The mechanism design approach to analyzing
trust mechanisms suggests that there are three
key components to the design decisions of
online platforms:
16
Parkes, David. “Iterative Combinatorial Auctions:
Achieving Economic and Computational Efficiency.”
Ph.D. dissertation, 2001. University of Pennsylvania.
17
Mount, Kenneth, and Stanley Reiter. "The
informational size of message spaces." Journal of
Economic Theory 8, no. 2 (1974): 161-192.
Key takeaways
We propose a new classification scheme for trust mechanisms, based on three categories:
participants, content and function.
We apply this classification scheme to five industries and twenty-nine platforms, with detailed
results in Appendix A.
This exercise illuminates several broader results about trust mechanisms which are relevant
for regulators, including
o Online platforms vary in levels of transparency about their trust mechanisms;
o Characteristics of trust mechanisms vary more between industry than within industry,
depending mostly on the nature of the trust problem; and
o Within-industry variation in trust mechanism design may be driven by competitive
forces.
1. Participants – who participates in the
platform’s trust mechanism?
This determines the ‘type space’ of the
mechanism design problem – the types of
consumers and private information that may
be used as part of the platform’s trust
mechanism.
2. Content – what information do market
participants provide to the platform in
the trust mechanism?
This determines the ‘action space’ for the
trust mechanism – what information
participants must provide as part of the
mechanism design and, therefore, what
information may be used by the platform to
determine outcomes for market participants.
3. Function – how is the informational
content of the trust mechanism used by
the online platform?
This determines the ‘outcome function’ for
the trust mechanism – the way in which
user’s actions are aggregated and then
used to affect the actual market outcomes
on the platform.
Field research
We used field research of existing trust
mechanisms in order to refine this broad
categorization of trust mechanisms based on
economic theory. We analyzed five industries –
ecommerce, ride-sharing, accommodation
services, online advertising platforms and
freelance labor hire – totaling twenty-nine trust
mechanisms to determine the specific ways in
which the participants, informational content
and function of trust mechanisms varied
between online platforms.
Detailed analysis of the five industries is
contained in Appendix A, but the key result of
this work is the classification schema itself in
Figure 4.
Error! Reference source not found.
Figure 3: Mount-Reiter diagram for trust mechanisms
Figure 4: Classification schema for trust mechanisms
Lessons from classification
Our field research on the characteristics of
online trust mechanisms also uncovered a
number of broader results which we believe
should inform regulation in platform markets.
The first important remark is that online
platforms differ in the level of transparency of
the features of the trust mechanisms they use.
In most cases, platforms widely share
information concerning the ‘participants’ and
‘informational content’ characteristics of their
trust mechanisms, but are reticent to share
publicly or readily full details on the ‘function’
characteristics of their trust mechanism. For
example, Uber does not share publicly the
‘rating floor’ for their drivers – the minimum star
rating that drivers require to remain on the
system. Neither Yelp nor Airbnb share
information about how ratings and reviews are
used to prioritize search results in their
algorithm (but do note that these are used as
inputs to their algorithm).
There are exceptions to this rule. Some
businesses promote widely the trust
mechanisms employed on their platform,
sharing substantial information about the way
their trust mechanisms operate. For example,
Toptal – a platform in the online labor industry –
define themselves on the basis of their trust
mechanism, claiming to offer “the Top 3% of
Freelance Talent”. Toptal’s trust mechanism
involves staff within the organization assessing
the qualifications and standards of work of
contractors on their platform, and directly
responding to feedback from the hiring parties.
The different levels of transparency on online
platforms served as a challenge for our
classification exercise – particularly in
determining ‘function’ characteristics –requiring
us to often rely on secondary evidence to
determine some characteristics of trust
18
We mean “strategic positioning” in the sense of
Besanko, David, David Dranove, Mark Shanley, and
mechanisms. This suggests that certain
characteristics of online platforms may be
unknown or opaque to members of the general
public and platform users. This may prevent
users from exercising informed choice on
platforms.
A second observation is that the design
features of trust mechanisms seem to vary
more significantly between industries (rather
than within industries). This confirms a premise
of the mechanism design approach to trust
mechanisms which is that the mechanism
depends fundamentally on the nature of the
trust problem being addressed. Different
industries tend to face different asymmetries of
information and thus different mechanism
design problems. Firms within a single industry
are more likely to face similar trust problems to
overcome.
Table 1 contains the key trust problems in the
industries analyzed and the common features of
trust mechanisms which may be seen as a
response to the trust problem encountered by
firms in that industry. Some trust problems (for
example, safety of market participants) may be
common to all online marketplaces, but Table 1
seeks to highlight the core trust problems that
are distinctive to these industries.
Although between-industry variation accounts
for many of the differences between trust
mechanisms, within certain industries, we have
observed that some heterogeneity may be
explained by competition between firms. In this
way, trust mechanisms may act, in some cases,
as a medium of market segmentation or
specialization by firms. We observe that firms
strategically positioning
18
themselves on the
higher end of the market often impose stricter
trust mechanisms on their users. By ‘stricter’,
Scott Schaefer. Economics of strategy. John Wiley &
Sons, 2009.
we mean that the level of burden associated
with the trust mechanism for the user or any
threshold used by the platform to qualify users
for certain benefits is higher.
For example, in the online retail market, there
are real and considerable differences in the
burden associated with the trust mechanisms
used between platforms. At one end, Craigslist
has one of the least burdensome trust
mechanisms in use of any platform we
analyzed, consisting only of the ability of the
platform to block users if severe complaints are
received or scam activity is suspected. On the
other end of the market, Amazon has a complex
trust mechanism including customer reviews of
sellers, ratings, meta-ratings of review
helpfulness, surveillance activities conducted by
Amazon and tiered user benefits on the basis of
the trust mechanism. eBay’s trust mechanism
shares some of these features, but not all of
19
“Marketplace Seller Fees Example | Where to sell
online,” 2017,
http://www.wheretosellonline.com/seller-fees-
example/. (Accessed 12 February 2018)
them. We argue that these differences may be
explained by the strategic positioning of these
firms within the online retailing industry.
Craigslist, with a less strict trust mechanism,
offers free postings for most users. In
comparison, on a $30 sale, Amazon
charges fees averaging 18.3% to their sellers,
where eBay’s fees average 12%
19
.
Within a platform, differences in trust
mechanisms may also arise as the result of
product segmentation. For example, Airbnb’s
premium product, Airbnb Plus, requires listings
to have a minimum rating requirements and to
undergo an inspection by an Airbnb
employee
20
. This is on top of the usual ratings-
based trust mechanism for Airbnb, which is
discussed in Appendix A.2.
A similar pattern emerges in several of the other
industries we analyzed. This result can be
illustrated graphically as in Figure 5.These
20
“Airbnb Plus Host Requirements,” Airbnb, 2018,
https://www.airbnb.com/plus/host/requirements/.
(Accessed 25 February 2018)
Table 1: Trust problem and design features of trust mechanisms by industry analyzed
Industry
Key trust problem
Common feature
Ride-sharing
Safety of passengers and drivers, and the quality of experience for
passengers
Two-sided reviews
Online retail
Ex-ante identification of seller and product/service quality prior to a
financial transaction.
Public ratings and
feedback
Casual labor
Ensuring freelancers have the capabilities to complete job in the
time and of the quality expected by client.
Skills tests and
endorsements
Accommodation
Identifying quality and ensuring personal and property safety for
travelers and hosts.
Double-blind reviews
Online
advertising
Ensuring advertisements are presented to appropriate audiences
which will drive engagement and revenue for advertisers
Granular, real-time
performance metrics
observations have a number of implications for
regulators. Firstly, a lack of information about
trust mechanisms may act as a barrier, both for
consumers to make informed choices and for
regulators to determine appropriate regulatory
responses. Secondly, a one-size-fits-all
approach would be an inappropriate regulatory
response to the challenges associated with trust
mechanisms, given the considerable variation in
trust mechanisms between firms in different
industries. Finally, regulating trust mechanisms
may have implications on competitive dynamics
and industrial organization in industries with
platform firms.
Figure 5: Relationship between trust mechanism strictness and positioning of firm within industry or
market segment
Benefits and costs associated with trust mechanisms
The benefits of trust mechanisms
The economic and social benefits arising from
trust mechanisms fall into at least three
categories:
1. Replacing existing laws and regulations;
2. Growing markets and increasing economic
welfare; and
3. Enabling superior targeting of government
spending.
Replacing existing laws and regulations
Trust mechanisms may substitute for a number
of consumer protection laws and public safety
regulations.
21
Since laws and regulations are
costly to enforce, one possible benefit of trust
mechanisms is to reduce or remove these
costs. By replacing regulations with their trust
21
See for example Lao et al., “Sharing Economy,” p.
60, 65; Martens, “Online Platforms,” p. 60; Cohen,
Molly, and Arun Sundararajan. "Self-regulation and
innovation in the peer-to-peer sharing economy." U.
Chi. L. Rev. Dialogue 82 (2015): 116; and Koopman,
Christopher, Matthew Mitchell, and Adam Thierer.
mechanisms in this way, online platforms would
exist in a regime of self-regulation.
However, it is difficult to measure the size and
scope of these benefits. This is partly because
trust mechanisms often interact with many
regulations. Consider for example the case of
ridesharing. As was discussed earlier in this
report, the trust mechanisms employed by
online platforms such as Uber and Lyft share
many similarities. However, the regulations they
interact with are numerous. For example, the
regulations that apply to medallion taxicab
services in New York City alone govern factors
including but not limited to maintenance and
record-keeping requirements, vehicle inspection
schedules, public accommodation laws, and
licensing fees.
22
Given this, the question of how
ridesharing platforms’ driver verification
"The sharing economy and consumer protection
regulation: The case for policy change." J. Bus.
Entrepreneurship & L. 8 (2014): 529.
22
Jonas, Alexandra. "Share and share dislike: The
rise of Uber and Airbnb and how New York City
should play nice." JL & Policy 24 (2015): 205, p. 213.
Key takeaways
Trust mechanisms have the potential to bring significant economic and social benefits to consumers
and businesses in the United States. However, they also bear inherent costs and risks for regulators
to manage.
Occupational licensing is a type of regulation where the benefits from trust mechanisms may be
broadly estimated. We estimate that trust mechanisms may reduce regulatory burden in
occupational licensing by potentially $794m-$1.94bn, depending on the scenario and time horizon.
Trust mechanisms may also help grow markets and increase economic welfare, and also have been
used by governments to enable the superior targeting of government spending.
However, possible costs of trust mechanisms include new forms of discrimination on online
platforms, the potential for welfare-reducing manipulation of ratings and certain design flaws that
may open up new forms of harm on online marketplaces.
systems, two-way ratings, and other trust
mechanisms interact with each of these
individual regulations is not one that can (or
should) be easily answered in general even for
a single industry within a single city.
An additional problem with identifying the
benefits arising from self-regulation is that there
are often inherent costs associated with
removing or supplanting existing
regulations. These may attenuate the
estimated benefit of trust mechanisms, make
these difficult to isolate, or make the prospect of
self-regulation practically or politically difficult to
realize.
A concrete illustration of this (again from New
York City) is the 2010 amendment to the city’s
Multiple Dwelling Law and its interaction with
short-term accommodation agreements
facilitated by Airbnb. The Multiple Dwelling Law
defines “illegal hotel activity” as “[w]hen
permanent residential apartments in buildings
with three units or more are rented out for less
than thirty days to transient visitors instead of
residents.”
23
In response to efforts by Airbnb to
overturn this law, New York State Senator Liz
Krueger highlighted one of the law’s purposes
to protect the safety of the city residents by
restricting strangers’ access to residential
properties, noting that
24
“Illegal hotel operations mean, at a
minimum, a regular stream of relatively
un-vetted strangers coming into and out
of residential buildings. That can create
23
Ibid., p 2018.
24
Krueger, Liz. “Answers for New Yorkers
Concerned or Confused About the Illegal Hotel Law |
NY State Senate,” 27 May, 2014,
https://www.nysenate.gov/newsroom/articles/liz-
krueger/answers-new-yorkers-concerned-or-
confused-about-illegal-hotel-law. (Accessed 15
February 2018).
25
As discussed below, occupational licensing may
also have other objectives that are not achieved by
trust mechanisms, e.g. political economy, and the
restriction of supply to increase prices.
serious quality-of-life problems and
safety for neighbors, at a minimum –
sleepless nights caused by overcrowded
neighboring apartments packed with
loud tourists, for example. But it can get
far worse. My office has heard of
buildings burglarized and neighbors
assaulted by strangers who might never
have had access to get inside, were it
not for illegal hotel activity.”
Occupational licensing and the benefits of trust
mechanisms
Occupational licensing is a key area in which
trust mechanisms may enable reduced
regulation. Occupational licensing and trust
mechanisms on online platforms share a
common goal of reducing information
asymmetries and establishing a minimum level
of safety and quality for consumers in
marketplaces
25
. There may thus be some scope
for substitution between the two in the modern
economy.
Occupational licensing is costly and generates
significant regulatory burden in the US
economy, with several researchers noting that
this form of regulation “has become one of the
most significant factors affecting labor markets
in the United States.”
26
In 2016, approximately a
quarter of US workers held an occupational
license,
27
(a proportion that has grown roughly
five-fold since the 1950s in large part due to
new regulations on previously unlicensed
occupations).
28
As well as its prevalence,
26
Hall et al. “Occupational Licensing of Uber
Drivers.” Unpublished working paper presented at
the ASSA 2018 Preliminary Program, 6 January,
2018.
27
Furman, J. and Laura Giuliano. “New Data Show
that Roughly One-Quarter of U.S. Workers Hold an
Occupational License,” 17 June, 2016,
https://obamawhitehouse.archives.gov/blog/2016/06/
17/new-data-show-roughly-one-quarter-us-workers-
hold-occupational-license. (Accessed 4 December
2017).
28
U.S. Department of the Treasury Office of
Economic Policy, Council of Economic Advisors, and
occupational licensing has been associated with
regulatory through-channels including increased
costs for consumers, mobility restrictions, and
lower wages for excluded workers.
29
One study
estimated this burden as comprising an annual
cost to consumers of $203 billion and 2.8 million
fewer jobs.
30
Trust mechanisms, where functioning
successfully, may represent a cheaper and
more effective way to ensure the quality and
safety of services in certain marketplaces.
However, there seem to be other relevant
factors that determine whether this form of self-
regulation is appropriate in a given marketplace,
namely—
Does the trust mechanism collect and
receive sufficient information about
operators? In order for the trust
mechanism to replace the safety and quality
assurance function of occupational
licensing, sufficient feedback and
information about past performance of
operators is necessary. This suggests that
regular contact with others is a bare
minimum in order for trust mechanisms to
operate effectively.
Does a platform effectively aggregate
and display information about operators
to allow consumers to make informed
decisions about operators on the
platform? In order for the quality and safety
of users on platforms to be assured, users
should have some confidence that the
platform is effectively and honestly
collecting, aggregating and displaying
information about operators on the platform.
What are the consequences of a failure
to filter out substandard operators on a
the Department of Labor. “Occupational Licensing: A
Framework for Policymakers,” July 2015, p. 17.
29
Ibid.
30
Kleiner, Morris M., Alan B. Krueger, and
Alexandre Mas. "A Proposal to Encourage States to
platform? We may be less inclined to allow
self-regulation by platforms in the case
where operators have significant
responsibilities for the health and safety of
others, or where the consequences of error
may be widespread or disastrous. As an
example, we might be less inclined to rely
on customer reviews when choosing a brain
surgeon, where the consequences of failure
could be fatal, as compared to a florist,
where the consequences of failure may be
more limited.
On the basis of this, we have sought to identify
which currently licensed industries have the
highest likelihood of having trust mechanisms
effectively ensuring quality and safety in the
future. The result is a ‘regulatory substitution
index’, which also helps us estimate the dollar
value of the regulatory burden that may be
reduced in this context.
We attain a quantitative estimate of the
regulatory burden online mechanisms’ trust
mechanisms may alleviate by utilizing recent
data, existing research on the burden of
licensing, and the findings of our primary
research. To our knowledge, no existing
exercise to quantify the potential for trust
mechanisms to substitute for regulations exists.
This means that the approach adopted in this
report should be interpreted as a proof-of-
concept calculation that lays the foundation for
further work.
