Order without Law: Reputation Promotes Cooperation in a Cryptomarket for Illegal Drugs

Abstract

The emergence of large-scale cooperation in humans poses a major puzzle for the social and behavioural sciences. Reputation formation—individuals’ ability to share information about others’ deeds and misdeeds—has been found to promote cooperation. However, these findings are mostly based on small-scale laboratory and field experiments or on data gathered from online markets embedded in functioning legal systems. Using a unique data set of transactions in a cryptomarket for illegal drugs, we analyse the effect of buyers’ ratings of finished transactions on sellers’ business success. Cryptomarkets are online marketplaces in the so-called Dark Web, which can only be accessed by means of encryption software that conceals users’ identities and locations. The encryption technology makes it virtually impossible for law enforcement to intervene in these market exchanges. We find that sellers with a better rating history charge higher prices and sell their merchandise faster than sellers with no or a bad rating history. Our results demonstrate how reputation creates real incentives for cooperative behaviour at a large scale, in the absence of law enforcement and among anonymous actors with doubtful intentions. Our results thus challenge the institutional and social embeddedness of actors as necessary preconditions for the emergence of social order in markets.

Full text

Order without Law: Reputation Promotes
Cooperation in a Cryptomarket for Illegal Drugs
Wojtek Przepiorka
1,2,
*, Lukas Norbutas
1,3
and Rense Corten
1
1
Department of Sociology / ICS, Utrecht University, Padualaan 14, Utrecht, 3584 CH, The Netherlands,
2
Nuffield College, New Road, Oxford OX1 1NF, UK and
3
Netherlands Institute for the Study of Crime and
Law Enforcement, De Boelelaan 1077a, Amsterdam 1081 HV, The Netherlands
*Corresponding author. E-mail: w.przepiorka@uu.nl
Submitted March 2017; revised July 2017; accepted September 2017
Abstract
The emergence of large-scale cooperation in humans poses a major puzzle for the social and behav-
ioural sciences. Reputation formation—individuals’ ability to share information about others’ deeds
and misdeeds—has been found to promote cooperation. However, these findings are mostly based
on small-scale laboratory and field experiments or on data gathered from online markets embedded
in functioning legal systems. Using a unique data set of transactions in a cryptomarket for illegal
drugs, we analyse the effect of buyers’ ratings of finished transactions on sellers’ business success.
Cryptomarkets are online marketplaces in the so-called Dark Web, which can only be accessed by
means of encryption software that conceals users’ identities and locations. The encryption technology
makes it virtually impossible for law enforcement to intervene in these market exchanges. We find
that sellers with a better rating history charge higher prices and sell their merchandise faster than sell-
ers with no or a bad rating history. Our results demonstrate how reputation creates real incentives for
cooperative behaviour at a large scale, in the absence of law enforcement and among anonymous
actors with doubtful intentions. Our results thus challenge the institutional and social embeddedness
of actors as necessary preconditions for the emergence of social order in markets.
Introduction
Humans’ ability to overcome individual self-interest to
create a larger benefit for the collective has received con-
siderable attention in the social and behavioural sciences
in the past three decades (Axelrod, 1984;Nowak, 2006;
Bowles and Gintis, 2011). A simple mechanism that has
been shown to promote cooperation in humans is our abil-
ity to share information about our peers’ deeds and mis-
deeds with third parties. Such information sharing
contributes to the formation of individuals’ reputations
(Dunbar, 2004;Sommerfeld et al., 2007;Feinberg, Willer
and Schultz, 2014). Individuals with a reputation to lose
have a strong incentive to behave cooperatively and are
therefore attractive partners in social and economic
exchange (Shapiro, 1983;Kollock, 1994;Sylwester and
Roberts, 2013;Milinski, 2016). More generally, it has
been argued that reputational incentives provide a more
efficient means to uphold norm compliance and order in
society than other forms of sanctioning (Ellickson, 1991;
Milinski, Semmann and Krambeck, 2002;Willer, 2009;
Grimalda, Pondorfer and Tracer, 2016;Wu, Balliet and
van Lange, 2016).
Throughout human history, reputation mechanisms
have also facilitated mutually beneficial economic
V
CThe Author 2017. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com
European Sociological Review, 2017, Vol. 33, No. 6, 752–764
doi: 10.1093/esr/jcx072
Advance Access Publication Date: 17 October 2017
Original Article
Downloaded from https://academic.oup.com/esr/article-abstract/33/6/752/4555481 by guest on 19 November 2018

exchanges. In the absence of formal institutions protecting
property rights and enforcing contractual commitments,
the transfer of reputation information through dense social
networks has promoted good business conduct (Hillmann,
2013). For example, Greif (1989) describes how Maghribi
traders in medieval Europe organized in coalitions in which
they exchanged information about their agents’ reputations
to reduce the trust problems arising in long-distance trade.
Hillmann and Aven (2011) describe the role reputation
played in the development of corporate capitalism in
Russia around the turn of the nineteenth century. At that
time, entrepreneurs faced the trade-off of limiting their
interactions to local partners to ensure compliance and
interacting with partners beyond their social networks
offering more profitable but riskier businesses.
These historical examples illustrate how the lack of for-
mal institutions promoting economic exchange is replaced
by informal institutions at work in social groups in which
information about actors’ deeds and misdeeds is exchanged
and selective incentives upheld (Nee 2005). However, there
are several historical accounts of centralized reputation sys-
tems which facilitated economic exchange without requir-
ing actors to be closely connected via a social network.
Also in the early middle ages, the Champagne Fairs in
France became a meeting point for traders from all over
Europe. Promoted by the use of bookkeeping and cashless
payment, a private adjudication system evolved that
allowed tracking fraudulent traders and excluding them
from future fairs (Milgrom et al., 1990). In the late nine-
teenth century, so-called credit bureaus started to emerge,
which collected and shared information about borrowers’
credit histories creating reputational incentives for timely
debt repayment (Jappelli and Pagano, 2002;Carruthers,
2013). These examples are prototypical for the centralized
reputation systems that are today’s standard for governing
online market exchanges (Dellarocas, 2003;Diekmann
and Przepiorka, forthcoming).
In online markets such as eBay, thousands of anony-
mous buyers and sellers trade with each other every day
across large geographic distances. Via an electronic feed-
back system, traders can comment on each other’s conduct
after finished transactions with positive or negative ratings
and short text messages, and these ratings constitute online
traders’ reputations. The advent of the Internet and the
emergence of online markets have created ample opportu-
nities to study the effectiveness of reputation systems to
promote cooperation among anonymous traders at a large
scale (Kollock, 1999;Resnick and Zeckhauser, 2002;
Dellarocas, 2003;Bolton, Greiner and Ockenfels, 2013;
Diekmann et al., 2014). In particular, it has been shown
how electronic reputation systems create incentives for
traders’ cooperative behaviour without requiring these
traderstobeembeddedinsocialnetworks(Granovetter,
1992;Diekmann et al., 2014). However, the working of
the reputation mechanism has thus far only been estab-
lished under favourable conditions. The majority of online
markets are embedded in functioning legal systems attract-
ing and backing up trades among individuals with mostly
good intentions. It is thus an open question whether repu-
tation formation net of legal and moral assurances is suffi-
cient to promote cooperation in a large group of strangers.
Here we study the functioning of the reputation mech-
anism in a cryptomarket for illegal drugs. Cryptomarkets
are online marketplaces in the Dark Web, which can only
be accessed by means of encryption software that con-
ceals users’ identities and locations (Martin, 2014).
Trades in cryptomarkets include forged personal docu-
ments, hacked user accounts, weapons, etc., with illegal
drugs constituting the largest proportion of trades
(Christin, 2012;Soska and Christin, 2015). Globally,
almost 10 per cent of drug users reported ever buying
drugs from cryptomarkets (Global Drug Survey, 2016).
The encryption technology makes it virtually impossible
for law enforcement to intervene. Hence, given the lack
of legal deterrent, traders’ good intentions are highly
uncertain at best. This creates severe trust problems
between buyers and sellers, as it makes buyers vulnerable
to sellers’ fraudulent transactions (Dasgupta, 1988;
Coleman, 1990). However, in the same way as online
markets for licit goods, cryptomarkets use electronic rep-
utation systems which allow buyers to rate sellers after
finished transactions (Bartlett 2014;Hardy and
Norgaard, 2016). This makes cryptomarkets the ideal set-
tings to test the potential of the reputation mechanism to
bring about ‘order without law’ (Ellickson, 1991).
1
In the next section, we describe the reputation mech-
anism, previous approaches to its studying, the set-up of
the cryptomarket that we study, and state our hypothe-
ses. In the ‘Data and Methods’ section, we describe the
data and data gathering process, the variables we used
in our analyses, and our model estimations. Thereafter
we present our results and, in the final section, we con-
clude with a discussion of our findings.
Reputation in Markets and the Problem of
Embeddedness
Acquiring a good seller reputation in online (and offline)
markets is costly because it can only be achieved
through good business conduct over an extended period
of time. Sellers who lack trustworthy-making properties,
such as long-term business interests or honest intentions,
will not bother to enter the market and build a good rep-
utation by behaving cooperatively. Hence, based on
European Sociological Review, 2017, Vol. 33, No. 6 753
Downloaded from https://academic.oup.com/esr/article-abstract/33/6/752/4555481 by guest on 19 November 2018