Our estimation methodology comprises four
broad steps. First, we build a new
occupation-level dataset of job
characteristics, regulatory burden, and the
presence of trust mechanisms. We do this by
combining data from three sources:
Rationalize Occupational Licensing
Practices." Paper submitted to the Brookings
Institution, Washington, DC, April (2011), p. 3.
Time and fee burden estimates associated
with 102 occupations compiled in late 2017
by the Institute of Justice in the second
edition of their ‘License to Work’ report;
31
Job characteristics from the US
Government’s Occupational Information
Network (O*NET) database;
32
and
Hand-coded data on whether an occupation
possesses an online platform.
Our rationale for using and combining data from
the O*NET database and second License to
Work report is that they are among the most
comprehensive, recent, and high quality data
relating to occupations and licensing burdens in
the United States. O*NET comprises granular
and regularly updated activity-level survey data
for over 1,000 occupations developed under the
sponsorship of the U.S. Department of
Labor/Employment and Training
Administration.
33
License to Work remains the
only nation-wide study on the burdens of
occupational licensing in the United States and
is widely cited by academics and policymakers.
For example, the FTC’s former Acting Chair,
Maureen K Olhausen, has even noted that “I
have called for meaningful occupational
licensing reform, often citing the Institute for
Justice’s original May 2012 License to Work
report to back up my arguments.”
34
To illustrate the nature of these data,
descriptive statistics for two frequently licensed
occupations – taxis drivers and school bus
drivers – are shown below in Table 2.
Second, we assess and rank the
‘substitution potential’ for each occupation
on the basis of six dimensions from our
data, summarized below at Table 3. The
31
Carpenter II, D.M., and Lisa Knepper. “License to
Work: A National Study of Burdens from
Occupational Licensing (2
nd
Edition)” Institute for
Justice Report, November 2017.
32
“O*NET Resource Center – Overview.” O*NET
Resource Center,
https://www.onetcenter.org/overview.html.
(Accessed 16 March 2018).
choice of these dimensions and their
comparison benchmarks reflect our primary
research findings (including stakeholder
interviews) and the data in the previous step (as
also noted in Table 3).
The creation of these variables allows for a
category-by-category comparison of how
amenable to substitution by trust mechanisms
the licensed occupations in our dataset are. The
best and worst professions in each category are
available in Appendix B of this report.
However, we go further than this, and
tentatively attempt an aggregation of these
variables. That is, the third step in our
methodology is to combine these six
dimensions into a “regulatory substitution
index” with values for each occupation that
that range from 1.00 (most substitutable) to 0
(least substitutable). The full details of how we
construct this index is described in Appendix C
(which includes links to our data and STATA
programs for replication).
33
“O*NET Resource Center – Overview.” O*NET
Resource Center,
https://www.onetcenter.org/overview.html.
(Accessed 16 March 2018).
34
Ohlhausen, Maureen K. “Foreword – Institute for
Justice,” November 2017, http://ij.org/report/license-
work-2/report/foreword/. (Accessed 20 March 2018).
Table 2: Selected data for taxi and school bus drivers
Variable
Taxi drivers
School bus drivers
No. of states licensed
(2017)
16
51
Average fees ($) (2017)
$47
$112
Average calendar days
lost to training (2017)
148
300
Mean exams
0
6
O*Net Job Zone (2016)
Job Zone One: Little or No
Preparation Needed
Job Zone Two: Some
Preparation Needed
Number employed (2016)
305,000
508,000
Performs or Works
Directly with the Public
(2016)
Yes
No
Responsible for others’
health and safety (2016)
35% responded “Very high
responsibility.”
40% responded “Very high
responsibility.”
Consequence of error
(2016)
Not reported
56% responded “Extremely
serious.”
Online platform exists?
Yes
No
Source: License to Work v2. (2017), O*NET (2016), authors’ primary research; image source: Getty Images.
Table 3: Variables that are relevant to trust mechanisms potentially substituting for occupational
licensing within a given occupation
Dimension/variable
Description and range
Interpretation
1. The occupation involves
working directly with the
public
(a survey question
O*NET)
Two categories (“Yes” or
“No”
)
Licensed occupations that work directly
with the public
(the variable = “Yes
”) are
more likely to be substitutable for trust
mechanisms.
2. The occupation’s
consequence of error
(a
survey question with a
numerical response in
O*NET)
Five categories ranging from
“
Not serious at all” to
“
Extremely serious” and
proportional responses,
including a weighted score
out of five where 5 equals the
highest possible
consequence of error.
Licensed occupations with less serious
consequences of error are more likely to
be substitutable for trust mechanisms.
3. The occupation’s
responsibility for the health
and safety of others
(a
survey question with a
numerical response in
O*NET)
Five categories ranging from
“
No responsibility” to
“
Extremely high
responsibility
” and
proportional responses,
including a weighted score
out of five where 5 equals the
highest level of responsibility.
Licensed occupations with less
responsibility for others’ health and
safety
are more likely to be substitutable
for trust mechanisms.
4. The level of the
occupation’s
contact with
others
(a survey question
with a numerical response in
O*NET)
Five categories ranging from
“
No contact with others” to
“
Constant contact with
others
” and proportional
responses, including a
weighted score out of five
where 5 equals the highest
possible level of contact with
others.
Licensed occupations with more frequent
contact with others are more likely to be
substitutable for trust mechanisms. This is
because it could be difficult to collect
feedback data for a mechanism if other
human beings are not present for the
whole time the occupation’s activity
occurs.
5. The presence of an online
platform
where the
occupation is offered on or
supplied through (a hand-
coded variable)
Two categories (“Yes” or
“
No”) with the name of the
platform (e.g. “
Uber”)
Licensed occupations with online
platforms (the variable = “
Yes”) are more
likely to be substitutable for trust
mechanisms.
6. Whether the online platform
in category 5 is an
on-
demand platform
(a hand-
coded variable)
Two categories (“Yes” or
“
No)
Example of “
Yes”: Uber
Example of “
No”: Yelp
Licensed occupations with on-demand
platforms (the variable = “
Yes”) are more
likely to be substitutable for trust
mechanisms. This variable comes from a
discussion we had with Professor Morris
Kleiner, who noted that occupations with
on-demand platforms (those with a labor
spot market) may possess characteristics
attractive for trust mechanisms to
substitute for licensing, including the
ability to collect information rapidly and
regularly.
One notable feature of our index is that there
are two conditions that we consider to be
minimum qualifying standards for an occupation
to be realistically able to substitute for an online
platform’s trust mechanism. These are:
1. The occupation involves working directly
with the public (the rationale for this is the
same as that noted in Table 3); and
2. The occupation cannot have both:
a. An “Extremely high” or “Very
high” responsibility for the safety of
others; and
b. An “Extremely high” or “Very high”
consequence of error.
This excludes the possibility of serious and
significant consequences to users, as
discussed in the brain surgeon example
above.
Occupations that fail to satisfy these criteria
have an index value set to 0.
A summary of our index for each occupation
with non-zero index values is in Figure 6. The
top ten occupations according to our index are
shown below in Table 4.
Finally, the potential reduction in regulatory
burden is computed on a dollar scale using
our dataset using the formula
In words, our burden metric is the sum of the
total licensing fees of workers across all
occupations and all states, which can be
interpreted as a lower bound on the point-in-
time estimate of the economy-wide lifetime
burden of licensing fees. This figure is an
underestimate of the lifetime burden, as many
licenses are not one-off and require renewal.
Table 4: The ten occupations with the highest “regulatory substitution index”
Occupation
Regulatory substitution index value (100
= perfect substitute)
Online platform exists?
(with example)
1. Barber
93.05
Yes (Shortcut)
2. Bartender
92.15
Yes (Saucey)
3. Shampooer
91.07
Yes (GlamSquad)
4. Massage
Therapist
85.43
Yes (StyleBee)
5. Cosmetologist
84.96
Yes (Vensette)
6. Manicurist
84.61
Yes (GlamSquad)
7. Truck Driver,
Other
84.38
Yes (Uber Freight)
8. Taxi
Driver/Chauffer
83.56
Yes (Uber)
9. Interior Designer
80.99
Yes (Homepolish)
10. Florist
80.70
Yes (BloomNation)
*See Appendix C for the full list.
Figure 6: Summary of substitution index values for occupations with non-zero index values
Data for average licensing fees for a given
occupation-state
35
pair come from License to
Work, and data on the number of workers
employed in an a given occupation-state pair
come from The US Bureau of Labor Statistics’
Occupational Employment Statistics.
36
For a
given set of occupations, the number arising
from this formula can be interpreted as the
maximum saving in total fees that could arise
from workers not having to register for a license
as a result of being able to use an online
platform (and its trust mechanism instead)
instead.
We calculate the potential reduction regulatory
burden for three sets of occupations:
1. Occupations that already have on-
demand online platforms at the time of
writing. We consider this to be a ‘lower-
end’ estimate of the maximum ‘short-run’
reduction in burden arising from the trust
mechanisms that exist today. This number
can be thought of as answering the
35
We thank a staff member from the FTC’s Bureau
of Economics who kindly offered comments on an
earlier version of this chapter and helpfully pointed
out that state-level data was available and would
result in less bias than an earlier national-level
calculation we attempted.
question, “What would be savings be for
workers if all taxi drivers, and other
occupations in similar circumstances,
avoided licensing fees by switching to Uber
(and other similar platforms) today?”
2. Occupations that have non-zero values
in our regulatory substitution index. We
interpret this as our ‘higher-end’ estimate of
the maximum ‘long-run’ reduction in
regulatory burden that would arise if all of
the occupations that could potentially
substitute licensing for trust mechanisms did
so.
3. Occupations in top 50% of index scores
for those occupations that have non-zero
values in our regulatory substitution
index. This estimate is a “medium-run”
estimate between the first two categories.
The results of these calculations are shown
below at Table 5.
36
U.S. Bureau of Labor Statistics. “Occupational
Employment Statistics – May 2016 National
Occupational Employment and Wage Estimates
United States.” Data download on XLS file from
https://www.bls.gov/oes/current/oes_nat.htm#00-
0000. (Accessed 24 March 2018).
Table 5: Burden calculations for our three scenarios
Occupation set/scenario
Maximum potential reduction in regulatory
burden ($)
(Short-run/Low) Occupations that already
have on-demand online platforms at the
time of writing
$794m
(Medium-run/Medium) Occupations in top
50% of index scores for those occupations
that have non-zero values in our
regulatory substitution index
$1.26bn
(Long-run/High) Occupations that have
non-zero values in our regulatory
substitution index
$1.94bn
On just the basis of reduced fees, the potential
reduction in regulatory burden from online
platforms’ trust mechanisms is large and in
excess of a billion dollars in two of three
scenarios. Although this is noticeably smaller
than the $200bn back-of-the-envelope
calculation in Kleiner, Kruger, and Mass,
37
our
estimate should be interpreted in light of the
following facts:
Our calculations do not take into
account the opportunity cost of days
lost. Doing so markedly increases the
burden estimate. The data allow for a rough
back-of-the-envelope calculation of this
because the License to Work data includes
a ‘days lost’ variable for each occupation.
Assuming that half of the lost days have an
opportunity cost of the 2017 median
daily wage ($123) results in an aggregate
burden for our ‘short-run’ calculation of
around $61bn ((0.992m hours*0.5 of
these*123 dollars in fees on average)).
Our estimate fails to take into account
other benefits associated with trust
mechanisms. These include market growth
and the superior targeting of government
spending and are discussed below.
Not all occupations may have licensing
replaced by trust mechanisms, as
reflected in the occupations that have their
index values set to zero.
Further details (including the underlying data)
are available at Appendix E.
37
Kleiner, Morris M., Alan B. Krueger, and
Alexandre Mas. "Occupational Licensing Practices,”
p. 3.
Market size and economic welfare
The core purpose of trust mechanisms, if
effectively designed and implemented, is that
they allow businesses to overcome the trust
problem of online exchange and enable
transactions to occur over the platform. A key
question is whether these transactions grow the
overall market, both in terms of transaction
volume and value, and how this impacts
consumer and producer surplus.
In terms of market volume, there are two
possibilities:
1. Substitution of transactions: Transactions
on online platforms enabled by trust
mechanisms displace transactions that
would have occurred offline in businesses
using traditional business models.
2. Addition of transactions: Exchange on
online platforms enabled by trust
mechanisms is so convenient, inexpensive
or otherwise preferred over traditional
business activity that there is overall growth
in the market, brought about by new market
participants or increased volume of activity
by existing market participants.
For any given platform or industry, it is likely
that both explanations will hold true to some
extent. Thus, what determines whether online
platforms enabled by trust mechanisms
contribute to market growth is which of these
two effects dominate in a given industry subject
to disruption by online platforms.
There are several reasons to believe that the
additionality effect may dominate in many
industries, leading to an increase in overall
quantity of transactions, including
Online platforms tend to reduce search
costs for consumers and therefore increase
the likelihood that the marginal benefit of a
purchase exceeds the marginal cost and
thus increasing demand,
38
and
Online platforms seems to lower the cost of
bringing an asset to market,
39
increasing the
supply in an industry and lead to an
increase in total transactions.
Given that online platforms are likely to increase
demand, the overall impact on prices, consumer
surplus and producer surplus depends on the
relative impacts. In terms of prices, there is
some evidence, based on studies of
ecommerce websites, that online platforms tend
to reduce the price of goods and services
compared to offline markets.
40
This would also
suggest that the overall impact on consumer
surplus was positive (with the impact on
producer surplus ambiguous). However, recent
research suggests that the impact of platforms
on prices varies substantially between sectors,
retailers and geographic markets, with 72% of
goods in a recent sample having the same price
online as in brick-and-mortar retailers.
41
Despite mixed evidence of the impact of online
platforms on prices, there have been several
studies to suggest that the overall impact of
platform entry into markets is an enhancement
of welfare. One study found that peer-to-peer
38
Goldmanis, Maris, Ali Hortaçsu, Chad Syverson,
and Önsel Emre. "E‐commerce and the Market
Structure of Retail Industries." The Economic
Journal 120, no. 545 (2010): 651-682.
39
Horton, John J., and Richard J.
Zeckhauser. Owning, Using and Renting: Some
Simple Economics of the" Sharing Economy". No.
w22029. National Bureau of Economic Research,
2016.
40
Lieber, Ethan, and Chad Syverson. "Online versus
offline competition." The Oxford handbook of the
digital economy(2012): 189.
41
Cavallo, Alberto. "Are online and offline prices
similar? evidence from large multi-channel
markets for durable goods, enabled by trust
mechanisms, increase consumer surplus by
0.8% to 6.6%.
42
In the accommodation market,
recent research on the ten largest US cities by
penetration of Airbnb found that the entry of the
platform increased total welfare by $352 million
(around $70 per night booked on the
platform).
43
The entry of Uber and Lyft in New
York was found to have a welfare gain of 72
cents per dollar spent on the platform.
44
This
welfare gain seems to be concentrated among
consumers, with each dollar spent on Uber
found to have generated $1.60 in consumer
surplus in the four major US cities in 2015.
45
Overall, early evidence suggests that trust
mechanisms enable the growth of online
platforms, which has had substantial positive
welfare implications on many of the industries
disrupted. However, these welfare implications
appear to vary by industry, and thus we must be
careful in applying these results more broadly to
new industries being impacted by platform
entry.
Superior targeting of government spending
Aside from regulating trust mechanism use in
markets, government may also wish to make
active use of trust mechanisms to improve
public policy. There have been several
applications of government using the
retailers." American Economic Review 107, no. 1
(2017): 283-303.
42
Fraiberger, Samuel P., and Arun Sundararajan.
"Peer-to-peer rental markets in the sharing
economy." Forthcoming. (2015).
43
Farronato, Chiara, and Andrey Fradkin. The
welfare effects of peer entry in the accommodation
market: The case of airbnb. No. w24361. National
Bureau of Economic Research, 2018.
44
Lam, Chungsang Tom, and Meng Liu. "Demand
and Consumer Surplus in the On-demand Economy:
The Case of Ride Sharing." Working paper. (2017).
45
Cohen, Peter et al. Using big data to estimate
consumer surplus: The case of uber. No. w22627.
National Bureau of Economic Research, 2016.
informational content from trust mechanisms
(particularly reviews and ratings) to better target
government activities.
Ratings and textual content from Yelp has been
shown to have strong predictive power on the
hygiene inspection outcomes for restaurants.
46
A learning model using Yelp data was found to
predict severe hygiene offenders with 82%
accuracy, suggesting that inspection activity
and public disclosure policy may be improved
by mining public opinions from informational
content collected in the trust mechanisms of
online platforms.