sellers’ reputations, buyers can infer these sellers’ trust-
worthiness and choose the sellers they prefer to buy
from. However, trustworthy sellers who enter the mar-
ket and, therefore, have not yet built their reputation,
are indistinguishable from their untrustworthy competi-
tors. New, trustworthy sellers must therefore allow pri-
ces to compensate potential buyers for the risk they take
when trading with ‘unknown’ sellers. Once these sellers
have built their reputations, they can charge higher pri-
ces, which will compensate them for their initial invest-
ment in reputation (Shapiro, 1983;Friedman and
Resnick, 2001;Przepiorka, 2013).
2
From this reasoning
it follows that sellers with a better reputation will
achieve higher prices because buyers are willing to pay
higher prices for these sellers’ products.
This conjecture has been corroborated in more than
two dozen studies, most of which analyse the effect of
sellers’ reputations on the probability of product sale
and selling price using eBay auction data (for a review,
see Diekmann et al., 2014). Although the costliness of
building a good reputation constitutes an important
deterrent for fraudulent sellers, the legal system in which
online markets are embedded deters fraud and promotes
trust and large-scale cooperation in its own right
(Diekmann and Przepiorka forthcoming; Fligstein,
2001;Gu¨ th and Ockenfels, 2003;Pavlou and Gefen,
2004;Beckert, 2009;Bakos and Dellarocas, 2011).
First, the legal system maintains a non-negligible threat
that fraudulent business conduct will be prosecuted and
punished. Secondly, online market platforms can be
made accountable for sellers’ misconduct by their com-
munity of buyers, who can easily turn to alternative
platforms. As a consequence, platform providers have a
strong incentive to protect buyers from fraud by, for
example, monitoring sellers’ activities and sanctioning
bad behaviour (e.g. by banning sellers). However, unlike
online markets for licit goods, cryptomarkets cannot
work with law enforcement to combat fraudulent
behaviour (Calkins et al., 2008). Thirdly, buyers are
insured against fraud to a certain extent if they use credit
card payment. Although such insurances do not elimi-
nate the trust problem, they reduce the material losses
buyers may expect when trading in online markets. In
sum, legal assurances preselect sellers with good inten-
tions, incentivize platform providers to enforce coopera-
tion, and reduce the risk of large monetary losses. As a
consequence, buyers will have high a priori expectations
as to online sellers’ trustworthiness (Gu¨ th and
Ockenfels, 2003;Lindenberg, 2017).
3
In the light of these considerations, it is an open ques-
tion whether the reputation mechanism, as instituted in
many electronic rating systems, promotes trust and
cooperation in online markets net of their embeddedness
in well-functioning legal systems. Backed by legal and
moral assurances, reputation may function as a mere
coordination device, which facilitates buyers’ choices
among the plethora of sellers offering the same products,
rather than solving cooperation problems (Beckert,
2009;Frey and van de Rijt, 2016;Przepiorka and
Aksoy, 2017). More importantly, in the absence of legal
and moral assurances, reputation systems may fail to
attract a critical mass of traders because they may be
regarded as an insufficient safeguard of mutually benefi-
cial economic exchange.
One way to address this question is to study the func-
tioning of the reputation mechanism in an extra-legal
context, for example, as we do, in a cryptomarket for
illegal goods. If we find reputation effects even in the
absence of legal and social conditions that deter oppor-
tunistic actors and promote trust, this would strongly
reinforce the idea that reputation systems enable the
bottom-up emergence of cooperation in large groups of
self-regarding actors. We thus re-evaluate the claim
from earlier research that reputation affects market out-
comes of sellers in the context of cryptomarkets for ille-
gal drugs, and derive hypotheses for our specific study
context, the Cryptomarket Silk Road 1.0.
4
The Cryptomarket Silk Road 1.0
We use data from the first cryptomarket, Silk Road 1.0
(Christin, 2012), to study in how far sellers reap the ben-
efits of a good reputation and in how far buyers take
into account sellers’ reputations when deciding which
seller to buy from. Silk Road 1.0 started operating in
February, 2011 and was closed after its owner was
arrested in October 2013 (Barratt and Aldridge, 2016).
Our data contain information on all item listings that
were online between 3 February 2012 and 24 July 2012,
including item names and descriptions, categories, pri-
ces, and item-specific feedback messages. Since each
item listing contains an encrypted vendor identifier,
feedback can be attributed to individual sellers who
were active during this period.
A typical transaction on Silk Road is initiated by the
seller, who decides on the number of items and the item
price of his or her product, and posts the offer online.
Buyers can then buy the item at the specified price as
long it is available. A buyer first sends the money to an
escrow service, which releases the money and transfers it
to the seller when the buyer confirms receipt of the item.
Although the escrow service mitigates the trust problem
to a certain extent as it protects the buyer from spending
the money without receiving anything in return, the trust
754 European Sociological Review, 2017, Vol. 33, No. 6
Downloaded from https://academic.oup.com/esr/article-abstract/33/6/752/4555481 by guest on 19 November 2018

problem remains with regard to the quality of the prod-
uct and the competence of the seller. Buyers cannot
withhold payment because of quality reasons, as this
would require using or testing the product. Moreover,
there is more to a transaction than the money and the
product. A seller’s good reputation also stands for his or
her professional handling of the transactions. Since
buyers have to provide a postal address the seller can
send the product to, buyers have to trust the seller to
wrap the product inconspicuously, send it to the right
address, and maybe provide a refund should the product
get lost on its way to the buyer (Bartlett 2014: Ch. 5).
In Silk Road 1.0, a finished transaction receives the
highest rating (i.e. five stars) by default. Buyers can
change the default rating to a four-, three-, two-, or one-
star rating and add a text comment. The sum of these
seller ratings establishes a seller’s reputation in the mar-
ket. The information about a seller’s reputation is con-
spicuously displayed and can be considered by buyers
deciding which seller to buy from. We use the sum of a
seller’s five-star ratings and non-five-star ratings as an
indicator of this seller’s reputation, and keep track of
sellers’ rating history over time (see ‘Data and Methods’
section). We use item prices, which are set by the sellers,
as an indicator for sellers’ cashing in the premium for
their good reputation or giving a discount in case of no
or a bad rating history. We use the speed at which items
are sold, which is determined by buyers, as an indicator
for the trust buyers have in sellers with a certain rating
history. Based on the theoretical argument in the first
paragraph of the previous section and the set-up of Silk
Road 1.0, we can formulate the following four
hypotheses:
H1: The more five-star ratings a seller has, the more he
or she will charge for his or her items.
H2: The more non-five-star ratings a seller has, the less
he or she will charge for his or her items.
H3: The more five-star ratings a seller has, the more
items he or she will sell per day.
H4: The more non-five-star ratings a seller has, the less
items he or she will sell per day.
Data and Methods
Data
We use a data set containing 24,385 items collected by
Christin (2012) on Silk Road 1.0 between 3 February
2012 and 24 July 2012. To avoid bias due to unobserved
item heterogeneity, we select a subset of illegal drug
items for our analyses (Diekmann et al., 2014). Some
types of drugs (e.g. LSD) are sold in different forms
(e.g. pills, powder, blotter), which vary in weight and
substance concentration, making the calculation and
comparability of item prices per gram more difficult.
Our subset comprises seven categories of illegal drugs:
weed, hash, cocaine, ketamine, MDMA, heroin, and
meth. We limit our analyses to these categories because
of their size in terms of the number of item listings and
the comparability of items within each category. Item
listings in these categories account for 24.6 per cent
(6,005) of all item listings and 37.3 per cent of all feed-
back messages. To this sample we add 211 items listed
in the general categories ‘Drugs’ and ‘Cannabis’, which
we could identify as also belonging to one of the seven
categories specified above. This results in 6,216 items.
Information on item weight is not available in a
standardized form in the original data set. Item weight
information is only provided as part of the item name or
item description. We extract item weight from the item
name or description manually and exclude items which
have no information on weight or have weight informa-
tion not comparable with the majority of items in the
category (e.g. pre-rolled joints of marijuana in category
‘weed’, where the majority of items are sold in grams of
loose marijuana). At this stage we exclude 430 items.
Moreover, we exclude 111 items that are given away
(e.g. as ‘freebies’, ‘samples’, or ‘lotteries’) or are included
as custom listings for a specific buyer. This leaves us
with 5,675 items offered by 550 different sellers.
Of these 5,675 items, 2,522 (44.4 per cent) received
no feedback messages during the time of data collection.
Since we use the number of feedback messages an item
received in total as a measure for the number of item
sales (see below), we have no evidence that these items
generated any sales. Note that 67 per cent of these items
were online for 5 days or less, whereas only 11 per cent
of the items for which at least one sale was recorded
were online 5 days or less. In other words, a large major-
ity of the items without recorded sales were online for a
relatively short time. We cannot be sure that the sellers
posting these 2,522 items had the intention to sell them
as specified. These sellers may have offered these items
but not reacted to any buyer requests or taken the items
offline before completing any transactions. We therefore
exclude these items from our main analyses. Our main
analyses are based on 3,153 items, for each of which at
least one sale was recorded. These items were offered by
445 different sellers. However, in the online
Supplementary Material we also discuss the results of
logistic regressions that are based on all 5,675 items and
estimate the probability that at least one sale was
European Sociological Review, 2017, Vol. 33, No. 6 755
Downloaded from https://academic.oup.com/esr/article-abstract/33/6/752/4555481 by guest on 19 November 2018