Similarly, the Behavioural Insights Team in the
United Kingdom has used data collected in the
trust mechanism employed by a medical
bookings platform for the National Health
Services (NHS Choices) to identify with 95%
accuracy doctor’s surgeries that would fail
46
Kang, Jun Seok, Polina Kuznetsova, Michael
Luca, and Yejin Choi. "Where not to eat? Improving
public policy by predicting hygiene inspections using
online reviews." In Proceedings of the 2013
Conference on Empirical Methods in Natural
Language Processing, pp. 1443-1448. 2013.
47
The Behavioural Insights Team. “Using Data
Science in Policy,” 14 December, 2017, report
random health inspections.
47
Yelp data have
also been used with some success to predict
changes in local economic activity, like
business openings and closures, without
incurring the cost of traditional data collection
and surveying.
48
However, shortcomings of this
approach have also been identified, with the
data from online platforms found to be most
informational in denser, wealthier and more
educated areas.
49
New creative applications of data collected via
trust mechanisms are likely to be found, offering
governments more ways to target spending in
the future. This could either serve to create
savings for taxpayers or improve the quality of
existing government programs.
available at:
behaviouralinsights.co.uk/publications/using-data-
science-in-policy/.
48
Glaeser, Edward L., Hyunjin Kim, and Michael
Luca. Nowcasting the Local Economy: Using Yelp
Data to Measure Economic Activity. No. w24010.
National Bureau of Economic Research, 2017.
49
Ibid.
The harms of trust mechanisms
The rising prominence of trust mechanisms in online commerce has also brought about harms to
consumers and businesses in some markets. In the Background section above, we discussed some
specific, high-profile examples of problems that trust mechanisms have caused in the last twelve
months. In this section, we will discuss methodically the different forms of harm that trust mechanisms
may bring about and what challenges these harms present to regulators.
Strategic manipulability
As the name suggests, the goal of trust
mechanisms is to improve trust in online
marketplaces. This suggests that the integrity of
the data collected and displayed through trust
mechanisms may be of high concern to
effective market functioning. The strategic
incentives of various market participants may
pose a threat to the integrity of trust
mechanisms.
There are three key groups that may have an
incentive to manipulate trust mechanisms:
sellers of online goods and services; buyers of
goods and services; and platforms themselves.
We consider each in turn.
1. Sellers of goods and services in online
platforms
Vendors in online platforms would like to
present a more positive image of
themselves than other users might provide
(for example, higher ratings or more glowing
reviews). A 2009 study of reviews on
Amazon, iTunes and Vanno (a now defunct
company reputation website) found that
50
Kornish, Laura J. "Are user reviews systematically
manipulated? Evidence from the helpfulness
ratings." Leeds School of Business Working
Paper (2009).
51
Woolacott, Emma. “Amazon's Fake Review
Problem Is Now Worse Than Ever, Study Suggests.”
Forbes, 9 September, 2017,
https://www.forbes.com/sites/emmawoollacott/2017/
09/09/exclusive-amazons-fake-review-problem-is-
now-worse-than-ever/#429dfc497c0f. (Accessed 14
February 2018).
52
Mayzlin, Dina, Yaniv Dover, and Judith Chevalier.
"Promotional reviews: An empirical investigation of
between 20% and 47% of reviews on the
platforms showed evidence of manipulation,
with this proportion varying both by platform
and by product on a platform.
50
A recent
study suggests that Amazon’s fake review
problem has only worsened in recent
months, despite the platform taking steps to
ban vendors from incentivizing reviews on
the site.
51
However, there is evidence to
suggest that the incidence of vendor
manipulation of trust mechanisms seems to
be worse on platforms that allow anyone to
submit data (as opposed to those that allow
only customers to review).
52
On Yelp, firms
with a weak reputation or facing high levels
of competition have been found to be more
likely to commit review fraud.
53
Firms have
been known to resort to extreme measures
in some cases, with small businesses like
the one illustrated in Figure 7 allegedly
operating as rankings manipulation
services.
54
The mattress firm Casper
resorted to legal action against an online
review platform in 2017 due to what it
perceived to be unfair rankings and
online review manipulation." American Economic
Review 104, no. 8 (2014): 2421-55.
53
Luca, Michael, and Georgios Zervas. "Fake it till
you make it: Reputation, competition, and Yelp
review fraud." Management Science 62, no. 12
(2016): 3412-3427.
54
Tweedie, Steven. “This disturbing image of a
Chinese worker with close to 100 iPhones reveals
how App Store rankings can be manipulated.”
Business Insider, 11 February, 2015,
http://www.businessinsider.com/photo-shows-how-
fake-app-store-rankings-are-made-2015-2
(Accessed 25 March 2018).
ultimately paid for its acquisition.
55
One
hotel has threatened to fine guests who post
negative reviews of their stay online.
56
Where identified, vendor manipulation may
lead to countervailing responses by users
and corrective action by platforms.
2. Buyers of goods and services on
platforms
In online platforms where the reputation of
buyers is a concern, buyers may have an
incentive to misreport information in a trust
mechanism. One cause is the opportunity
for retaliatory feedback, estimated to
comprise around 1.2% of mutual feedback
data on eBay in 2013.
57
However, this may
underestimate the degree that retaliation
affects the integrity of trust mechanism, with
a laboratory experiment on Airbnb
suggesting the threat of retaliatory feedback
is a significant driver of upward bias in
55
McKay, Tom. “Mattress Startup Casper Sued a
Mattress Review Site, Then Paid for Its Acquisition.”
Gizmodo, 24 September, 2017,
https://gizmodo.com/mattress-startup-casper-sued-
a-mattress-review-site-th-1818703265. (Accessed14
February 2018.)
56
Alter, Charlotte. 2014. ‘”Historic’ inn charges $500
per negative online review.” Time, 4 August, 2014,
http://time.com/3079343/union-street-guest-house-
negative-review/ (Accessed 14 February 2018).
reviews.
58
Because social interaction often
occurs as a side-product to the main
transaction on online platforms, socially-
induced reciprocity may also affect the
information provided to trust mechanisms.
Together, the threat of retaliatory feedback
and socially-induced reciprocity effects are
estimated to upwardly bias around 15% of
reviews on Airbnb.
3. The platform itself
Online platforms may also face incentives to
manipulate their own trust mechanisms.
Yelp has been accused of giving
preferential treatment in their algorithm to
businesses who advertise with them,
including hiding unflattering reviews and
altering star ratings. Although an empirical
study of Yelp reviews found no evidence to
suggest advertisers were treated differently
to non-advertisers on the platform,
59
these
57
Bolton, Gary, Ben Greiner, and Axel Ockenfels.
"Engineering trust: reciprocity in the production of
reputation information." Management science 59, no.
2 (2013): 265-285.
58
Fradkin, Andrey. "Search, matching, and the role
of digital marketplace design in enabling trade:
Evidence from airbnb." Working paper. (2017).
59
Luca, Michael. "Reviews, reputation, and revenue:
The case of Yelp. com." (Working Paper) (2016).
Figure 7: Image of an alleged app store ranking manipulation service
Image Source: Weibo
accusations do demonstrate the theoretical
possibility of an incentive for platforms to
bias trust mechanisms in certain
circumstances. Competition between
platforms may also generate a similar
incentive – for example, if Platform A seeks
to claim that their average service level is
better than Platform B’s then Platform A
may seek to inflate their overall ratings
artificially. Because there is usually
substantial proprietary data underlying trust
mechanisms, it will be difficult to tell in many
circumstances whether any such tactics are
being employed by firms. On the other
hand, competition between firms for users
may mediate this effect.
These three sources of strategic manipulation in
trust mechanisms are likely to lead to diminution
of informed choice and may therefore pose a
threat to consumer welfare. However, there are
several mitigating arguments to consider,
including:
The possibility that consumers are aware of
possible strategic biases and factor these
into their purchasing decisions;
Reversion to the mean – where a large
amount of information is collected about
participants, individual instances of
manipulation are less likely to impact
aggregate reviews; and
Market forces which may offset the effect of
strategic manipulation – for instance,
unhappy buyers tricked by positive reviews
may feel a greater responsibility to leave
negative reviews in response.
60
60
Some evidence for this effect in online forums has
been presented in Dellarocas, Chrysanthos.
"Strategic manipulation of internet opinion forums:
Implications for consumers and firms." Management
science 52, no. 10 (2006): 1577-1593.
61
Smith, Aaron and Monica Anderson. “Online
Shopping and E-Commerce.” Pew Research Center,
19 December, 2016,
http://www.pewinternet.org/2016/12/19/online-
shopping-and-e-commerce/. (Accessed 20 February
2018).
Despite these arguments, ensuring that an
accurate information is provided to users at the
point of sale is likely to be in the public interest.
Around half of Americans who read online
reviews say they “generally give an accurate
picture of the true quality of the product” and
some 82% of Americans check online reviews
before purchasing a product for the first time.
61
Supporting informed choice by consumers may
require government intervention to lower the
risk and incidence of strategic manipulation.
Discriminatory behavior
Discrimination is a concern in all marketplaces,
and there is good reason to worry that new
technologies, like trust mechanisms, open up
new modes for discriminatory behavior. In a
recent study on Airbnb, guests with names
perceived to be African-American were found to
be 16% less likely
to be accepted by vendors than guests with
names perceived to be characteristically
white.
62
On Uber, the cancellation rate for
drivers with names perceived to be African-
American has been found to be around twice as
high as the cancellation rate for perceived
white-sounding names.
63
62
Edelman, Benjamin, Michael Luca, and Dan
Svirsky. "Racial discrimination in the sharing
economy: Evidence from a field
experiment." American Economic Journal: Applied
Economics9, no. 2 (2017): 1-22.
63
Ge, Yanbo, Christopher R. Knittel, Don
MacKenzie, and Stephen Zoepf. Racial and gender
discrimination in transportation network companies.
No. w22776. National Bureau of Economic
Research, 2016.
While these discriminatory outcomes are
concerning, it is not immediately clear that trust
mechanisms are the cause of them. For
instance, on Uber, which uses a matching
algorithm which is blind to gender, male drivers
have been found to earn 7% more than female
drivers, but this gap is driven by differential
levels of experience, preferences over
when/where to work and preferences for driving
speed.
64
Additionally, discriminatory behavior
has existed in marketplaces well before online
platforms.
However, there are several reasons to believe
that trust mechanisms may open up a new and
concerning medium of discrimination:
There is evidence of discrimination in
ratings on some online platforms. On the
casual labor site TaskRabbit, women have
been shown to receive 10% fewer reviews
64
Cook, Cody, Rebecca Diamond, Jonathan Hall,
John List, and Paul Oyer. The Gender Earnings Gap
in the Gig Economy: Evidence from over a Million
Rideshare Drivers. No. 00634. The Field
Experiments Website, 2018.
65
Hannák, Anikó, Claudia Wagner, David Garcia,
Alan Mislove, Markus Strohmaier, and Christo
Wilson. "Bias in Online Freelance Marketplaces:
Evidence from TaskRabbit and Fiverr." In CSCW,
pp. 1914-1933. 2017.
than men.
65
On Fiverr, black workers are
found to receive 32% fewer reviews than
white workers and have an average star
rating 9% lower. Combined with evidence
that ratings and review numbers both
contribute to higher earnings on platforms
where users initiate matchings,
66
this would
suggest worse outcomes for women and
black workers on these platforms.
Where trust mechanisms are used by
platforms to determine qualifications, bans
or matches for their users,
nondiscriminatory practices within
companies may become
discriminatory.
67
For example, Rosenblat
et al. write
“Uber's rating system may, thus, present a
facially neutral route for discrimination to
“creep in” to employment decisions.
Through a rating system, consumers can
66
See, for example, Luca, Michael. "Reviews,
reputation, and revenue: The case of Yelp. com."
(Working Paper) (2016).
67
Rosenblat, Alex, Karen EC Levy, Solon Barocas,
and Tim Hwang. "Discriminating Tastes: Uber's
Customer Ratings as Vehicles for Workplace
Discrimination." Policy & Internet 9, no. 3 (2017):
256-279.
Figure 8: Protest against alleged discriminatory behavior on Uber
Image Source: United for Equal Access NY
directly assert their preferences and biases
in ways that companies would be prohibited
from doing directly. In effect, companies
may perpetuate bias without being liable for
it, as the grounds for firing or “deactivating”
a particular driver may be derived from a
large corpus of systemically biased
consumer ratings.”
Platforms themselves may not be legally
liable for discriminatory outcomes in this
case, which may lower the incentive to
debias their platforms.
A number of platforms have moved to remove
or lower discriminatory behavior by redesigning
elements of their trust mechanism. For
instance, Airbnb has changed the way that
reviews and information are presented in an
attempt to lower discrimination on their
platform.
68
Still, regulators should be aware of
the potential for trust mechanisms to bring
about new forms of discriminatory behavior in
markets.
Imperfections in trust mechanisms
Several other imperfections in trust
mechanisms has been identified. Some of the
main ones include:
Selection bias – contribution to optional
trust mechanisms may be thought of as a
68
McGee, Chantel. “How Airbnb’s redesign aims to
combat discrimination on the service.” CNBC, 8
April, 2017,
https://www.cnbc.com/2017/04/07/airbnb-
experimenting-with-site-design-to-fight-
discrimination.html.(Accessed 14 February 2018.)
69
Resnick, Paul, and Richard Zeckhauser. "Trust
among strangers in Internet transactions: Empirical
analysis of eBay's reputation system." In The
Economics of the Internet and E-commerce, pp.
127-157. Emerald Group Publishing Limited, 2002.
public good problem,
69
and therefore may
be subject to freeriding. This is particularly
the case for optional trust mechanisms
which are burdensome for contributors. The
result may be that only very motivated
users, and in particular those with very good
or very bad experiences, may contribute to
the trust mechanism, leading to a bias
towards extremes. On the other hand,
contributions to trust mechanisms which are
exclusive to market participants are also
likely to be generally biased toward positive
experiences because the types of people
who contribute have demonstrated
themselves as people inclined to purchase
that kind of good or service. This often leads
to a J-shaped distribution of data, and has
been found to negatively impact product
demand, firm profit and consumer surplus in
such industries
70
.
Cold start – in many trust mechanisms, the
online platform has no information to display
about new users, which may put them at a
disadvantage compared to established
users on the platform. There is significant
evidence that users depend both on
informational content and the number of
reviews when making ecommerce
transactions.
71
For example, on Airbnb,
there is a significant positive association
between the number of months of
membership on the platform and the price
charged by vendors.
72
However, goods and
70
Hu, Nan, Paul A. Pavlou, and Jie Zhang. "On self-
selection biases in online product reviews." MIS
Quarterly 41, no. 2 (2017): 449-471.
71
Flanagin, Andrew J., Miriam J. Metzger, Rebekah
Pure, Alex Markov, and Ethan Hartsell. "Mitigating
risk in ecommerce transactions: perceptions of
information credibility and the role of user-generated
ratings in product quality and purchase
intention." Electronic Commerce Research 14, no. 1
(2014): 1-23.
72
Teubner, Timm, Florian Hawlitschek, and David
Dann. "PRICE DETERMINANTS ON AIRBNB: HOW
REPUTATION PAYS OFF IN THE SHARING
services with little data are found to have a
weaker relationship between rating and
other more objective measures of quality.
73
Final period problem – concern for future
reputation is less likely to be a motivator for
users who intend to leave a platform in the
near future. As a result, we may expect to
see a decline in service quality prior to
platform exit. For example, sellers leaving
eBay have 25% more negative reviews in
final week of trading than their long-term
averages.
74
Reputation inflation – over time, user
responses to trust mechanisms may change
for a variety of reasons. There is evidence
that the average rating on star rating
platforms increases over time, an effect
referred to as ‘reputation inflation’.
75
Figure
9 illustrates this effect on a number of online
platforms. The result may be that that
informativeness of the reputation
mechanism as a whole diminishes over
time. This may be due to pressure on users
to leave above average feedback out of fear
of retaliation.
These imperfections vary in magnitude
depending on the type of trust mechanism used
which makes the design choices of online
platforms particularly consequential for user
welfare
ECONOMY." Journal of Self-Governance &
Management Economics 5, no. 4 (2017).
73
De Langhe, Bart, Philip M. Fernbach, and Donald
R. Lichtenstein. "Navigating by the stars:
Investigating the actual and perceived validity of
online user ratings." Journal of Consumer
Research 42, no. 6 (2015): 817-833.