recorded conditional on covariates. The results of this
analysis are consistent with the results reported here.
Variables and Model Estimations
The unit of analysis is an item offered by a seller. Our
two target variables are item price per gram (in USD)
and number of item sales per day. We calculate item
price per gram by dividing item price by item weight.
Since no direct information on the number of sales per
item is available in Silk Road 1.0, we derive the number
of item sales from the total number of ratings a seller
received for a particular item. Since a five-star rating is
automatically awarded after a finished transaction, this
approach provides an accurate estimate of the number
of item sales (Soska and Christin, 2015;De´cary-He´tu,
Paquet-Clouston and Aldridge, 2016). To obtain the
number of item sales per day, we divide the number of
item sales by the time in days the item was observed
online.
To test Hypotheses 1 and 2, we use a regression model
with the log-transformed item price per gram (in USD) as
the dependent variable; to test Hypotheses 3 and 4, we
use a regression model with the log-transformed number
of sales per day as dependent variable. In both models,
we use the same explanatory and control variables with
two exceptions. In our model of item price, we use the
log number of total item sales as a measure for the quan-
tity of an item initially in the seller’s possession. Recall
that item prices are set by the sellers. Since sellers are
likely to pass on quantity discounts to their customers,
the quantity of an item in the seller’s possession will nega-
tively affect item price. For obvious reasons, we do not
use the log number of item sales as an explanatory varia-
ble in our models of sales per day, but we use log item
price per gram for it will negatively affect item sales per
day.
Our main explanatory variables are sellers’ positive
and negative reputation scores. We calculate a seller’s
reputation scores by summing up the number of five-star
ratings and non-five-star ratings the seller has received
by the time his or her item is first listed online. Note that
we aggregate the five-star and non-five-star ratings
across all items of a seller, also the items not included in
our analyses. This is done to replicate the aggregated
seller reputation score displayed in Silk Road 1.0 as
closely as possible. We use the log number of five-star
ratings as a measure of positive seller reputation and the
log number of non-five-star ratings as a measure of neg-
ative seller reputation. Five-star ratings account for 95.8
per cent of all ratings in our sample; non-five-star rat-
ings are given in extraordinary cases and might have
negative impact on a seller’s reputation irrespective of
the actual number of stars.
We include several control variables in our models.
To reduce the number of control variables, we pool the
seven categories of illegal drugs in three price categories:
Low price (weed and hash), medium price (cocaine, ket-
amine, and MDMA), and high price (heroin and meth).
We use a set of three dummy variables to account for
the three different categories with low price items as the
reference category. Items in Silk Road 1.0 are offered in
quantities ranging from 0.05 g to 1,000 g (see Table 1).
We use log-transformed item weight (in grams), which
we expect to have a negative effect on item price because
of quantity discounts offered by sellers. Item weight will
also have a negative effect on the number of sales per
day because of a lower demand for bulk offers (Aldridge
and De´cary-He´tu, 2016). We also use dummy variables
for shipping locations of each item as offered by the sell-
ers. Since sellers who ship their items internationally
face additional risks (De´cary-He´tu et al., 2016), these
sellers might charge a price premium. For most sellers
we have information on their country of origin and the
countries and regions these sellers ship their items to.
Based on this information, we distinguish between sell-
ers who only ship domestically (i.e. within their country
of origin), sellers who also ship their items abroad and
sellers with unknown shipping preferences. Sellers who
only ship domestically constitute the reference category.
One dummy variable accounts for items of poor quality,
an attribute sometimes specified in the titles of low price
items only. Finally, since the data are right censored, we
include a dummy variable to mark items that were listed
online on the last 2 days of data collection. Table 1 con-
tains the descriptive statistics of the main variables used
in our analyses.
We do not have information on sellers’ ratings from
before the start of data collection. We therefore calcu-
late sellers’ reputations by summing up the number of
five-star ratings and non-five-star ratings these sellers
receive throughout the observation period. Since we per-
form our data analysis at the item level, we use the sum
of ratings a seller received by the time an item was first
listed in the market as an indicator of the seller’s reputa-
tion (also see above). In other words, our data allow us
to capture the change in sellers’ reputations over time;
our ability to compare sellers based on their number of
five-star and non-five-star ratings is rather limited. We
therefore estimate regression models with seller fixed
effects (FE), which are solely based on the within-seller
variation in dependent and explanatory variables
(Allison, 2009). These models estimate how changes in
seller’s reputation, compared to seller’s average
756 European Sociological Review, 2017, Vol. 33, No. 6
Downloaded from https://academic.oup.com/esr/article-abstract/33/6/752/4555481 by guest on 19 November 2018

reputation across all observations, affect changes in sell-
er’s item prices and number of sales.
Results
Table 2 shows the regression model estimations with
seller FE. In both the model of item price (M1) and the
model of sales per day (M2), the coefficient estimates of
the log number of five-star ratings are positive and the
coefficient estimates of the log number of non-five-star
ratings are negative. The four coefficients are statisti-
cally significant providing first evidence in support of
our hypotheses.
Recall from Table 1 that the range in the number of
five-star ratings in our data is considerably larger than
the range in the number of non-five-star ratings. In what
follows, we use 10-fold increases in five-star ratings (e.g.
50 vs. 500) and 3-fold increases in non-five-star ratings
(e.g. 7 vs. 21) to calculate the effect of sellers’ rating his-
tories on item price and selling speed.
In Model M1, if the number of five-star ratings
increases by a factor 10, sellers increase the item price
by 100 [exp (0.029 ln(10)) 1] ¼6.8 per cent, and
if the number of non-five-star ratings increases by a fac-
tor 3, sellers decrease the item price by 100 [exp
(0.044 ln(3)) 1] ¼() 4.7 per cent. Based on the
average selling price of a medium-price item of USD
92.26 (see Table 1), these changes correspond to USD
6.29 and USD 4.35, respectively. These results clearly
support our first two hypotheses (H1 and H2) and show
that sellers’ price adjustments in response to changes in
their reputation can be substantial. These results are
visualized in Figure 1A.
The coefficients of the control variables point in the
expected directions. The higher its product category and
the higher its quality, the higher is the price the seller asks
for the item. The negative and statistically significant
coefficients of the log number of sales and item weight
both indicate that sellers give quantity discounts to
buyers. Although sellers who also ship their items abroad
tend to charge higher prices than sellers who only ship
domestically, the difference is statistically insignificant.
Setting the price of an item is a deliberate choice
made by sellers. Therefore, changes in item price do not
directly tell us whether buyers infer trustworthiness
from sellers’ reputations and act accordingly. Given sell-
ers’ reputations and item characteristics, it is mainly the
buyers who determine how quickly an item sells. Selling
speed thus constitutes a better measure for how sellers’
reputations help buyers to overcome the trust problem.
In Model M2, we find similar results in terms of item
sales per day, as we found in our model for item price
(M1). The coefficient estimates of the log number of
five-star ratings and log number of non-five-star ratings
Table 1. The table lists descriptive statistics of the main variables used in our analyses
Variable name NMean SD Median Minimum Maximum
Item sales and duration online
# item sales 3,153 20.72 58.94 5 1 1501
item online in days 3,153 50.44 56.17 28 0.5 382
# item sales per day 3,153 0.45 0.69 0.25 0.01 10.83
Seller ratings at time item was first seen
# five-star ratings 3,153 148.3 279.9 43 0 2615
# non-five-star ratings 3,153 4.89 12.36 0 0 149
Low-price products (weed, hash)
weight in g 2,297 18.15 63.57 5 0.25 1000
price in USD per gram 2,297 15.50 7.30 14.61 1.46 115.8
Medium-price products (ketamine, MDMA, cocaine)
weight in g 562 7.17 45.00 1 0.05 1000
price in USD per gram 562 92.26 57.52 80.58 8.41 464.1
High-price products (meth, heroin)
weight in g 294 1.40 3.87 0.5 0.10 56
price in USD per gram 294 217.6 140.4 173.8 33.72 992.8
Note: The data comprise N¼3153 items which were online for 50 days and generated 21 sales on average. Only items for which at least one transaction was
recorded are considered in this analysis (see ‘Data and Methods’ section). For each item, we calculated the number of five-star and non-five-star ratings the seller of an
item had received up to the time point the item was first seen online. The seller of an item has 148 five-star and 5 non-five-star ratings on average. We divided items in
three price categories. Low-price items (N¼2297) comprise weed and hash and are sold for USD 16 per gram in packages of 18 g on average. Medium-price items
(N¼562) comprise ketamine, MDMA, and cocaine, and are sold for USD 92 per gram in packages of 7 g on average. High-price items (N¼294) comprise meth and
heroin, and are sold for USD 218 per gram in packages of 1 g on average.
European Sociological Review, 2017, Vol. 33, No. 6 757
Downloaded from https://academic.oup.com/esr/article-abstract/33/6/752/4555481 by guest on 19 November 2018

can be interpreted as follows. If the number of five-star
ratings increases by a factor 10, item sales per day
increase by 100 [exp (0.060 ln(10)) 1] ¼14.7 per
cent, and if the number of non-five-star ratings increases
by a factor 3, item sales per day decrease by 100 [exp
(0.188 ln(3)) 1] ¼()18.7 per cent. Based on these
changes, a median seller, who sells one item in 4 days
(see Table 1), would need 3.5 or 5 days, respectively, to
sell that item. These results clearly support hypotheses
H3 and H4. These results are visualized in Figure 1B.
The control variables exhibit the same effects as in our
model of item price. Unsurprisingly, item price has a sig-
nificantly negative effect on the number of sales.
The effects of sellers’ reputation on selling speed
appear relatively small. To a large extent, this is a result
of our accounting for differences in unobserved, time-
constant seller characteristics in our model estimations
(Allison, 2009). However, buyers choosing a seller com-
pare offers from different sellers (Snijders and Weesie,
2009). Therefore, we also estimated ordinary least square
regression models, which take between-seller variability
of seller and item characteristics into account (see online
Supplementary Material). Based on these estimations, the
effects of five-star ratings on selling speed are up to three
times larger than the effects shown in Figure 1B, but these
effects can only partly be interpreted causally.
Table 2. Regression models of item price and sales per day with seller FE
Variable name
Log(item price per gram in USD) Log(# item sales per day)
M1 M2
Const. 2.974*** 1.402***
(0.036) (0.259)
Item variables
Log(item price per gram in USD) 0.823***
(0.089)
Log(weight in gram) 0.208*** 0.502***
(0.009) (0.079)
Low price (reference) (reference)
Medium price 1.506*** 0.908***
(0.065) (0.218)
High price 2.035*** 1.271***
(0.095) (0.310)
Poor quality (weed and hash) 0.756*** 0.190
(0.078) (0.196)
Last 2 days 0.114*** 0.174*
(0.027) (0.077)
Seller variables
Log(# five-star ratings þ1) 0.029*** 0.060*
(0.007) (0.028)
Log(# non-five-star ratings þ1) 0.044** 0.188***
(0.016) (0.057)
Log(# item sales) 0.035***
(0.006)
Seller ships to
Unknown 0.012 0.006
(0.064) (0.207)
Domestic only (reference) (reference)
Foreign 0.035 0.063
(0.036) (0.130)
N
1
3153 3153
N
2
445 445
Adjusted R
2
0.77 0.22
Notes: The table lists coefficient estimates and cluster–robust standard errors (***P<0.001, **P<0.01, *P<0.05, for two-sided tests) of FE regression models.
The target variable in Model M1 is the log-transformed item price per gram in USD. The target variable in Model M2 is the log-transformed number of item sales per
day. N
1
denotes the number of cases (items), and N
2
denotes the number of clusters (sellers). Both models include seller FE. Figures 1 A and 1B are based on Models
M1 and M2, respectively.
758 European Sociological Review, 2017, Vol. 33, No. 6
Downloaded from https://academic.oup.com/esr/article-abstract/33/6/752/4555481 by guest on 19 November 2018

A
B
Figure 1. Shows the changes in item price (A) and selling speed (B) due to chang es in the number of five-star and non-five-star rat-
ings of a seller. In (A), changes in item price are calculated relative to the average price of a low-price, medium-price, and high-price
product; in (B) changes in selling speed are calculated relative to sellers at the 25th, 50th, and 75th percentile in terms of selling
speed. Due to a 10-fold increase in the number of five-star ratings, a seller increases the item price by USD 1.06, USD 6.29, and
USD 14.83 for a low-price, medium-price, and high-price product, respectively. Correspondingly, due to a 3-fold increase in the
number of non-five-star ratings, a seller decreases the item price by USD 0.73, USD 4.35, and USD 10.26 for a low-pric e, medium-
price, and high-price product, respectively. These results support hypotheses H1 and H2: sellers reap the benefits of a good reputa-
tion by increasing prices and give a discount if they have not yet established a good reputation or their reputation decreased due
to non-five-star ratings. Due to a 10-fold increase in the number of five-star ratings, a median seller sells 0.37 items more in 10 days,
and due to a 3-fold increase in the number of non-five-star ratings the median seller sells 0.47 items less in 10 days. These effects
are larger for 75th percentile sellers, who have a higher frequency of sales, and smaller for 25th percentile sellers, who have a
lower frequency of sales than a median seller. These results support hypotheses H3 and H4: Buyers are more eager to buy from
sellers with a good reputation and more reluctant to buy from sellers who have not yet established a good reputation or who have
received non-five-star ratings in the past.
European Sociological Review, 2017, Vol. 33, No. 6 759
Downloaded from https://academic.oup.com/esr/article-abstract/33/6/752/4555481 by guest on 19 November 2018

Discussion
We study how cooperation is established between anon-
ymous individuals in a cryptomarket for illegal goods.
We use longitudinal data on market transactions from
the first major cryptomarket, called ‘Silk Road’, to test
whether reputation formation can promote cooperation
between buyers and sellers in an environment of high
uncertainty and in the absence of a centralized author-
ity. We use all available market transactions in the seven
largest categories of illegal drugs to test whether high
(low) buyer ratings increase (harm) sellers’ market suc-
cess in terms of pricing and sales. We find that sellers’
rating histories affect the behaviour of both sellers and
buyers. Sellers react to changes in their reputation by
adjusting the prices of their goods. Well-reputed sellers
reap market benefits by increasing prices, while sellers
with lower reputations decrease their prices to compen-
sate potential buyers for the risk they take when buying
from them (Shapiro, 1983;Friedman and Resnick,
2001;Przepiorka, 2013). We also find that sellers with
better reputations sell more goods over the same period
of time. Although we do not observe buyers’ choices of
particular sellers, the higher selling speed of well-
reputed sellers’ items suggests that buyers choose sellers
based on these sellers’ rating histories. Finally, we find
that negative ratings have a larger absolute effect on sell-
ers’ prices and sales than positive ratings. Negative
asymmetry, or a large impact of negative information
about partner’s trustworthiness on withholding trust,
has been observed previously in online markets and
experimental settings (Standifird, 2001;Bozoyan and
Vogt, 2016). Our findings suggest that in cryptomarkets
too damaged reputations are hard to repair (Matzat and
Snijders, 2012).
Our research contributes to the agenda of the new
institutionalism in economic sociology (Nee 2005;
Beckert, 2009;Hillmann, 2013) in at least two ways.
First, we show that reputation formation is a robust
mechanism to foster trust and cooperation in online
markets net of legal assurances, verifiable identities, or a
positive self-selection of mostly law-abiding citizens. In
online markets for licit goods, the trust problem inherent
in economic exchanges is mitigated but not entirely dis-
solved by legal and moral assurances. Our results thus
corroborate that reputation systems in general can be an
essential organizational assurance which, if well-
designed, protect online traders from being cheated or in
other ways dissatisfied by their peers. Secondly, in a his-
torical perspective, cryptomarkets constitute a next
phase in the evolution of market institutions. In our
article, we describe a case that illustrates how
cooperative market exchanges are possible at a large
scale in the absence of formal institutions established by
nation states and informal institutions at work in small
social groups (Nee 2005). In fact, formal and informal
institutions such as legal systems and social norms,
respectively, have not disappeared, but they have lost
their influence in the governing of online market
exchanges (Przepiorka and Aksoy 2017).
Our findings also illustrate more generally the poten-
tial of data generated by the hidden corners of the
World Wide Web for studying fundamental social proc-
esses. Since the shutdown of Silk Road 1.0, many new
cryptomarkets have emerged, which have developed a
wide range of institutional arrangements that promote
trust between buyers and sellers, but also online traders’
trust in cryptomarket platforms. Traders’ trust in market
platforms has become essential for establishing coopera-
tion on the Dark Web, since absence of law enforcement
and full anonymity also brought uncertainty with regard
to trustworthiness of cryptomarkets as institutions.
After several cases of large-scale fraud committed by
owners of major cryptomarkets, more recent marketpla-
ces signal their trustworthiness to potential traders by
investing in technical innovations. Such innovations
include, for example, multi-signature escrow systems
that prevent marketplaces from abusing funds in the
escrow system (Bartlett 2014: Ch. 5; Barratt and
Aldridge, 2016), more complex website code bases that
require cryptomarket owners to have higher levels of
technical knowledge (Branwen, 2016) or finances to
start and maintain the marketplace, but also in social
innovations. The latter is exemplified by some crypto-
markets’ introduction of hierarchical systems among
buyers and sellers, where only well-reputed buyers can
gain access to goods of well-reputed sellers. In other
words, only after having engaged in some illicit transac-
tions are buyers trusted with transactions for higher
stakes (Gambetta, 2009). Such institutional arrange-
ments have an effect on how traders in a particular cryp-
tomarket perceive the shadow of the future and,
accordingly, on their strategies when interacting with
each other (Axelrod, 1984;Guala, 2012).
On the one hand, these developments show that cryp-
tomarket traders can never fully rely on platform pro-
viders to take the role of rule enforcers, making trust
between buyers and sellers an essential part of maintain-
ing cooperation in the Dark Web. On the other hand,
such developments open avenues to analyse how institu-
tional innovations affect cooperation between individuals
in an otherwise largely de-centralized and unregulated
environment. The ongoing growth and diversification of
cryptomarkets (Soska and Christin, 2015), thus makes
760 European Sociological Review, 2017, Vol. 33, No. 6
Downloaded from https://academic.oup.com/esr/article-abstract/33/6/752/4555481 by guest on 19 November 2018