74
Cabral, Luis, and Ali Hortacsu. "The dynamics of
seller reputation: Evidence from eBay." The Journal
of Industrial Economics 58, no. 1 (2010): 54-78.
75
Filippas, Apostolos, John Horton, and Joseph M.
Golden. Reputation in the Long-Run. No. 6750.
CESifo Working Paper, 2017.
Figure 9: Reputation inflation in platform markets
Source: Filippas, Apostolos, John Horton, and Joseph M. Golden. Reputation in the Long-Run. No. 6750.
CESifo Working Paper, 2017.
Assessing the costs and benefits of trust mechanisms
As we have demonstrated, the benefits
associated with trust mechanisms enabling new
business models may be substantial.
Consumers, businesses and governments are
all likely to enjoy increased economic activity
and better quality goods and services enabled
by new technology. Despite this, there are new
harms to consumer welfare and competition
that are made possible by trust mechanisms,
and the costs of these may also be substantial.
In many cases, it is in the platform’s interest to
minimize the potential for harm caused by their
trust mechanisms. Market forces may
incentivize businesses to act against
discriminatory behavior, for example, or to
minimize some of the biases and imperfections
that may exist in their trust mechanisms. In
these cases, the argument for regulatory action
is weaker – collaboration with online platforms
may be the major role for government
regulators.
However, there are some cases where market
forces alone may be insufficient to force action
to reduce harms. Incentives for some platforms
to manipulate their trust mechanisms for
strategic or competitive gains seem particularly
concerning. Instances where platforms benefit
from reducing informed choice to consumers
may also present a case for intervention.
We believe the overarching goal for regulators
should be to minimize the potential costs
associated with trust mechanisms, while
maximizing the upside. In the following chapter,
we discuss steps regulators can take to achieve
this.
Recommendations
The broad implications of our findings for our report’s recommendations
Two stylized facts emerged from our analysis
that are consequential for all of the
recommendations that follow. These are:
1. There is (probably) no such thing as a
single “perfect” or “optimal” trust
mechanism that applies to a given
industry or an online platform within it.
Rather, as shown by our classification
schema and field research, a wide variety of
mechanisms exist within and across
industries. These each have their strengths
and weaknesses in solving the trust
problems a particular online platform faces.
This implies that specific mechanism design
rules or mandates may be extremely
difficult, if not impossible, to compose.
76
“Building trust with a new review system.” Airbnb
Blog, 10 July, 2014,
https://blog.atairbnb.com/building-trust-new-review-
system/. (Accessed 16 December, 2017).
2. Although all trust mechanisms have
imperfections, it is not always clear that
a market failure exists in regards to
correcting them. In many cases, platforms’
incentives are well-aligned to address
these. For example, to incentivize honest
and combat retaliatory reviews, Airbnb
amended its trust mechanism to include a
‘double-blind’ review policy in July 2014,
76
and many platforms have in-house teams
and procedures for removing fake reviews.
77
A number of existing laws, such as the
Consumer Review Fairness Act passed in
2016, also address these issues. Similarly,
other imperfections such as ‘reputation
inflation’
78
are recently documented
phenomenon. Regulatory interventions in
some areas may thus run the risk of being
77
See Appendix A for several examples of these.
78
Horton, John, and Joseph Golden. "Reputation
inflation: Evidence from an online labor
market." Working Paper, NYU (2015).
Key takeaways
Consumer-facing recommendations:
1. Regulators should investigate the development of an online database of information about the
characteristics and function of trust mechanisms employed by platforms.
2. Regulators should require businesses to make public information about the characteristics and
function of trust mechanisms employed on their platform.
Business-facing recommendations:
3. Regulators issue guidelines via its Business Center to businesses concerning how to minimize
potential harms caused by trust mechanisms in online platforms.
Recommendations for government relationships:
4. (In respect to occupational licensing) Regulators write to state and local authorities about
appropriate ways to minimize regulatory burden in the face of trust mechanisms.
5.(In respect to better targeting government policy) Regulators should investigate areas where
consumer protection activities could be better targeted using data from trust mechanisms.
premature or inferior to what platforms may
produce on their own.
Nevertheless, we believe there are a number of
valuable areas where regulators can and should
act to facilitate and encourage the healthy
development and use of trust mechanisms in
line with its mission. These are discussed
below.
Guidance for regulating trust mechanisms
We evaluated policy options related to trust
mechanisms using several criteria which are
relevant for consideration by business
regulators:
1. Likelihood to reduce potential for harm to
consumers on platforms – we seek
regulatory interventions that show real
promise of improving consumer outcomes
on platforms;
2. Ability to allow for and promote platform
competition – we seek policies that do not
constrict competition between platforms and
do not entrench unfair competitive
advantages;
3. Level of burden for platform – we seek to
minimize red tape and ensure that the
benefits of any policy proposals outweigh
the burden imposed on businesses; and
4. Support for economic growth and success
of platform markets – we seek policies that
support the economic benefits of trust
mechanisms.
We have also taken into account the three
elements of Moore’s strategic triangle – the
79
Moore, Mark Harrison. Creating public value:
Strategic management in government. Harvard
University Press, 1995.
authorizing environment, operational capacity,
and the creation of value.
79
This suggests a fifth
criteria:
5. Political and operational feasibility – we
seek policy changes that are implementable
– both in the political context of regulators
and taking into account their operational
capabilities.
We have developed our policy proposals in
response to the challenges identified in the
previous sections, and have informed them on
the basis of discussions with businesses and
policy-makers, and other field research. Our
recommendations fall into three categories:
those geared towards consumer welfare,
businesses and government. These, are
summarized and discussed below.
Consumer-facing recommendations
We see one of the key challenges facing
consumers as the lack of information that
consumers have access to regarding the
characteristics and function of trust
mechanisms in online platforms. As discussed
above, one of the key causes of this is a lack of
transparency on the part of online platforms,
and this may be an area in which the strategic
interests of platform businesses and consumer
welfare are aligned. Greater awareness and
transparency would help address the potential
harms which may not be in the platform’s own
interest to address. Our first recommendations
address this problem.
Addressing a lack of transparency – options
considered:
1. No change;
2. Regulators to develop and make
available for public consumption a
database of information about trust
mechanisms employed by platforms;
3. Regulators to require businesses to
publicize information about the trust
mechanisms employed on their
platforms, including their characteristics
and function.
We believe that the no change option is
insufficient to reduce the potential of harm to
consumers. With trust mechanisms playing an
increasingly influential role in the economy, a
continued lack of transparency regarding the
trust mechanisms enabling these transactions
may lead to irreversibly negative consequences
for the evolution of online marketplaces. We
also note that a continued lack of transparency
may also hinder competition of platforms on the
basis of trust mechanisms. We thus do not
recommend the no change option.
Recommendation 1
We do recommend that regulators investigate
the development of an online database of
information about the characteristics and
function of trust mechanisms employed by
platforms. We believe that consumers would
benefit from data on the characteristics of trust
mechanisms on platforms, including an
assessment of the potential to be manipulated
and information about complaints regarding the
platform’s trust mechanism. This would allow
consumers to make more informed choices in
online marketplaces. We also believe that
regulators are well-placed to provide this data
in an objective and fair way between platforms.
Appendix C contains an example of what data
would be displayed as part of this database.
Recommendation 2
We do recommend that regulators require
businesses to make public information about
the characteristics and function of trust
mechanisms employed on their platform. Even
under a model where regulators publicize data,
there would still need to be a requirement for
transparency, as some aspects (particularly
regarding the ownership and use of trust
mechanism data) are not currently available
from many platforms. We believe that the
minimum requirement for publication should be
the information included in the example
database entry in Appendix C.
Business-facing recommendations
As discussed above, there are a number of
potential harms of trust mechanisms which
businesses may already be motivated to
address. However, there may be a lack of
information in some businesses, particularly
startups, as to the best way to address these
problems (or of the hidden costs and
unintended consequences of leaving them
unaddressed). It is in the interests of both
consumers and businesses to improve the
function of trust mechanisms, and we believe
that regulators may effectively play an
informational and advocacy role in this process.
Improving trust mechanisms employed by
businesses – options considered:
1. No change.
2. Development of ISO standard(s) for trust
mechanisms.
3. Regulators to issue guidance to
businesses concerning how to minimize
potential harms, including how to
combat fake reviews and prevent
discrimination.
We do not recommend the no change option as
there may be a lack of information in some
platforms as to how to protect their platforms
from problems like fake reviews and
discriminatory behavior. In addition, providing
advice on these topics places the responsibility
of action squarely on the platform to improve
their trust mechanisms.
We also do not recommend the development of
an ISO standard for trust mechanisms. This is a
current proposal of the European Commission
with a committee at the International
Organization for Standardization currently
investigating standard principles and
requirements for the collection, moderation and
publication of online reviews, to be codified in
ISO/FDIS 20488.
80
We believe that an ISO
standard would not be sufficiently flexible to
cater the different kinds of trust mechanisms in
the market, as described in Chapter 2 above. In
particular, since trust mechanisms may act as a
medium for competition in some industries, the
requirement for a standardized trust mechanism
may inhibit competition in those markets. An
ISO standard is also likely to be overly
burdensome for many platforms given the
heterogeneity of trust problems that they may
seek to solve, which may apply even within an
industry (as discussed in Chapter 2 and shown
in our field research).
Recommendation 3
We recommend that regulators issue guidelines
to businesses concerning how to minimize
potential harms caused by trust mechanisms in
online platforms. Guidelines could be developed
on a number of topics, including how to
maximize informativeness, prevent reputation
inflation, and discriminatory behavior. Five
principles based upon our mechanism
classification scheme in Chapter 2 and the
associated platform profiles (Appendix A) that
could be included in such guidelines are:
Know what your [the platform’s]
design goals are.
There are many ways to generate
trust.
Different parties may have different
incentives.
80
International Organization for Standardization.
“ISO/FDIS 20488: Online consumer reviews --
Principles and requirements for their collection,
It’s not always a good idea to copy
other companies.
Let people speak honestly about your
products and their experience with
your company.
Recommendations for government
relationships
Some government practices may benefit from
change in the face of trust mechanisms. As
highlighted above, the economy stands to
benefit from a substantial decrease in
regulatory burden due to a lowered need for
occupational licensing in some industries.
Occupational licensing – options considered
1. No action.
2. Regulators to write to state and local
authorities about areas in which
occupational licensing laws could be
weakened in response to the
emergence of trust mechanisms.
Given the substantial economic benefits that
decreased regulatory burden of occupational
licensing may provide, we do not recommend
regulators take no action with respect to
occupational licensing.
Recommendation 4
We do recommend that regulators write to state
and local authorities about appropriate ways to
minimize regulatory burden in the face of trust
mechanisms. Because state and local
authorities often have jurisdiction for most forms
of occupational licensing, there are limited ways
regulators can unilaterally change the
regulatory burden in affected industries.
However, regulators could be a powerful voice
for change in some states and heighten the
moderation and publication.” Available at:
iso.org/standard/68193.html. (Accessed 15 February
2018.)
prominence of this issue in the minds of voters.
The political forces protecting occupational
licensing in some industries are significant, but
the economic impact is so substantial that
reform efforts are warranted.
Targeting government policy – options
considered
1. No action.
2. Investigate areas where regulators’
activities could be better targeted using
data from trust mechanisms.
There are few downsides to experimentation
with better targeted government activities, and
so we do not recommend the no action option
here.
Recommendation 5
We do recommend that the regulators should
investigate areas where their activities could be
better targeted using data from trust
mechanisms. For example, data from online
company review sites could be used to inform
surveillance activities targeted at identifying
scammers.
Bibliography
“Airbnb Plus Host Requirements,” Airbnb, 2018, ttps://www.airbnb.com/plus/host/requirements/.
(Accessed 25 February 2018)
Acemoglu, Daron, Ali Makhdoumi, Azarakhsh Malekian, and Asuman Ozdaglar. Fast and slow learning
from reviews. No. w24046. National Bureau of Economic Research, 2017.
Akerlof, George A. "The market for “lemons”: Quality uncertainty and the market mechanism."
In Uncertainty in Economics, pp. 235-251. 1978.
Alexa Internet. “Alexa Top 500 Global Sites.” https://www.alexa.com/topsites, 2018, (Accessed 4
January 2018).
Alter, Charlotte. 2014. ‘”Historic’ inn charges $500 per negative online review.” Time, 4 August, 2014,
http://time.com/3079343/union-street-guest-house-negative-review/ (Accessed 14 February 2018).
Bahlavan, Pahlavan. “Building trust and increasing transparency with MRC-accredited measurement.”
Google Agency Blog, 21 February, 2017, https://agency.googleblog.com/2017/02/building-trust-and-
increasing.html. (Accessed 16 March, 2018).
Barach, Moshe, Joseph M. Golden, and John J. Horton. "“Skin-in-the-Game” and Platform Credibility:
Evidence from Field Experiments." Working paper. (2017).
Besanko, David, David Dranove, Mark Shanley, and Scott Schaefer. Economics of strategy. John Wiley
& Sons, 2009.
Bolton, Gary, Ben Greiner, and Axel Ockenfels. "Engineering trust: reciprocity in the production of
reputation information." Management science 59, no. 2 (2013): 265-285.
“Building trust with a new review system.” Airbnb Blog, 10 July, 2014,
https://blog.atairbnb.com/building-trust-new-review-system/. (Accessed 16 December, 2017).
Cabral, Luis, and Ali Hortacsu. "The dynamics of seller reputation: Evidence from eBay." The Journal of
Industrial Economics 58, no. 1 (2010): 54-78.
Carpenter II, D.M., and Lisa Knepper. “License to Work: A National Study of Burdens from
Occupational Licensing (2
nd
Edition)” Institute for Justice Report, November 2017.
Cavallo, Alberto. "Are online and offline prices similar? evidence from large multi-channel
retailers." American Economic Review 107, no. 1 (2017): 283-303.
Cohen, Molly, and Arun Sundararajan. "Self-regulation and innovation in the peer-to-peer sharing
economy." U. Chi. L. Rev. Dialogue 82 (2015): 116.
Cohen, Peter, Robert Hahn, Jonathan Hall, Steven Levitt, and Robert Metcalfe. Using big data to
estimate consumer surplus: The case of uber. No. w22627. National Bureau of Economic Research,
2016.
Cook, Cody, Rebecca Diamond, Jonathan Hall, John List, and Paul Oyer. The Gender Earnings Gap in
the Gig Economy: Evidence from over a Million Rideshare Drivers. No. 00634. The Field Experiments
Website, 2018.
De Langhe, Bart, Philip M. Fernbach, and Donald R. Lichtenstein. "Navigating by the stars:
Investigating the actual and perceived validity of online user ratings." Journal of Consumer
Research 42, no. 6 (2015): 817-833.
Dellarocas, Chrysanthos. "Analyzing the economic efficiency of eBay-like online reputation reporting
mechanisms." In Proceedings of the 3rd ACM Conference on Electronic Commerce, pp. 171-179.
ACM, 2001.
Dellarocas, Chrysanthos. "Strategic manipulation of internet opinion forums: Implications for consumers
and firms." Management science 52, no. 10 (2006): 1577-1593.
Dellarocas, Chrysanthos. "The digitization of word of mouth: Promise and challenges of online
feedback mechanisms." Management science 49, no. 10 (2003): 1407-1424.
“Distance to Frontier – Doing Business.” The World Bank Group, 2017,
http://www.doingbusiness.org/data/distance-to-frontier. (Accessed 25 February 2018).
Edelman, Benjamin, Michael Luca, and Dan Svirsky. "Racial discrimination in the sharing economy:
Evidence from a field experiment." American Economic Journal: Applied Economics 9, no. 2 (2017): 1-
22.
European Commission. “Online Platforms.” Commission Staff Working Document accompanying the
report Communication on Online Platforms and the Digital Single Market {COM (2016) 288}, 2016.
Fair, Lesley. “Three FTC actions of interest to influencers.” FTC Business Blog, 7 September, 2017,
https://www.ftc.gov/news-events/blogs/business-blog/2017/09/three-ftc-actions-interest-influencers
(Accessed 6 January 2018).
Farronato, Chiara, and Andrey Fradkin. The welfare effects of peer entry in the accommodation market:
The case of airbnb. No. w24361. National Bureau of Economic Research, 2018.
Federal Trade Commission. “FTC Stops False Advertising, Phony Reviews by Online Trampoline
Sellers,” https://www.ftc.gov/news-events/press-releases/2017/05/ftc-stops-false-advertising-phony-
reviews-online-trampoline (Accessed 8 January 2018).
Federal Trade Commission. “About the FTC | Federal Trade Commission,” https://www.ftc.gov/about-
ftc. (Accessed 8 January 2018).
Filippas, Apostolos, John Horton, and Joseph M. Golden. Reputation in the Long-Run. No. 6750.
CESifo Working Paper, 2017.
Flanagin, Andrew J., Miriam J. Metzger, Rebekah Pure, Alex Markov, and Ethan Hartsell. "Mitigating
risk in ecommerce transactions: perceptions of information credibility and the role of user-generated
ratings in product quality and purchase intention." Electronic Commerce Research 14, no. 1 (2014): 1-
23.
Fradkin, Andrey. "Search, matching, and the role of digital marketplace design in enabling trade:
Evidence from airbnb." Working paper. (2017).
Fraiberger, Samuel P., and Arun Sundararajan. "Peer-to-peer rental markets in the sharing economy."
Forthcoming. (2015).
Furman, J. and Laura Giuliano. “New Data Show that Roughly One-Quarter of U.S. Workers Hold an
Occupational License,” 17 June, 2016, https://obamawhitehouse.archives.gov/blog/2016/06/17/new-
data-show-roughly-one-quarter-us-workers-hold-occupational-license. (Accessed 4 December 2017).
Ge, Yanbo, Christopher R. Knittel, Don MacKenzie, and Stephen Zoepf. Racial and gender
discrimination in transportation network companies. No. w22776. National Bureau of Economic
Research, 2016.
Glaeser, Edward L., Hyunjin Kim, and Michael Luca. Nowcasting the Local Economy: Using Yelp Data
to Measure Economic Activity. No. w24010. National Bureau of Economic Research, 2017.
Goldmanis, Maris, Ali Hortaçsu, Chad Syverson, and Önsel Emre. "Ecommerce and the Market
Structure of Retail Industries." The Economic Journal 120, no. 545 (2010): 651-682.
“Google and Facebook Tighten Grip on US Digital Ad Market: Duopoly to grab more than 60% of the
2017 digital ad spend.” eMarketer, 21 September, 2017, https://www.emarketer.com/Article/Google-
Facebook-Tighten-Grip-on-US-Digital-Ad-Market/1016494. (Accessed 21 March 2018).
Greif, Avner. "The fundamental problem of exchange: a research agenda in historical institutional
analysis." European Review of Economic History 4, no. 3 (2000): 251-284.
Hall, Jonathan V., Jason Hicks, Morris M. Kleiner, and Rob Solomon. “Occupational Licensing of Uber
Drivers.” Unpublished working paper presented at the ASSA 2018 Preliminary Program, 6 January,
2018.
Hannák, Anikó, Claudia Wagner, David Garcia, Alan Mislove, Markus Strohmaier, and Christo Wilson.
"Bias in Online Freelance Marketplaces: Evidence from TaskRabbit and Fiverr." In CSCW, pp. 1914-
1933. 2017.
Horton, John, and Joseph Golden. "Reputation inflation: Evidence from an online labor
market." Working Paper, NYU (2015).
Horton, John J., and Richard J. Zeckhauser. Owning, Using and Renting: Some Simple Economics of
the" Sharing Economy". No. w22029. National Bureau of Economic Research, 2016.
Hu, Nan, Paul A. Pavlou, and Jie Zhang. "On self-selection biases in online product reviews." MIS
Quarterly 41, no. 2 (2017): 449-471.
International Organization for Standardization. “ISO/FDIS 20488: Online consumer reviews -- Principles
and requirements for their collection, moderation and publication.” Available at:
iso.org/standard/68193.html. (Accessed 15 February 2018.)
Jonas, Alexandra. "Share and share dislike: The rise of Uber and Airbnb and how New York City
should play nice." JL & Policy 24 (2015): 205, p. 213.
Kang, Jun Seok, Polina Kuznetsova, Michael Luca, and Yejin Choi. "Where not to eat? Improving
public policy by predicting hygiene inspections using online reviews." In Proceedings of the 2013
Conference on Empirical Methods in Natural Language Processing, pp. 1443-1448. 2013.
Kleiner, Morris M., Alan B. Krueger, and Alexandre Mas. "A Proposal to Encourage States to
Rationalize Occupational Licensing Practices." Paper submitted to the Brookings Institution,
Washington, DC, April (2011), p. 3.
Koopman, Christopher, Matthew Mitchell, and Adam Thierer. "The sharing economy and consumer
protection regulation: The case for policy change." J. Bus. Entrepreneurship & L. 8 (2014): 529.
Kornish, Laura J. "Are user reviews systematically manipulated? Evidence from the helpfulness
ratings." Leeds School of Business Working Paper (2009).
Kotila, M., Ruben C. Rumin, and Shailin Dhar. “Compendium of ad fraud knowledge for media
investors.” WFA and The Advertising Fraud Council Report, 2017,
ttps://www.wfanet.org/app/uploads/2017/04/WFA_Compendium_Of_Ad_Fraud_Knowledge.pdf.
Krueger, Liz. “Answers for New Yorkers Concerned or Confused About the Illegal Hotel Law | NY State
Senate,” 27 May, 2014, https://www.nysenate.gov/newsroom/articles/liz-krueger/answers-new-yorkers-
concerned-or-confused-about-illegal-hotel-law. (Accessed 15 February 2018).
Lam, Chungsang Tom, and Meng Liu. "Demand and Consumer Surplus in the On-demand Economy:
The Case of Ride Sharing." Working paper. (2017).
Lao, M., Andrew I. Gavil, Tara Isa Koslov, Andrew E. Stivers, William F. Adkinson, Jr., Derek W. Moore,
Nathan E. Wilson, Julie A. Goshorn, Megan Cox, Keith Fentonmiller, Cecelia M. Waldeck, and
Christopher Bryan. “The ‘Sharing’ Economy: Issues Facing Platforms, Participants & Regulators.” A
Federal Trade Commission Staff Report. 2016.
Lee, T. H. "Online reviews could help fix medicine." Harvard Business Review (2014).
Lieber, Ethan, and Chad Syverson. "Online versus offline competition." The Oxford handbook of the
digital economy(2012): 189.
Luca, Michael, and Georgios Zervas. "Fake it till you make it: Reputation, competition, and Yelp review
fraud." Management Science 62, no. 12 (2016): 3412-3427.
Luca, Michael. "Reviews, reputation, and revenue: The case of Yelp. com." (Working Paper) (2016).
Manville, Brook. “Are Platform Businesses Eating the World?”. Forbes, 14 February, 2016,
https://www.forbes.com/sites/brookmanville/2016/02/14/are-platform-businesses-eating-the-world/.
(Accessed 14 December 2017).
“Marketplace Seller Fees Example | Where to sell online,” 2017,
http://www.wheretosellonline.com/seller-fees-example/. (Accessed 12 February 2018)
Marinescu, Ioana Elena, Nadav Klein, Andrew Chamberlain, and Morgan Smart. "Incentives Can
Reduce Bias in Online Reviews." No. w24372. National Bureau of Economic Research, 2018.
Martens, Bertin. “An Economic Policy Perspective on Online Platforms.” Institute for Prospective
Technological Studies, Digital Economy Working Paper 2016/05. 2016.
Mayzlin, Dina, Yaniv Dover, and Judith Chevalier. "Promotional reviews: An empirical investigation of
online review manipulation." American Economic Review 104, no. 8 (2014): 2421-55.
McGee, Chantel. “How Airbnb’s redesign aims to combat discrimination on the service.” CNBC, 8 April,
2017, https://www.cnbc.com/2017/04/07/airbnb-experimenting-with-site-design-to-fight-
discrimination.html.(Accessed 14 February 2018.)
McKay, Tom. “Mattress Startup Casper Sued a Mattress Review Site, Then Paid for Its Acquisition.”
Gizmodo, 24 September, 2017, https://gizmodo.com/mattress-startup-casper-sued-a-mattress-review-
site-th-1818703265. (Accessed14 February 2018.)
Moore, Mark Harrison. Creating public value: Strategic management in government. Harvard University
Press, 1995.
Mount, Kenneth, and Stanley Reiter. "The informational size of message spaces." Journal of Economic
Theory 8, no. 2 (1974): 161-192.
Nicas, Jack. “Google Rival Yelp Claims Search Giant Broke Promise Made to Regulators.” The Wall
Street Journal, 11 September, 2017, https://www.wsj.com/articles/google-rival-yelp-claims-search-
giant-broke-promise-made-to-regulators-1505167498. (Accessed 9 January 2018).
Ohlhausen, Maureen K. “Advancing Economic Liberty.” Speech given at the George Mason Law
Review's 20th Annual Antitrust Symposium on 23 February, 2017. Transcript available at:
https://www.ftc.gov/system/files/documents/public_statements/1098513/ohlhausen_-
_advancing_economic_liberty_2-23-17.pdf.
Ohlhausen, Maureen K. “Foreword – Institute for Justice,” November 2017, http://ij.org/report/license-
work-2/report/foreword/. (Accessed 20 March 2018).
O’Kane, Sean. “China will ban people with poor ‘social credit’ from planes and trains.” The Verge, 16
March, 2018, https://www.theverge.com/2018/3/16/17130366/china-social-credit-travel-plane-train-
tickets. (Accessed 25 March 2018).
“O*NET Resource Center – Overview.” O*NET Resource Center,
https://www.onetcenter.org/overview.html. (Accessed 16 March 2018).
Parker, Geoffrey G., Marshall W. Van Alstyne, and Sangeet Paul Choudary. Platform Revolution: How
Networked Markets Are Transforming the Economy and How to Make Them Work for You. WW Norton
& Company, 2016.
Parkes, David. “Iterative Combinatorial Auctions: Achieving Economic and Computational Efficiency.”
Ph.D. dissertation, 2001. University of Pennsylvania.
PwC and Interactive Advertising Bureau. IAB internet advertising revenue report: 2016 full year results.
2017.
Resnick, Paul, and Richard Zeckhauser. "Trust among strangers in Internet transactions: Empirical
analysis of eBay's reputation system." In The Economics of the Internet and E-commerce, pp. 127-157.
Emerald Group Publishing Limited, 2002.
Rosenberg, M., Nicholas Confessore, and Carole Cadwalladr. “How Trump Consultants Exploited the
Facebook Data of Millions.” The New York Times, 17 March, 2018,
https://www.nytimes.com/2018/03/17/us/politics/cambridge-analytica-trump-campaign.html. (Accessed
17 March 2018).
Rosenblat, Alex, Karen EC Levy, Solon Barocas, and Tim Hwang. "Discriminating Tastes: Uber's
Customer Ratings as Vehicles for Workplace Discrimination." Policy & Internet 9, no. 3 (2017): 256-
279.
Shannon, Claude Elwood. "A mathematical theory of communication." Bell system technical journal 27,
no. 3 (1948): 379-423.
Smith, Aaron and Monica Anderson. “Online Shopping and E-Commerce.” Pew Research Center, 19
December, 2016, http://www.pewinternet.org/2016/12/19/online-shopping-and-e-commerce/. (Accessed
20 February 2018).
Smith, Tom W, Peter Marsden, Michael Hout, and Jibum Kim. “General Social Surveys.” National
Opinions Research Center. 1972-2016. Data accessed from the GSS Data Explorer website at
gssdataexplorer.norc.org.
Soberats, Annette. “It’s illegal to ban honest reviews.” FTC Business Blog, 21 February, 2017,
https://www.ftc.gov/news-events/blogs/business-blog/2017/02/its-illegal-ban-honest-reviews. (Accessed
2 March 2018).
Steiner Peter. ‘”On the Internet, nobody knows you’re a dog”’ [Cartoon]. The New Yorker, 5 July, 1993.
Tadelis, Steven. "Reputation and feedback systems in online platform markets." Annual Review of
Economics 8 (2016): 321-340.
Tanz, Jason. “How Airbnb and Lyft finally got Americans to trust each other.” Wired, 23 April, 2014,
https://www.wired.com/2014/04/trust-in-the-share-economy/. (Accessed 19 March 2018).
Telles Jr, R. “Digital Matching Firms: A New Definition in the ‘Sharing Economy’ Space.” Office of the
Chief Economist, Economics and Statistics Administration, U.S. Department of Commerce, ESA Issue
Brief #01-16, 2016.
Teubner, Timm, Florian Hawlitschek, and David Dann. "PRICE DETERMINANTS ON AIRBNB: HOW
REPUTATION PAYS OFF IN THE SHARING ECONOMY." Journal of Self-Governance & Management
Economics 5, no. 4 (2017).
The Behavioural Insights Team. “Using Data Science in Policy,” 14 December, 2017, report available
at: behaviouralinsights.co.uk/publications/using-data-science-in-policy/.
Thierer, Adam, Christopher Koopman, Anne Hobson, and Chris Kuiper. "How the Internet, the sharing
economy, and reputational feedback mechanisms solve the lemons problem." U. Miami L. Rev. 70
(2015): 830.
U.S. Bureau of Labor Statistics. “Occupational Employment Statistics – May 2016 National
Occupational Employment and Wage Estimates United States.” Data download on XLS file from
https://www.bls.gov/oes/current/oes_nat.htm#00-0000. (Accessed 24 March 2018).
U.S. Department of the Treasury Office of Economic Policy, Council of Economic Advisors, and the
Department of Labor. “Occupational Licensing: A Framework for Policymakers,” July 2015.
Weyl, E.G. and Alexander White. “Let the Right 'One' Win: Policy Lessons from the New Economics of
Platforms.” Coase-Sandor Working Paper Series in Law and Economics No. 709, 2014.
White Ops/Association of National Advertisers. “The Bot Baseline 2016-17: Fraud in Digital
Advertising,” https://www.ana.net/content/show/id/botfraud-2017. (Accessed 16 March, 2018).
Woolacott, Emma. “Amazon's Fake Review Problem Is Now Worse Than Ever, Study Suggests.”
Forbes, 9 September, 2017, https://www.forbes.com/sites/emmawoollacott/2017/09/09/exclusive-
amazons-fake-review-problem-is-now-worse-than-ever/#429dfc497c0f. (Accessed 14 February 2018).
Zacharia, Giorgos, and Pattie Maes. "Trust management through reputation mechanisms." Applied
Artificial Intelligence 14, no. 9 (2000): 881-907.
Appendix A.1 – Online retail platform profiles
Amazon
(Marketplace)
eBay
Etsy
Jet.com
eBid
Description
Fixed-price marketplace
where 3
rd
-party sellers
sell to all Amazon
customers
One of the world’s first
online auction and fixed-
price marketplaces
Marketplace focused on
handmade and vintage
goods like art and
jewellery
E-commerce site with a
focus on consumer
discounts; has a retail
partnership program
Online auction site
founded in 1998 and
operating in 23
countries
Participants
Sides
One-sided: buyers rate
goods and sellers
Two-sided: buyers and
sellers
One-sided: buyers rate
sellers
One-sided: buyers rate
sellers
Two-sided: buyers and
sellers
Access
Only transacting
customers
Any registered eBay
member may leave
feedback
Only transacting
customers
Any registered Jet.com
member may leave
feedback
Only transacting
customers
Obligation
Optional
Optional
Optional
Optional
Optional
Content
Format
Star ratings, comments,
tags for some products
(e.g. fit for clothing),
“helpful” tags
Seller ratings, product star
ratings, badges, return
and refund guarantees
Star ratings, comments,
and photos
Star ratings and
comments
Net review score,
comments, activity
time, address
verification
Scale
Five stars
Product stars: out of five
Sellers: integer score with
categories; various
badges
Five stars
Five stars
Net score of positive
(+1), neutral (+0), and
negative (-1)
Subcategories
Product-dependent (e.g.