these Dark Web marketplaces particularly interesting for
studying issues of cooperation and institutional evolution
(Beckert and Wehinger, 2013).
Notes
1 In his seminal book, ‘Order without Law: How
Neighbors Settle Disputes’, Ellickson (1991)
describes how cooperation and social order are
maintained in a close-knit community of cattle
farmers in Shasta County, California. In his
account, Ellickson repeatedly emphasizes the
importance of gossip and reputation for promoting
cooperative behaviour (e.g. p. 232): ‘The residents
of rural Shasta County gossip all the time. Indeed,
any close-knit group is likely to have procedural
norms that ask members to help spread truthful
information about the prior prosocial or antisocial
behavior of other members. By facilitating the
flow of reputational information, these norms
deter future uncooperative behavior by increasing
an actor’s estimates of the probability that infor-
mal enforcers would eventually catch up with
him’. Although we study a ‘community’ of anony-
mous traders that engage in illegal market
exchanges without ever meeting each other in per-
son, the mechanisms by which cooperation is
maintained are the same albeit digitalized in form
of an electronic reputation system.
2 Note that this is different from the ‘reputation’
model put forward by Kreps et al. (1982).Kreps
et al. (1982) start from the observation that in
finitely repeated prisoners’ dilemma (PD) games
players cooperate more than the Nash equilibrium
would predict. They explain this observation by
suggesting that if players assume that with a small
probability their interaction partner prefers to
cooperate as long as they do, it may be rational
for them to cooperate as well. In other words, a
self-regarding player, who would defect in the
one-shot PD, mimics a cooperative type in order
not to forgo the higher benefits that result from
several rounds of mutual cooperation as compared
to mutual defection. Only once the sequence of
interactions approaches the end, it becomes benefi-
cial to defect and thereby reveal one’s true type.
Kreps and Wilson (1982) call this initial phase of
mimicry ‘reputation effect’. In contrast, we argue
that reputation can be conceived as a costly signal
because it is costly to acquire, which deters
untrustworthy sellers to enter the market
(Przepiorka and Berger, 2017).
3 Legal systems do not completely prevent scamming
behaviour of online sellers, especially for transac-
tions of small value, where potential costs of litiga-
tion become relatively high compared to potential
losses for buyers. It has been estimated, based on
buyer survey data, that 1–2 per cent of transac-
tions in eBay contain fraudulent seller behaviour
(Bauerly, 2009). While embeddedness of licit mar-
kets in legal systems cannot fully eradicate oppor-
tunistic behaviour, it provides strong assurances for
buyers, which might account for a large part of
trust facilitation that has been attributed to reputa-
tion systems in previous research.
4 Another way to address this question is to con-
duct laboratory experiments, in which a basic
market environment can be staged, and the main
tenets of the reputation mechanism can be put to
an empirical test under controlled conditions
(Falk and Heckman, 2009;Greiner and
Ockenfels, 2009). The effectiveness of the reputa-
tion mechanism to promote trust and cooperation
in social and economic exchange has been
corroborated in a number of experiments (Bolton
et al., 2004;Yamagishi et al., 2009;Kuwabara,
2015;Abraham et al., 2016; although see Corten
et al., 2016). However, this approach still resem-
bles the set-up of online markets for licit goods
in many relevant respects. First, participants are
recruited from a selective sample of mostly uni-
versity students motivated to participate in exper-
imental research. Secondly, although decisions in
the laboratory are financially incentivized, the
stakes are relatively low and potential losses lim-
ited to the opportunity costs of participation.
Third, at the end of the day, participants are
socially embedded in the environment of their
universities which, in most cases, are embedded
in well-functioning legal systems. In other words,
actors are less likely to be blackmailed or even
killed as a consequence of participating in experi-
mental as compared to illegal markets.
Supplementary Data
Supplementary data are available at ESR online.
European Sociological Review, 2017, Vol. 33, No. 6 761
Downloaded from https://academic.oup.com/esr/article-abstract/33/6/752/4555481 by guest on 19 November 2018