“comfort” for headphones)
Yes (4): description,
communication, shipping
time, shipping charges
N/A
N/A
N/A
Frequency
Once per transaction
(within 90 days of order
for products and sellers)
Reviews can be left at any
time
Once per transaction
(within 100 days of the
estimated delivery date)
Once per transaction
Once per transaction
Fun
ction
Visibility
Ratings and reviews are
visible to the public
All seller and product
ratings are public; ratings
of buyers are not
Ratings and reviews are
visible to the public
Ratings and reviews are
visible to the public
Ratings and reviews
are visible to the public
Anonymity
Reviewers can change
their “public name”
Detailed seller ratings
(subcategories)
anonymous; others not
Not anonymous
Not anonymous
Not anonymous
Weighting
Unweighted, but seller-
feedback shows average
of last 12 months
Sellers unweighted;
buyer-given feedback
distribution for last 12
months
Unweighted
Unweighted
Overall average and
five most recent
reviews shown
Influence
Affects search ranking,
customer
recommendations, may
lead to user removals
Affects search rankings,
sellers ratings, refunds,
member suspensions
Affects search rankings,
featured products and
sellers; other uses
unclear
Affects search filter
options; other uses of the
information is unclear
Affects search
rankings; other uses
unclear
Filtering
Sellers may respond;
reviews can be flagged
Sellers may respond,
reviews can be flagged
Sellers may respond to
<=3 stars; platform may
remove
Reviews are accompanied
by “verified buyer” status
Hosts may respond;
address verification
tags
Continued:
Craigslist
Facebook
Marketplace
Groupon
Shopify
Magento (API)
Description
Classifieds advertisement
website founded in 1995
with annual revenue
~700m USD
Classifieds advertisement
section of Facebook for
inter-user exchanges
Site with deals becoming
available if a minimum
number of people sign up
Platform for online stores
and POS with >600k firms
and >55bn USD annual
revenue
Open-source e-
commerce
platform/content
management system
Participants
Sides
Two-sided: buyers and
sellers
Two-sided: buyers and
sellers
One-sided: buyers rate
sellers
One-sided: buyers rate
products
One-sided: buyers rate
products
Access
Anyone
Potentially transacting
parties
Only redeeming
customers
Registered online store
users
Depends on online
store creator/operator
Obligation
Optional
Optional
Optional
Optional
Optional
Content
Format
Photos, location, contact
details for buyers,
“prohibited” goods flag
Photos, location, seller
profile, private messages
between buyers and
sellers
Star ratings and written
comments; reviewer tags
(e.g. “Top Reviewer”)
Star ratings and written
comments
Star ratings and written
comments (but can
vary by store)
Scale
N/A
N/A
Five stars
Five stars
Five stars
Subcategories
N/A
N/A
N/A
N/A
Determined by online
store operator
Frequency
Varies
Varies
Once per transaction
Once per transaction
Once per transaction
Function
Visibility
All elements visible to
public except private
messages
All elements visible to
other Facebook users
except private messages
Visible to the public
Visible to public upon
approval of online store
operator
Visible to public upon
approval of online store
operator
Anonymity
User identities
anonymous
Non-anonymous
Non-anonymous
Non-anonymous
Depends on online
store creator/operator
Weighting
N/A
N/A
Unweighted
Unweighted
Unweighted
Influence
N/A
N/A
Affects search rankings,
featured deals
Influence determined by
online store operator and
users who see reviews
Influence determined
by online store
operator and users
who see reviews
Filtering
Craigslist offers a
‘prohibited’ goods/service
flag
User interactions linked to
Facebook user profiles
Formal dispute process
where both parties
notified
Online store operator may
remove (or flag)
comments
Online store operator
may remove comments
Core trust problem facing online retail platforms
Ex-ante identification of seller and product/service quality prior to a financial transaction.
Industry level findings: stylized facts and selected examples
1. Star ratings (out of five) + buyer comments are among most common form of mechanism in
this industry. In fact, it is generally difficult to find online retail platforms that lack this mechanism
(particularly when conditioning on there being some kind of feedback mechanism). Some examples are
given below.
Platform
Star rating appearance
Amazon
eBay
Etsy
Intra-industry differences usually focus on reported distributional and sub-group information (e.g. other
products reviewed by reviewers) and seller ratings in addition to product ratings (such as eBay’s
feedback scores).
2. Transparency of inputs to the trust mechanism is generally high across the industry. For
many platforms, aggregate scores are able to be reconstructed from individual ratings and reviews,
which are all publicly available. Platforms’ use of review information – e.g. in regard to seller sanctions
– is also often publicized.
However, this is not always the case. Consider for example the Shopify API, which is used by over
600,000 businesses worldwide. Its API allows platform managers (usually product sellers) to unilaterally
censor and delete feedback without reason:
3. Market participants are usually not compelled to input information, but many platform-
initiated mechanisms exist to encourage these. A common form of these is a reminder email to
leave a rating for a purchase (see example below). Encouragement may also come directly from a
seller on some platforms.
4. Platforms with less-rich trust mechanisms are associated with a higher frequency of lower
quality and lower-priced items. As discussed earlier in this report, this may be a consequence of
competitive positioning. An example of this is Facebook Marketplace (an illustrative listing is below)
which has no ratings or review system, and includes a user option to “Only show free listings”.
Appendix A.2 – Short-term accommodation platform profiles
Airbnb
VRBO
Homeaway
Flipkey
Couchsurfing.com
Description
Largest short-term rental
marketplace with >4m
shared and full-space
listings
Airbnb competitor with
focus on vacation rentals
and large sq. ft. full-
space listings
Vacation rental
marketplace owned by
VRBO’s parent company,
Expedia
Vacation rental
marketplace with >300k
listings worldwide;
subsidiary of TripAdvisor
Shared lodging platform
with >15m members; no
host charges allowed
Participants
Sides
Two-sided: guests and
hosts
Two-sided
Two-sided
One-sided (travelers
only)
Two-sided: all community
members regardless of
type on the site
Access
Only transacting parties
Only transacting parties
Transacting parties for
ratings, any registered
member for the
guestbook
Anyone may write a
review
Transacting parties for
surf/host refs, registered
for other references
Obligation
Optional, but incentivized
via email on check-out
day
Optional, but incentivized
via email three days
post-stay
Optional, but incentivized
via email three days
post-stay
Optional, but incentivized
via post-stay emails
Optional, but incentivized
via post-stay emails
Content
Format
Comments by guests
and hosts (with word
limit), star ratings, user
verification, list photos,
and host badges
Comments by guests and
hosts (with word limit),
star ratings, user
verification, list photos,
and host badges
Reviews by guests and
hosts, star ratings, user
verification, photos,
badges, and “guestbook
comments”
Star ratings, written
reviews, review count,
photos, amenities lists,
“payment protection”
status
Two types of references
(surf/host and personal),
+/- feedback tags,
verification systems
Scale
Five stars, various
badges and tags (e.g.
‘Superhost,’ ‘Highly
Rated,’ ‘Rare Find’)
Five stars, various
badges (e.g. ‘Premier
Partner,’ ‘Popular,’
‘Excellent’)
Five stars, various
badges and tags (e.g.
‘Superhost,’ ‘Popular,’
‘Excellent’)
Five stars (aggregate and
individual)
Four reference categories
(e.g. Don’t Recommend),
other discrete tags
Subcategories
Yes (for guests):
accuracy, location,
communication, check in,
cleanliness, value
Yes (for hosts):
cleanliness,
communication, and
adherence to house rules
Yes (for hosts):
cleanliness,
communication, and
adherence to house rules
No
Yes: 17 categories split by
+/- reference, e.g.
“punctual,” “unfriendly”
Frequency
Once after transaction
(within 14 days of
checkout)
Once after transaction
(within one year of
checkout)
Once after transaction
(within one year of
checkout)
Usually one review per
transaction (but more
possible)
Surf/host refs: once after
trans. (within 14 days)
Other refs: at any time
Function
Visibility
Public + private host;
double-blind period;
traveler review (by host)
private
Public + private host; 14-
day double-blind period;
traveler review (by host)
private
Public + private host; 14-
day double-blind period;
traveler review (by host)
private
Reviews are public after
approval by platform
Surf/host positive refs ad
tags visible, negative refs
only seen by the platform
Anonymity
Not anonymous
Not anonymous
Not anonymous
Anonymous
Tags anonymous; refs not
Weighting
Unweighted (but
aggregate host scores
only visible after >3
reviews)
Unweighted (but
aggregate host scores
only visible after >3
reviews)
Unweighted (but
aggregate host scores
only visible after >3
reviews)
Unweighted (simple
average star rating
shown)
Unweighted (total
proportion of positive
references shown)
Influence
Affects placement in
search, host badges,
may lead to access being
revoked
Affects placement in
search, tags, host
comparison dashboard,
access decisions
Affects placement in
search, tags, host
comparison dashboard,
access decisions
Affects placement in
search, eligibility for
featuring and awards,
warning labels
Affects search rankings,
ability to send messages,
support, favourites, more
Filtering
Hosts may respond to
comments; reviews can
be left for partial stays;
content flagging and
dispute system
Hosts may respond to
comments; reviews can
be flagged for removal if
they violate content
guidelines
Hosts may respond to
comments; reviews can
be flagged for removal if
they violate content
guidelines
Reviewers may delete;
fraud-detection system
operated by platform;
suspect reviews may be
flagged by anyone
Users may respond to
references; formal dispute
forms available for
violating guidelines
Core trust problem facing short-term accommodation platforms
Identifying quality and ensuring personal and property safety for travelers and hosts.
Industry level findings: stylized facts and selected examples
1. “Double-blind” review processes are commonly used to mitigate retaliatory reviews. Most
platforms we examined hide travelers’ and hosts’ feedback from each other until both have submitted
feedback on the other or a preset time window for providing feedback (usually 14 days) elapses. This
aims to incentivize honesty by preventing retaliation. As Airbnb noted when introducing the system:
81
Both hosts and guests may worry that if they leave an honest review that includes praise and
criticism, they might receive an unfairly critical review in response. To address this concern,
reviews will be revealed to hosts and guests simultaneously. Starting today, hosts and their
guests will only see reviews they receive from a completed trip after both participants have
completed their assessment of the experience.
2. Rating subcategories and qualitative tags are frequently used and allow multiple quality
dimensions of a user to be communicated. For example, VRBO has ratings of travelers along
“cleanliness,” “communication,” and “adherence to house rules.” There is heterogeneity of
subcategories and tags across platforms, and the named dimensions could potentially involve trade-
offs in some cases that users may weight differently (e.g. “value” vs. “cleanliness” as shown below).
3. There does not seem to be a norm for the side of the market the trust mechanism draws
information, the anonymity of responses, and access to the platform. Some platforms draw
feedback from both travelers and hosts (e.g. Airbnb), whereas others only elicit responses from
travelers (e.g. Flipkey). Even among platforms with two-sided mechanisms, the nature and volume of
information is not the same across all of them. For example, VRBO and Homeaway feature sub-
category ratings for travelers only; Airbnb’s sub-categories apply to hosts only. In regards to anonymity
and access, Flipkey allows for anonymous reviews from anyone who visits the site, and
Couchsurfing.com only makes public feedback positive. This lack of “convergence” stands in contrast to
several other industries we examine.
4. Platforms appear to place a high level of importance on identity verification mechanisms.
Identity verification usually occurs through the provision of personal information from government IDs,
credit cards, social media profiles, email addresses, often in combination with each other. This are
often an explicit requirement to use the platform, and can also serve as a gateway to other features (as
in the case of Couchsurfing.com below).
81
Airbnb, "Building Trust with a New Review System."
Platform functionality benefits of identity verification on Couchsurfing.com
Appendix A.3 – Ridesharing platform profiles
Uber
Lyft
Juno
Via
Description
Largest ride-sharing company in US
Uber’s major competitor
NYC ridesharing app,
merged with former
competitor Gett
NYC, Chicago and DC-based
app focused on pooled
transportation
Participants
Sides
Two-sided
Two-sided
One-sided: passengers rate
drivers
One-sided: passengers rate
drivers
Access
Only transacting parties
Only transacting parties
Only transacting parties
Only transacting parties
Obligation
Optional (was mandatory)
Optional
Optional
Optional
Content
Format
Star rating, optional comments, tags
Star rating, optional comments
Star ratings
Star rating, optional comments
Scale
Five stars
Five stars
Five stars
Five stars
Subcategories
No, although tags may ‘modify’ ratings
Yes, including cleanliness, safety,
navigation, friendliness
No
No
Frequency
After each ride
After each ride
After each ride
After each ride with a new
driver
Function
Visibility
Users see own profile and those they are
matched with
Users see own profile and those
they are matched with
Users see own profile and
those they are matched with
Not seen by users
Anonymity
Yes
Yes
Yes
Yes
Weighting
Only most recent 500 ratings
Only most recent 100 ratings
Unknown
Unknown
Influence
Drivers banned below certain level, users
may cancel after match but before
pickup, some benefits for highly rated
drivers
Drivers banned below certain level,
users may cancel after match but
before pickup, some benefits for
highly rated drivers
Platform uses ratings to
qualify drivers
Platform uses feedback to
qualify drivers and ensure
ongoing driver quality
Filtering
“Ratings protection” – consistent low
raters discarded as are low ratings with
certain banned reasons (e.g. GPS route)
Strict policy against review removal
/ filtering
Unknown
Unknown
Core trust problem facing ridesharing platforms
Ensuring the safety of passengers and drivers, and the quality of experience for passengers.
Industry level findings: stylized facts and selected examples
1. Trust mechanisms operate as the ‘qualifying requirement’ for ridesharing drivers. Ridesharing
platforms tend to require drivers to maintain a certain minimum rating in order to continue to operate on
the platform. This allows platforms to filter out those drivers who may be endangering passengers or
offering a low quality service.
Example Uber driver deactivation message
2. Changes to trust mechanism are hotly contested among drivers and passengers, indicating
perceived importance by both parties. A number of very active forums exist where drivers and
passengers on ridesharing platforms debate changes and proposals to change ridesharing ratings
systems. An example is the ratings sub-forum on uberpeople.net which attracts around 100 posts and
comments per week. Uber is aware of the need to maintain confidence and satisfaction with their rating
system, as evidenced by the major announcements included in their ‘180 Days of Change’ campaign
which included ratings changes geared towards improving driver satisfaction.
180 Days of Change campaign changes to ratings system
3. Profiles are not public, but are made available to transacting parties.
Ridesharing platforms tend to limit the ratings data made available publicly about their drivers and
passengers. Because, in general, users are not able to choose their own matches on these platforms
(geographical location is more important), making ratings data available only after matching seems to
be sufficient for most ridesharing platforms to enable transactions to occur.
Appendix A.4 – Online freelance labor platform profiles
Upwork
Freelancer.com
Fiverr
Toptal
Crew.co
Description
Global freelance labor
website with $1bn in annual
billings
Global freelance labor
website based on
freelancers bidding for
work
Claims to be ‘world’s
largest online
marketplace for freelance
services’
Vetted high-end
freelance labor
marketplace
Claims to offer ‘top 3% of
freelancers’ in the world
Participants
Sides
Two-sided: freelancers and
clients
Two-sided
Two-sided
Platform acts as a filter
and no feedback is taken
All prospective
freelancers are required
to take tests and provide
portfolios to Toptal
Two-sided
Access
Only contracting parties
Only contracting parties
Only contracting parties
Only contracting parties
Obligation
Optional, but incentivized
Optional, but incentivized
Optional, but incentivized
Optional
Content
Format
Written feedback on clients,
star ratings and comments on
freelancers; results on tests
offered by Upwork
Star ratings, written
feedback, skill tests
offered by freelancer.com
Star ratings, written
feedback, private
feedback on clients, “Pro”
badge
Platform assesses
applicants for freelance
roles and determines
suitability, which is
guaranteed to clients
All freelancers must
complete tests in order to
become eligible for jobs
A portfolio of previous
work is displayed to
prospective clients
Score, written feedback
Scale
5 stars, 0-100 job success
score; test pass badges
5 stars, test pass badges
Five stars
Score out of 10
Subcategories
No
Yes for freelancers:
quality, communication,
expertise,
professionalism, hire
again?
Yes for freelancers:
Communication, service,
buy again or recommend
No
Frequency
Once at end of transaction
Once per transaction
Once per transaction
Once at end of
transaction
Function
Visibility
Freelancer feedback
aggregated into public ‘job
success score’ out of 100.
Client reviews private to
Upwork
Freelancer reviews
public, client reviews
private to freelancer.com
Public
Reviews are not used to
filter. A private algorithm
is used to assess
suitability of freelancers
for client jobs
Not public, visible only to
crew.co
Anonymity
Not anonymous, some
contract details are visible
Not anonymous, some
contract details are
visible
Not anonymous
No
Weighting
Weighted by recency (best of
6,12 and 24 months)
Aggregated into
reputation score based
on recency, number of
reviews, size of projects
and quality of reviewer
None
Unknown
Influence
Affects placement in search
rankings, top performer
badges
Used in ranking algorithm
in searches, top
performer badges
Used to create ‘level’ of
freelancer, affecting
ranking algorithm
If rating falls below a
certain level, access can
be revoked
Filtering
Dispute resolution system,
top performers may remove
one review
Not possible to remove
negative ratings
Sellers may invite users
to edit reviews of each
other
None
Core trust problem facing platforms
Ensuring freelancers have the capabilities to complete job in the time and of the quality expected by
client.