Acknowledgement
The authors are grateful to Nicolas Christin for providing them
with the data and very helpful comments. They would also like
to thank three anonymous reviewers for their perceptive
comments and suggestions.
Funding
LN gratefully acknowledges support from the Netherlands
Organisation for Scientific Research (NWO) [grant number:
406-12-004].
References
Abraham, M. et al. (2016). Reputation formation in economic
transactions. Journal of Economic Behavior & Organization,
121, 1–14.
Aldridge, J. and De´ cary-He´tu, D. (2016). Hidden wholesale: the
drug diffusing capacity of online drug cryptomarkets.
International Journal of Drug Policy,35, 7–15.
Allison, P. D. (2009). Fixed Effects Regression Models.
Thousand Oaks, CA: SAGE.
Axelrod, R. (1984). The Evolution of Cooperation. New York,
NY: Basic Books.
Bakos, Y. and Dellarocas, C. (2011). Cooperation without
enforcement? A comparative analysis of litigation and online
reputation as quality assurance mechanisms. Management
Science,57, 1944–1962.
Barratt, M. J. and Aldridge, J. (2016). Everything you always
wanted to know about drug cryptomarkets* (*but were afraid
to ask). International Journal of Drug Policy,35, 1–6.
Bartlett, J. (2014). The Dark Net: Inside the Digital
Underworld. London: Melville House.
Bauerly, R. J. (2009). Online auction fraud and eBay. Marketing
Management Journal,19, 133–143.
Beckert, J. (2009). The social order of markets. Theory and
Society,38, 245–269.
Beckert, J. and Wehinger, F. (2013). In the shadow: illegal mar-
kets and economic sociology. Socio-Economic Review,11,
5–30.
Bolton, G. E., Greiner, B. and Ockenfels, A. (2013). Engineering
trust: reciprocity in the production of reputation information.
Management Science,59, 265–285.
Bolton, G. E., Katok, E. and Ockenfels, A. (2004). How effective
are electronic reputation mechanisms? An experimental inves-
tigation. Management Science,50, 1587–1602.
Bowles, S. and Gintis, H. (2011). A Cooperative Species:
Human Reciprocity and its Evolution. Princeton: Princeton
University Press.
Bozoyan, C. and Vogt, S. (2016). The impact of third-party
information on trust: valence, source, and reliability. PLoS
One,11, e0149542.
Branwen, G. (2016). Black-market risks: analysis of lifespans,
deaths, and predictive factors, gwern.net Blogs. Available from:
<http://www .gwern.net/Black-m arket%20surviv al >[accessed
19 December 2016].
Calkins, M. M., Nikitkov, A. and Richardson, V. (2008).
Mineshafts on Treasure Island: a relief map of the eBay fraud
landscape. Pittsburgh Journal of Technology Law and Policy,
8, 1–52.
Carruthers, B. G. (2013). From uncertainty toward risk: the case
of credit ratings. Socio-Economic Review,11, 525–551.
Christin, N. (2012). Traveling the Silk Road: A measurement
analysis of a large anonymous online marketplace. arXiv,
1207.7139.
Coleman, J. S. (1990). Foundations of Social Theory. Cambridge,
MA: The Belknap Press of Harvard University Press.
Corten, R. et al. (2016). Reputation effects in social networks do
not promote cooperation: an experimental test of the Raub &
Weesie model. PLoS One,11, e0155703.
Dasgupta, P. (1988). Trust as a commodity. In Gambetta, D.
(Ed.), Trust: Making and Breaking Cooperative Relations.
Oxford: Basil Blackwell, pp. 49–72.
De´ cary-He´ tu, D., Paquet-Clouston, M. and Aldridge, J. (2016).
Going international? Risk taking by cryptomarket drug ven-
dors. International Journal of Drug Policy,35, 69–76.
Dellarocas, C. (2003). The digitization of word of mouth: prom-
ise and challenges of online feedback mechanisms.
Management Science,49, 1407–1424.
Diekmann, A. et al. (2014). Reputation formation and the evolu-
tion of cooperation in anonymous online markets. American
Sociological Review,79, 65–85.
Diekmann, A. and Przepiorka, W. (forthcoming). Trust and rep-
utation in markets. In Giardini, F. and Wittek, R. (Eds.), The
Oxford Handbook of Gossip and Reputation. Oxford:
Oxford University Press.
Dunbar, R. I. M. (2004). Gossip in evolutionary perspective.
Review of General Psychology,8, 100–110.
Ellickson, R. C. (1991). Order without Law: How Neighbors
Settle Disputes. Cambridge: Harvard University Press.
Falk, A. and Heckman, J. J. (2009). Lab experiments are a major
source of knowledge in the social sciences. Science,326, 535–538.
Feinberg, M., Willer, R. and Schultz, M. (2014). Gossip and
ostracism promote cooperation in groups. Psychological
Science,25, 656–664.
Fligstein, N. (2001). The Architecture of Markets: An Economic
Sociology of Twenty-First-Century Capitalist Societies.
Princeton, NJ: Princeton University Press.
Frey, V. and van de Rijt, A. (2016). Arbitrary inequality in repu-
tation systems. Scientific Reports,6, 38304.
Friedman, E. J. and Resnick, P. (2001). The social cost of cheap
pseudonyms. Journal of Economics & Management Strategy,
10, 173–199.
Gambetta, D. (2009). Codes of the Underworld: How Criminals
Communicate. Princeton: Princeton University Press.
Global Drug Survey. (2016). Key findings from the Global Drug
Survey 2016. London: Global Drug Survey. Available from:
<https://www.globaldrugsurvey.com/past-findings/the-global-
drug-survey-2016-findings/>[accessed 12 December 2016].
762 European Sociological Review, 2017, Vol. 33, No. 6
Downloaded from https://academic.oup.com/esr/article-abstract/33/6/752/4555481 by guest on 19 November 2018