Industry level findings: stylized facts and selected examples
1. In marketplace freelance platforms, ratings have a two-fold impact on freelancers. Ratings
affect both the positioning in search algorithm (controlled by the platform) and the likelihood clients will
hire the freelance if presented in search results (controlled by the client). In thick marketplaces, this
double impact of ratings seems to strengthen the incentive for freelancers to earn good reviews. For
example, Upwork not only promotes highly rated freelancers higher in their search algorithm, they also
offer a separate ‘suggested freelancers’ section for clients on their homepage and a ‘suggested
freelancers’ email for clients after they have posted a job.
Example suggested freelancers section
2. Where the platform is more closely involved in each individual transaction, ratings seem to
play a smaller role. Where search is not the primary method of matching on a platform, platforms
seem to absorb a greater proportion of the risk associated with bad transactions. For example, on
Toptal, portfolio work rather than previous feedback is provided to prospective clients on the platform
and any feedback after a transaction is only used by the platform to determine the suitability of
freelancers for future projects. These kinds of platforms may be seen as closer to a traditional agency
model in the services industry, rather than the marketplace model that has come to characterize many
parts of the sharing economy.
3. Non-anonymized feedback seems to be expected by clients on freelance platforms. Because
clients tend to have specialized requests for freelancers on these platforms, more information about the
types of jobs completed by freelancers previously is likely to help clients identify whether a given
freelancer has the skills necessary to complete the specialized task at hand. Non-anonymized
feedback, which includes details about contracts previously completed, and portfolios of work are likely
to help clients overcome the trust problem inherent in this market.
Appendix A.5 – Online advertising platform profiles
Facebook Advertising
Google Adwords
Bing Ads
Snap Inc
Outbrain
Description
Targeted and pay-per-click
advertising on world’s largest
social network
Pay-per-click and search
advertising on world’s
largest search engine
Search advertising by
Microsoft
Youth-targeted video and
integrated advertising on
messaging platform
Online advertiser
specializing in presenting
sponsored website links
Participants
Sides
Two-sided: advertiser and ad
viewer
Two-sided: advertiser
and ad viewer
Two-sided: advertiser
and ad viewer
One-sided: to advertiser
One-sided: to advertiser
Access
Only parties who have
created or viewed an ad
Only parties who have
created or viewed an ad
Only parties who have
created or viewed an ad
Only parties who have
created or viewed an ad
Only parties who have
created or viewed an ad
Obligation
Impressions data collected
automatically, ad viewer
feedback optional
Click data collected
automatically, ad viewer
feedback optional
Click data collected
automatically, ad viewer
feedback optional
Viewership data collected
automatically
Click data collected
automatically
Content
Format
For advertiser: report on
impressions, engagement
broken down by demographic
For ad viewer: “report ad”
feature, like, share, comment
For advertiser: report on
number of clicks, click-
through-rate, conversion
rate
For ad viewer: report ad
feature
For advertiser: report on
number of clicks, click-
through-rate, conversion
rate, relative success
rates of ad by website
For ad viewer: report ad
feature
Report on number of
impressions, swipe ups,
click through and time
spent watching videos
Report on number of
clicks, click-through-rate,
conversion rate
Scale
Advertiser: Number of
impressions, engagement
Viewer; categories of
complaint, like / not
Advertiser: number and
rate, $/click
Viewer: written
complaints
Advertiser: number and
rate, $/click
Viewer: written
complaints
Number, rate, seconds
Number, rate, seconds
Subcategories
Demographic breakdown,
detailed data
Very limited geographic
information, broken down
by key word
Very limited geographic
information, broken down
by key word
Demographic reporting
(age, location, gender)
Geographic reporting
Frequency
Updated continuously
throughout advertising
campaigns
Updated continuously
throughout advertising
campaign
Updated continuously
throughout advertising
campaign
Updated continuously
throughout advertising
campaign
Updated continuously
throughout advertising
campaign
Fun
ctio
n
Visibility
Private
Private
Private
Private
Private
Anonymity
Viewer interactions are
anonymous by default, unless
viewers choose to engage
publicly with ad using social
tools.
Anonymous
Anonymous
Anonymous
Anonymous
Weighting
None.
None
None
None
None
Influence
Placement in Newsfeed
algorithm, cost of advertising
Placement in search, cost
of advertising
Placement in search,
cost of advertising
Cost and placement
Ad placement and cost
Filtering
Unknown – it is not clear
whether Facebook excludes
certain types of engagement
from advertising manager
Unknown – it seems
unlikely clicks are
excluded, Google
response to complaints
unknown.
Unknown – it seems
unlikely clicks are
excluded, Google
response to complaints
unknown.
Unknown
Unknown
Core trust problem facing online advertising platforms
Ensuring advertisements are presented to appropriate audiences which will drive engagement and
revenue for advertisers
Industry level findings: stylized facts and selected examples
1. Advertisers expect more granular and detailed data on the effectiveness of their advertising
in order to enable trust in a platform. Compared to other industries profiled in this document, the
trust mechanisms on online advertising platforms are substantially more sophisticated than star ratings
and similar mechanisms. This may reflect the fact that the revenue-generating side of the market for
advertising is businesses rather than consumers, who may have more resources to analyze feedback
than many consumers. Trust in online advertising platforms seems to require demographic information
and detailed click-through data to be provided as a minimum.
2. Anonymity is a common feature of trust mechanisms on online advertising platforms.
Platforms rarely provide information about the identity of users who have engaged with advertising to
advertisers. Indeed, attempts to access identifying data may constitute a breach of advertiser’s terms
and conditions, as in the recent case of Cambridge Analytica and Facebook.
82
In many cases,
advertisers would be willing to pay for further information about viewers of their advertisements which
may help improve targeting and marketing efforts. However, protecting certain aspects of the identity of
users seems necessary to maintain thickness of the advertising market, to ensure that the user base
from which platforms derive revenue remains large. This tradeoff is reflected in the trust mechanisms of
online advertising platforms, which will display various levels of demographic data to advertisers, but
are unlikely to provide individualized and identifiable data.
3. Despite efforts by advertising platforms, trust on online platforms seems to be significantly
influenced by instances of strategic manipulation. One common problem on online advertising
platforms that diminishes perceptions of trust by advertisers is instances of strategic manipulation by
other vendors on online platforms. A recent report on by the Association of National Advertisers found
that ad bots inflate monetized audiences on online platforms by between 5 and 50%, at a cost of
around $6.3 billion per annum to advertisers.
83
This has led some advertisers to decrease advertising
spending on online platforms, with the World Federation of Advertisers (whose members include some
of the largest advertisers including McDonalds’s, Visa and Unilever) warning members
“Until the industry can prove that it has the capability to effectively deal with ad fraud, advertisers
should use caution in relation to increasing their digital media investment, to limit their exposure to
fraud.”
84
82
Rosenberg, M., Nicholas Confessore, and Carole Cadwalladr. “How Trump Consultants Exploited the
Facebook Data of Millions.” The New York Times, 17 March, 2018,
https://www.nytimes.com/2018/03/17/us/politics/cambridge-analytica-trump-campaign.html. (Accessed 17 March
2018).
83
White Ops/Association of National Advertisers. “The Bot Baseline 2016-17: Fraud in Digital Advertising,”
https://www.ana.net/content/show/id/botfraud-2017. (Accessed 16 March, 2018).
84
Kotila, M., Ruben C. Rumin, and Shailin Dhar. “Compendium of ad fraud knowledge for media investors.” WFA
and The Advertising Fraud Council Report, 2017,
https://www.wfanet.org/app/uploads/2017/04/WFA_Compendium_Of_Ad_Fraud_Knowledge.pdf. (Accessed 16
March, 2018).
In response, the Media Rating Council developed a third-party accreditation system for measurement of
the effectiveness of ads, which is now used by Facebook, Twitter, Google and others.
85
85
Bahlavan, Pahlavan. “Building trust and increasing transparency with MRC-accredited measurement.” Google
Agency Blog, 21 February, 2017, https://agency.googleblog.com/2017/02/building-trust-and-increasing.html.
(Accessed 16 March, 2018).
Appendix B – Top and bottom professions on O*Net variables
Contact with others
Consequence of error
Responsibility for others’ health and safety
Appendix C – Benefits estimation methodology and the full regulatory
substitution index
This section provides further notes on the regulatory burden reduction estimates in Chapter 3.
Replication: dataset construction and associated STATA files
Data sources
The variables in our main dataset come from three sources:
1. Data from License to Work, Second Edition can be found at the following URL:
http://ij.org/report/license-work-2/ltw2-data/
2. Data from O*NET can be found at the following URL: https://www.onetonline.org/
3. The remaining data about platform presence were hand-collected and are available in the Stata
file “l2w_onet_data_0318.dta.”
Additionally, for the burden calculation, we use state-level occupation employment data from the US
Bureau of Labor Statistics’ National Occupational Employment and Wage Estimates for May 2016 (the
most recent period at the time of writing). These data are contained within the Stata file “burden.dta”
and are available in raw form at https://www.bls.gov/oes/current/oes_nat.htm#00-0000.
Occupations were manually matched between O*NET and License to Work, Second Edition by
occupation name. The variables “occname_l2w,” “occname_onet,” and “id_onet” track these matches
for reproducibility. The only occupation where an O*NET match was not found was for an “Auctioneer,”
so this was dropped from the dataset when preparing the regulatory substitution index.
STATA files
The file “l2w_onet_data_0318.dta” is the starting dataset which includes selected variables from
the previous section.
The file “occlicensev2.do” conducts all of the calculations and charts in Chapter 3 of the report
The file “burden_calc.do” calculates the reduction in regulatory burden from the report. It should
be run after “occlicensev2.do.”
The remaining .dta files are each input and intermediate output data for “occlicensev2.do.”
The regulatory substitution index
Objective
The intent of the regulatory substitution index is to provide a parsimonious snapshot of the potential for
online platforms’ trust mechanisms to substitute for occupational licensing in a given industry.
The index takes values between 0-1, with 1.00 being an “ideal” occupation for trust mechanisms to
substitute for. It represents a simple average of nine components that are also scaled to take scores
between 0-1.00 (with the same interpretation).
Methodology
The construction of the index uses a “distance-to-frontier” approach used, for example, by the World
Bank’s Doing Business Rankings.
86
The calculation begins by defining a “best” and “worst” value of
each dimension present within the dataset. These were shown in Table 2 of Chapter 3, which is
reproduced below.
Topic/dimension
Best value (w example)
Worst value
1. Performs or Works
Directly with the Public
Yes
(Florist)
No
(Animal Breeder)
2. Contact with others
(category)
Constant contact
(Shampooer)
Occasional contact
(Taxidermist)
3. Contact with others
(frequency)
98%
(Gaming Cage Worker)
24%
(Paving Contractor)
4. Responsible for others’
health and safety (category)
No Responsibility
(Plant Nursery Worker)
Very High Responsibility
(School Bus Driver)
5. Responsible for others’
health and safety (frequency)
22%
(Plant Nursery Worker)
80%
(Emergency Medical
Technician)
6. Consequence of error
(frequency)
Not Serious at All
(Funeral Attendant)
Extremely Serious
(Emergency Medical
Technician)
7. Consequence of error
(proportion)
23%
(Funeral Attendant)
94%
(Midwife, Direct Entry)
8. Online platform exists
Yes
(Taxi Drivers)
No
(Milk Sampler)
9. Online platform is an on-
demand platform
Yes
(Taxi Drivers)
No
(Door Repair Contractors)
The score (denote this ) for each of the nine categories for each occupation is then normalized to a
common unit via the following linear transformation:
Finally, the individual scores are aggregated by simple averaging to produce the rankings of
occupations in Chapter 3. The relevant code for the creation of the index and its component scores is
at lines 61-75 of “occlicense.do.”
The full regulatory substitution index (including components)
These are generated by running “occlicense.do” with index values being contained in the variable
“reg_sub_index and are shown below.
86
See “Distance to Frontier – Doing Business.” The World Bank Group, 2017,
http://www.doingbusiness.org/data/distance-to-frontier. (Accessed 25 February 2018).
Occupation name
Index
score
Works direct w
the public?
Contact
(category)
Resp. for health/safety
(category)
Consequence of error
(category)
Contact
(score)
Resp
(score)
Error
(score)
Platfo
rm?
Platform
name
On
demand
?