Granovetter, M. (1992). Problems of explanation in economic
sociology. In Nohria, N. and Eccles, R. G. (Eds.), Networks
and Organizations: Structure, Form, and Action. Boston:
Harvard Business School Press, pp. 25–56.
Greif, A. (1989). Reputation and coalitions in medieval trade:
evidence on the Maghribi traders. Journal of Economic
History,49, 857–882.
Greiner, B. and Ockenfels, A. (2009). Vom Labor ins Feld: Die
O
¨konomik des Vertrauens. Ko¨ lner Zeitschrift Fu¨ r Soziologie
Und Sozialpsychologie, Sonderheft,49, 219–242.
Grimalda, G., Pondorfer, A. and Tracer, D. P. (2016). Social
image concerns promote cooperation more than altruistic
punishment. Nature Communications,7, 12288.
Guala, F. (2012). Reciprocity: weak or strong? What punish-
ment experiments do (and do not) demonstrate. Behavioral
and Brain Sciences,35, 1–15.
Gu¨ th, W. and Ockenfels, A. (2003). The coevolution of trust
and institutions in anonymous and non-anonymous commun-
ities. In Holler, M.J., Kliemt, H., Schmidtchen, D. and Streit,
M. (Eds.), Jahrbuch fu¨ r Neue Politische O
¨konomie.
Tu¨ bingen: Mohr Siebeck, pp. 157–174.
Hardy, R. A. and Norgaard, J. R. (2016). Reputation in the inter-
net black market: an empirical and theoretical analysis of the
deep web. Journal of Institutional Economics,12, 515–539.
Hillmann, H. (2013). Economic institutions and the state:
insights from economic history. Annual Review of Sociology,
39, 251–273.
Hillmann, H. and Aven, B. L. (2011). Fragmented networks and
entrepreneurship in late imperial Russia. American Journal of
Sociology,117, 484–538.
Jappelli, T. and Pagano, M. (2002). Information sharing, lend-
ing and defaults: cross-country evidence. Journal of Banking
and Finance,26, 2017–2045.
Kollock, P. (1994). The emergence of exchange structures: an
experimental study of uncertainty, commitment, and trust.
American Journal of Sociology,100, 313–345.
Kollock, P. (1999). The production of trust in online markets. In
Lawler, E. J. and Macy, M. W. (Eds.), Advances in Group
Processes, Vol. 16. Amsterdam: JAI Press, pp. 99–123.
Kreps, D. M. et al. (1982). Rational cooperation in the finitely
repeated prisoners’ dilemma. Journal of Economic Theory,
27, 245–252.
Kreps, D. M. and Wilson, R. (1982). Reputation and imperfect
information. Journal of Economic Theory,27, 253–279.
Kuwabara, K. (2015). Do reputation systems undermine trust?
Divergent effects of enforcement type on generalized trust and
trustworthiness. American Journal of Sociology,120,
1390–1428.
Lindenberg, S. (2017). The dependence of human cognitive and
motivational processes on institutional systems. In Jann, B.
and Przepiorka, W. (Eds.), Social Dilemmas, Institutions and
the Evolution of Cooperation. Berlin: De Gruyter
Oldenbourg, pp. 85–106.
Martin, J. (2014). Drugs on the Dark Net: How Cryptomarkets
are Transforming the Global Trade in Illicit Drugs.
Basingstoke: Palgrave Macmillan.
Matzat, U. and Snijders, C. (2012). Rebuilding trust in online
shops on consumer review sites: sellers’ responses to
user-generated complaints. Journal of Computer-Mediated
Communication,18, 62–79.
Milgrom, P. R., North, D. C. and Weingast, B. R. (1990). The
role of institutions in the revival of trade: the law merchant,
private judges, and the champagne fairs. Economics and
Politics,2, 1–23.
Milinski, M. (2016). Reputation, a universal currency for
human social interactions. Philosophical Transactions of the
Royal Society B,371, 20150100.
Milinski, M., Semmann, D. and Krambeck, H.-J. (2002).
Reputation helps solve the 0tragedy of the commons0.Nature,
415, 424–426.
Nee, V. (2005). The new institutionalisms in economics and
sociology. In Smelser, N. J. and Swedberg, R. (Eds.), The
Handbook of Economic Sociology. Princeton, NJ: Princeton
University Press, pp. 49–74.
Nowak, M. A. (2006). Five rules for the evolution of coopera-
tion. Science,314, 1560–1563.
Pavlou, P. A. and Gefen, D. (2004). Building effective online
marketplaces with institution-based trust. Information
Systems Research,15, 37–59.
Przepiorka, W. (2013). Buyers pay for and sellers invest in a
good reputation: more evidence from eBay. Journal of
Socio-Economics,42, 31–42.
Przepiorka, W. and Aksoy, O. (2017). Social Order in Online
Markets and the ‘Collapse’ of Institutions. Unpublished
manuscript, The Netherlands: Department of Sociology/ICS,
Utrecht University.
Przepiorka, W. and Berger, J. (2017). Signalling theory evolving:
signals and signs of trustworthiness in social exchange. In
Jann, B. and Przepiorka, W. (Eds.), Social Dilemmas,
Institutions and the Evolution of Cooperation. Berlin: De
Gruyter Oldenbourg, pp. 373–392.
Resnick, P. and Zeckhauser, R. (2002). Trust among strangers
in internet transactions: empirical analysis of eBay’s reputa-
tion system. In Baye, M. R. (Ed.), The Economics of the
Internet and E-Commerce. Amsterdam: Elsevier, pp.
127–157.
Shapiro, C. (1983). Premiums for high quality products as
return to reputation. Quarterly Journal of Economics,98,
659–680.
Snijders, C. and Weesie, J. (2009). Online programming mar-
kets. In Cook, K. S., Snijders, C., Buskens, V. and Cheshire,
C. (Eds.), eTrust: Forming Relationships in the Online
Wor ld . New York, NY: Russell Sage Foundation, pp.
166–185.
Sommerfeld, R. D. et al. (2007). Gossip as an alternative for
direct observation in games of indirect reciprocity.
Proceedings of the National Academy of Sciences of the USA,
104, 17435–17440.
Soska, K. and Christin, N. (2015). Measuring the longitudinal
evolution of the online anonymous marketplace ecosystem. In
24th USENIX Security Symposium (USENIX Security 15).
Washington, DC: USENIX Association, pp. 33–48.
European Sociological Review, 2017, Vol. 33, No. 6 763
Downloaded from https://academic.oup.com/esr/article-abstract/33/6/752/4555481 by guest on 19 November 2018

Standifird, S. S. (2001). Reputation and e-commerce: eBay auc-
tions and the asymmetrical impact of positive and negative
ratings. Journal of Management,27, 279–295.
Sylwester, K. and Roberts, G. (2013). Reputation-based partner
choice is an effective alternative to indirect reciprocity in solving
social dilemmas. Evolution and Human Behavior,34, 201–206.
Willer, R. (2009). Groups reward individual sacrifice: the status
solution to the collective action problem. American
Sociological Review,74, 23–43.
Wu, J., Balliet, D. and van Lange, P. A. M. (2016). Gossip versus
punishment: the efficiency of reputation to promote and main-
tain cooperation. Scientific Reports,6, 23919.
Yamagishi, T. et al. (2009). Solving the lemons problem with repu-
tation. In Cook, K. S., Snijders, C., Buskens, V. and Cheshire, C.
(Eds.), eTrust: Forming Relationships in the Online World.New
York, NY: Russell Sage Foundation, pp. 73–108.
Wojtek Przepiorka is Assistant Professor at the
Department of Sociology in Utrecht. Before moving to
the Netherlands, he held a research fellow position at
Nuffield College and the Department of Sociology in
Oxford. His research interests are in analytical and eco-
nomic sociology, game theory, and quantitative method-
ology, in particular experimental methods. His most
recent publications include ‘“Take One for the Team!”
Individual heterogeneity and the emergence of latent
norms in a volunteer’s dilemma’ (Social Forces 94, with
A. Diekmann), and ‘Generosity is a sign of
trustworthiness—the punishment of selfishness is not’
(Evolution and Human Behavior, 37, with U. Liebe).
Lukas Norbutas is a PhD candidate at the Department
of Sociology of Utrecht University, and the Netherlands
Institute for the Study of Crime and Law Enforcement.
His research interests include economic sociology, social
network analysis, cooperation and trust, and online
criminal networks.
Rense Corten is Associate Professor at the Department
of Sociology of Utrecht University. His research revolves
around the themes of cooperation, trust, and (the
dynamics of) social networks, with empirical applica-
tions including adolescent networks, social media, the
sharing economy, online criminal networks, and labora-
tory experiments. His work has been published in jour-
nals such as the American Sociological Review,Social
Networks,Rationality and Society, and others.
764 European Sociological Review, 2017, Vol. 33, No. 6
Downloaded from https://academic.oup.com/esr/article-abstract/33/6/752/4555481 by guest on 19 November 2018