Barber
0.9305
148
1
Constant contact
Some responsibility for
others
Some consequence of
error
4.83
2.05
2.36
1
Shortcut
1
Bartender
0.9214
658
1
Constant contact
Moderate responsibility
for others
Little consequence of
error
4.7
2.78
1.42
1
Saucey
1
Shampooer
0.9107
04
1
Constant contact
Moderate responsibility
for others
Some consequence of
error
4.92
2.98
1.82
1
GlamSqu
ad
1
Massage Therapist
0.8542
917
1
Contact most of
the time
Moderate responsibility
for others
Moderate
consequence of error
4.41
2.6
2.5
1
StyleBee
1
Cosmetologist
0.8496
007
1
Constant contact
Very high responsibility
for others
Some consequence of
error
4.82
3.52
2.26
1
Vensette
1
Manicurist
0.8461
846
1
Constant contact
Very high responsibility
for others
Some consequence of
error
4.5
3.55
1.69
1
GlamSqu
ad
1
Truck Driver, Other
0.8437
837
1
Constant contact
Some responsibility for
others
Moderate
consequence of error
4.67
2.47
3.37
1
Uber
Freight
1
Taxi Driver/Chauffeur
0.8356
429
1
Contact most of
the time
Moderate responsibility
for others
Some consequence of
error
4.49
3.23
2.28
1
Uber
1
Interior Designer
0.8099
01
1
Contact most of
the time
Moderate responsibility
for others
Moderate
consequence of error
4.48
3.38
2.62
1
Homepoli
sh
1
Florist
0.8070
254
1
Constant contact
Very high responsibility
for others
Some consequence of
error
4.53
3.83
2.23
1
BloomNat
ion
1
Travel Guide
0.7834
177
1
Constant contact
Moderate responsibility
for others
Very high
consequence of error
4.51
2.92
3.79
1
Dopios
1
Security Guard, Unarmed
0.7540
812
1
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
4.44
3.79
3.22
1
Bannerm
an
1
Truck Driver, Tractor-Trailer
0.7231
444
1
Contact most of
the time
Moderate responsibility
for others
Very high
consequence of error
3.93
2.96
3.91
1
Uber
Freight
1
Home Entertainment Installer
0.6948
397
1
Contact most of
the time
Some responsibility for
others
Moderate
consequence of error
4.18
2
2.64
1
Yelp
0
Landscape Contractor
(Commercial)
0.6839
122
1
Contact about
half the time
Moderate responsibility
for others
Moderate
consequence of error
3.19
3.31
2.91
1
Plowz &
Mowz
1
Landscape Contractor
(Residential)
0.6839
122
1
Contact about
half the time
Moderate responsibility
for others
Moderate
consequence of error
3.19
3.31
2.91
1
Plowz &
Mowz
1
Makeup Artist
0.6747
867
0
Constant contact
Moderate responsibility
for others
Some consequence of
error
4.68
3.36
2.36
1
Vensette
1
Locksmith
0.6568
448
1
Contact most of
the time
Moderate responsibility
for others
Moderate
consequence of error
4.39
2.83
2.83
1
Yelp
0
Travel Agency
0.6522
074
1
Contact most of
the time
Moderate responsibility
for others
Moderate
consequence of error
4.45
2.55
3.38
1
TripAdvis
or
0
Door Repair Contractor
(Residential)
0.6460
181
1
Contact most of
the time
Moderate responsibility
for others
Moderate
consequence of error
4.38
2.87
2.99
1
Yelp
0
Bill Collection Agency
0.6128
443
1
Constant contact
Some responsibility for
others
Some consequence of
error
4.88
1.84
2.38
0
Animal Trainer
0.6018
339
1
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
4.23
3.6
2.75
1
Yelp
0
Painting Contractor
(Residential)
0.6016
12
1
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
4.34
3.85
2.66
1
Yelp
0
Animal Control Officer
0.5916
71
1
Constant contact
Moderate responsibility
for others
Very high
consequence of error
4.76
3.25
4.39
1
Yelp
0
Social and Human Service
Assistant
0.5814
488
1
Constant contact
Some responsibility for
others
Moderate
consequence of error
4.96
2.08
2.9
0
HVAC Contractor (Residential)
0.5707
766
1
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
4.34
4.17
2.92
1
Yelp
0
Optician
0.5678
94
1
Constant contact
Moderate responsibility
for others
Some consequence of
error
4.81
2.7
2.15
0
Taxidermist
0.5675
037
1
Contact about
half the time
Some responsibility for
others
Moderate
consequence of error
3.12
2.42
2.8
1
Yelp
0
Glazier Contractor (Residential)
0.5564
181
1
Contact most of
the time
Very high responsibility
for others
Very high
consequence of error
4.39
3.82
3.74
1
Yelp
0
Interpreter, Sign Language
0.5527
314
1
Constant contact
Some responsibility for
others
Moderate
consequence of error
4.86
2.43
2.89
0
Gaming Dealer
0.5308
572
1
Constant contact
Moderate responsibility
for others
Moderate
consequence of error
4.59
2.66
2.56
0
Tree Trimmer
0.5212
944
1
Contact most of
the time
Very high responsibility
for others
Very high
consequence of error
4.37
4.34
3.81
1
Yelp
0
Gaming Cage Worker
0.5082
188
1
Constant contact
Moderate responsibility
for others
Very high
consequence of error
4.97
2.87
3.5
0
Child Care Home, Family
0.5064
598
1
Constant contact
Moderate responsibility
for others
Moderate
consequence of error
4.95
3.15
3.16
0
Gaming Supervisor
0.5043
186
1
Constant contact
Moderate responsibility
for others
Moderate
consequence of error
4.89
2.92
3.37
0
Coach, Head (High School
Sports)
0.4905
837
1
Constant contact
Very high responsibility
for others
Moderate
consequence of error
4.92
3.75
2.71
0
Skin Care Specialist
0.4885
957
1
Contact most of
the time
Moderate responsibility
for others
Moderate
consequence of error
4.24
2.78
2.64
0
Dietetic Technician
0.4849
877
1
Constant contact
Moderate responsibility
for others
Moderate
consequence of error
4.68
3.34
2.87
0
Preschool Teacher, Public
School
0.4834
979
1
Constant contact
Very high responsibility
for others
Moderate
consequence of error
4.73
3.51
2.79
0
Dental Assistant
0.4818
338
1
Constant contact
Very high responsibility
for others
Moderate
consequence of error
4.89
4.02
2.51
0
Paving Contractor (Residential)
0.4800
791
1
Contact most of
the time
Extremely high
responsibility for others
Moderate
consequence of error
3.86
4.58
3.42
1
Yelp
0
Door Repair Contractor
(Commercial)
0.4793
514
1
Contact most of
the time
Moderate responsibility
for others
Moderate
consequence of error
4.38
2.87
2.99
0
Pharmacy Technician
0.4763
361
1
Constant contact
Moderate responsibility
for others
Very high
consequence of error
4.79
2.96
3.72
0
Packer
0.4681
875
0
Contact most of
the time
Very high responsibility
for others
Some consequence of
error
4.15
3.53
1.99
1
Yelp
0
Pest Control Applicator
0.4677
482
1
Contact most of
the time
Moderate responsibility
for others
Moderate
consequence of error
4.38
2.84
3.27
0
Milk Sampler
0.4620
626
1
Constant contact
Moderate responsibility
for others
Moderate
consequence of error
4.61
3.43
3.11
0
Slot Supervisor
0.4524
173
1
Constant contact
Very high responsibility
for others
Moderate
consequence of error
4.72
3.67
3.23
0
Upholsterer
0.4481
729
0
Contact most of
the time
Moderate responsibility
for others
Some consequence of
error
3.65
2.84
2.3
1
Yelp
0
Funeral Attendant
0.4437
314
1
Contact most of
the time
Moderate responsibility
for others
Moderate
consequence of error
4.33
3.36
3.05
0
Nursery Worker
0.4419
584
1
Contact most of
the time
Moderate responsibility
for others
Some consequence of
error
3.59
3.11
1.99
0
Forest Worker
0.4414
311
1
Constant contact
Very high responsibility
for others
Moderate
consequence of error
4.7
4.11
2.89
0
Painting Contractor
(Commercial)
0.4349
454
1
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
4.34
3.85
2.66
0
Mason Contractor (Residential)
0.4333
839
0
Contact most of
the time
Moderate responsibility
for others
Moderate
consequence of error
3.79
3.15
2.5
1
Yelp
0
Bus Driver, City/Transit
0.4272
411
1
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
4.45
3.66
3.26
0
Title Examiner
0.4263
835
0
Contact most of
the time
Some responsibility for
others
Moderate
consequence of error
3.65
2.26
3.46
1
Yelp
0
Log Scaler
0.4121
049
1
Contact most of
the time
Moderate responsibility
for others
Moderate
consequence of error
3.8
3.13
2.99
0
Sheet Metal Contractor, HVAC
(Residential)
0.4063
599
0
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
4.28
3.74
3.28
1
Yelp
0
Sheet Metal Contractor, Other
(Residential)
0.4063
599
0
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
4.28
3.74
3.28
1
Yelp
0
HVAC Contractor (Commercial)
0.4041
1
1
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
4.34
4.17
2.92
0
Insulation Contractor
(Residential)
0.3955
123
0
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
3.72
3.59
2.63
1
Yelp
0
Psychiatric Technician
0.3928
584
1
Constant contact
Very high responsibility
for others
Moderate
consequence of error
4.65
4.42
3.44
0
Midwife, Direct Entry
0.3865
021
1
Constant contact
Moderate responsibility
for others
Extremely high
consequence of error
4.76
3.48
4.92
0
Floor Sander Contractor
(Residential)
0.3820
669
0
Contact about
half the time
Moderate responsibility
for others
Moderate
consequence of error
3.31
3.29
2.5
1
Yelp
0
Iron/Steel Contractor
(Residential)
0.3684
348
0
Contact most of
the time
Very high responsibility
for others
Very high
consequence of error
4.39
4.33
3.57
1
Yelp
0
Mobile Home Installer
0.3572
313
1
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
3.75
3.83
3.2
0
Vegetation Pesticide Applicator
0.3465
61
1
Contact most of
the time
Moderate responsibility
for others
Very high
consequence of error
3.62
3.3
3.82
0
Terrazzo Contractor
(Residential)
0.3399
91
0
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
3.55
3.8
3.22
1
Yelp
0
Cement Finishing Contractor
(Residential)
0.3142
417
0
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
3.52
4.34
3.05
1
Yelp
0
Paving Contractor (Commercial)
0.3134
124
1
Contact most of
the time
Extremely high
responsibility for others
Moderate
consequence of error
3.86
4.58
3.42
0
Fisher, Commercial
0
0
Contact most of
the time
Very high responsibility
for others
Very high
consequence of error
3.91
3.99
3.55
0
Electrical Helper
0
0
Contact most of
the time
Very high responsibility
for others
Very high
consequence of error
4.3
3.95
3.7
0
Athletic Trainer
0
0
Constant contact
Very high responsibility
for others
Very high
consequence of error
4.85
4.12
3.55
0
Pipelayer Contractor
0
0
Constant contact
Very high responsibility
for others
Moderate
consequence of error
4.67
4.44
3.46
0
Iron/Steel Contractor
(Commercial)
0
0
Contact most of
the time
Very high responsibility
for others
Very high
consequence of error
4.39
4.33
3.57
0
Weigher
0
0
Contact most of
the time
Moderate responsibility
for others
Very high
consequence of error
4.34
2.71
3.53
0
Sheet Metal Contractor, HVAC
(Commercial)
0
0
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
4.28
3.74
3.28
0
Conveyor Operator
0
0
Contact about
half the time
Very high responsibility
for others
Moderate
consequence of error
3.43
4.17
3.28
0
Carpenter/Cabinet Maker
Contractor (Residential)
0
0
Contact most of
the time
Very high responsibility
for others
Some consequence of
error
4.37
4.22
2.49
0
Drywall Installation Contractor
(Residential)
0
0
Contact most of
the time
Moderate responsibility
for others
Moderate
consequence of error
3.69
3.1
2.79
0
Sheet Metal Contractor, Other
(Commercial)
0
0
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
4.28
3.74
3.28
0
Drywall Installation Contractor
(Commercial)
0
0
Contact most of
the time
Moderate responsibility
for others
Moderate
consequence of error
3.69
3.1
2.79
0
Floor Sander Contractor
(Commercial)
0
0
Contact about
half the time
Moderate responsibility
for others
Moderate
consequence of error
3.31
3.29
2.5
0
Glazier Contractor
(Commercial)
0
1
Contact most of
the time
Very high responsibility
for others
Very high
consequence of error
4.39
3.82
3.74
0
Mason Contractor (Commercial)
0
0
Contact most of
the time
Moderate responsibility
for others
Moderate
consequence of error
3.79
3.15
2.5
0
Wildlife Control Operator
0
1
Constant contact
Very high responsibility
for others
Very high
consequence of error
4.67
4.23
4.26
0
Fire Alarm Installer
0
1
Constant contact
Very high responsibility
for others
Very high
consequence of error
4.55
3.8
3.61
0
Cement Finishing Contractor
(Commercial)
0
0
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
3.52
4.34
3.05
0
Veterinary Technician
0
1
Constant contact
Very high responsibility
for others
Extremely high
consequence of error
4.9
4.12
4.66
0
Crane Operator
0
0
Constant contact
Extremely high
responsibility for others
Very high
consequence of error
4.55
4.73
3.99
0
Farm Labor Contractor
0
0
Contact most of
the time
Extremely high
responsibility for others
Moderate
consequence of error
4.49
4.58
2.5
0
Animal Breeder
0
0
Contact most of
the time
Moderate responsibility
for others
Moderate
consequence of error
3.68
3.31
2.63
0
School Bus Driver
0
0
Contact most of
the time
Moderate responsibility
for others
Very high
consequence of error
4.22
3.34
3.86
0
Carpenter/Cabinet Maker
Contractor (Commercial)
0
0
Contact most of
the time
Very high responsibility
for others
Some consequence of
error
4.37
4.22
2.49
0
Earth Driller, Water Well
0
0
Contact most of
the time
Very high responsibility
for others
Very high
consequence of error
3.99
3.72
3.95
0
Still Machine Setter, Dairy
Equipment
0
0
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
4.26
3.55
3.01
0
Security Alarm Installer
0
1
Constant contact
Very high responsibility
for others
Very high
consequence of error
4.55
3.8
3.61
0
Emergency Medical Technician
0
1
Constant contact
Extremely high
responsibility for others
Extremely high
consequence of error
4.59
4.72
4.59
0
Terrazzo Contractor
(Commercial)
0
0
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
3.55
3.8
3.22
0
Teacher Assistant, Non-
Instructional
0
0
Constant contact
Moderate responsibility
for others
Some consequence of
error
4.72
2.51
1.97
0
Insulation Contractor
(Commercial)
0
0
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
3.72
3.59
2.63
0
Psychiatric Aide
0
0
Contact most of
the time
Very high responsibility
for others
Moderate
consequence of error
4.37
4.45
2.98
0
79 Mitchell Watt and Hubert Wu
Appendix D – Example database profile
The following is an example of an entry containing information about the operation of trust mechanisms
on a platform that could be provided on a business regulator’s website.
Upwork
www.upwork.com
A global freelancing platform where businesses and
independent professionals connect and collaborate
remotely
How does Upwork enable trust on their platform?
Upwork collects reviews from freelancers and clients
about their experiences with other users on the
platform. Only transacting parties can review other
parties. These reviews are optional, but incentivized.
What type of information does the platform collect
on transactions?
Upwork’s users can submit written feedback and star
ratings, once per transaction. Freelancers may also
take tests on skills which may be displayed on their
profile.
Star ratings are scored out of 5, and aggregated into a
job success score between 0 and 100.
How does Upwork use user feedback?
Feedback about freelancers is used to create a ‘job
success score’ which affects the placement of
freelancers in searches. Highly rated freelancers may
be promoted in ‘suggested freelancers’ emails and
promotions to clients. Freelancers who have completed
Upwork tests also have badges showing their
competencies as evaluated by Upwork.
Feedback about clients is used privately by Upwork to
prevent misbehavior on their platform and may be used
to remove clients from Upwork.
How is user feedback filtered or altered by Upwork?
There is a dispute resolution system where freelancers
may contest reviews, which is moderated by Upwork.
‘Top performers’ as determined by Upwork are also
able to remove one review of the platform. Only recent
reviews (6, 12 or 24 months) are used to calculate
average ratings and job success scores.
Things to keep in mind on Upwork:
• For freelancers, reviews may affect your
placement in searches on the site and
therefore your probability of being hired.
• Some reviews may have been removed by
Upwork or their clients. Be aware while
making hiring decisions on the platform.
• Eligibility for skill badges is assessed by
Upwork and so should not be treated like
independently verified qualifications.
Appendix E – Design guidelines for businesses
The style, tone, and formatting of these guidelines are based on existing informational publications by the
Federal Trade Commission and other business regulators.
Building Trust on Your Online Platform: Design
Tips for Reputation and Review Systems
Customer reviews, ratings, guarantees, verification, and other reputation and “trust
mechanisms” are increasingly important in the digital economy. We have tips to help your
company make the most of these and comply with the law.
Reputation and review systems are an important way of building trust and credibility in the digital
economy. Is your company doing the best it can in this area?
When used as intended, reviews, ratings, and other tools that build trust and reputation can
increase sales, consumer welfare, and allow companies to engage with their target customers in
meaningful and bespoke ways. But these systems can sometimes have unintended consequences
that can harm both businesses and consumers.
Here are some tips for getting the most out of them, and complying with the law.
WHAT ARE ONLINE PLATFORMS (AND ARE THEY RELEVANT TO
MY COMPANY)?
Online platforms are businesses that create value by facilitating exchanges between distinct groups
via the web. Today, the majority of the world’s most valuable companies and well-known brands
operate in this way.
But you don’t have to be a billion-dollar business for the idea to be relevant to your company. If you
conduct any type of e-commerce, have a listing of your business online, or interact with customers
via the web in any way, then this article may be relevant for you. Consumers may also find this
article helpful. But we also have a separate guide for you here.
WHAT ARE REPUTATION AND REVIEW SYSTEMS?
Reputation and review systems are ways to discern information about factors such as product
quality, service standards, customer/business reliability, safety, and the risk of fraud.
These systems can take many forms. Examples include star ratings, written reviews, money-back
guarantees, identity verification systems, and skill tests. They are important because without them,
platform users may the lack the confidence to engage in online interactions. This is be a particularly
important problem online because of user anonymity.
WHAT ARE SOME COMMON PROBLEMS WITH THESE
SYSTEMS?
Harm to consumers and businesses can result from misleading information, which can result from:
Deliberately manipulated reviews and ratings
Responses from a non-representative or ideal pool of responders
Bias against certain groups and demographics
Inflated or not up-to-date ratings
Insufficient ratings which can send a signal of poor quality
Many of these problems can be addressed or mitigated from considered design choices.
WHAT CAN ONLINE PLATFORMS DO TO GET THE MOST OUT OF
THEIR REPUTATION AND REVIEW SYSTEMS?
In designing reputation and rating systems for your business, here are five principles to keep in
mind.
Know what your goal is. Reputation systems can serve different purposes. Is yours to
signal quality? To allow users find specific products that suit them? To promote safety?
There are many ways to generate trust. For example, numerical ratings may be
complemented with tags and badges, and guarantees from your company.
Different parties may have different incentives. Not all of these may be “to be honest.”
It’s not always a good idea to copy other companies. Recent research has found that
different systems can be suited to different types of companies. And even within industries,
trust mechanisms can be imperfect even if they are common.
And most importantly:
Let people speak honestly about your products and their experience with your
company.
To put these principles into practice, we recommend that you keep the following questions in mind
when designing reputation and rating systems:
Category
Question
For example
Participants
Who participates?
Who has access?
Is participation required?
Businesses and customers
Only transacting parties
Optional vs. mandatory
Content
What is the format and scale used?
Are there subcategories for ratings?
Star ratings out of five
Timeliness, value, location
Function
Who may view the content?
Is information anonymous?
Are responses weighted?
How is information used by the platform?
Can the credibility of responses be checked?
Public vs. private
Anonymous vs. identifiable
Recent X number of reviews
Improvements to the platform
Detecting anomalous reviewers
