Offline ‘stranger’ and online lurker: methods for an ethnography of illicit transactions on the darknet

Introduction

Drug transactions have existed in the digital sphere since the infancy of Arpanet (Markoff, 2005: 109). For a long time these deals were privately arranged, such as the alleged marijuana deal between Stanford and MIT students in the early 1970s, as described in Markoff (2005: 109). In the last decade or so, public marketplaces popped up on the clearnet (the regular internet), but were short-lived. In the last five years we have seen the emergence of retail- and small-wholesale online drug marketplaces located on the darknet and, more recently on smartphone apps. The first major darknet marketplace (DNM), Silk Road, opened for business in February 2011, and captured the public’s attention when Adrian Chen wrote of the site for Gawker a few months later (Chen, 2016).

Silk Road (and its ilk) is best described as a ‘risk management platform for participants in (mostly illegal) transactions’ (Soska and Christin, 2015). DNMs are no more than middleman services, decentralized operations connecting sellers and buyers in a communal space, ‘providing the means for transactions and review systems.’ (Munksgaard et al., 2016: 2). The infrastructure is similar to that of eBay or Amazon Marketplace, with mechanisms in place to handle disputes, a feedback system, and forums where the community discusses buyers, vendors, drugs and drug use, and everyday life. ¹

The darknet (deep web; hidden internet) is accessible using the Tor browser. Tor (The Onion Router) essentially anonymizes user IP addresses, which for DNM users provides a layer of protection against law enforcement detection (Barratt and Maddox, 2016: 3; Tor Project. 2011). For all of the attention paid to DNMs by law enforcement, the media, and academics, these marketplaces are not the only digital option for those seeking to either sell or buy illegal drugs. We know people are also able to buy drugs from offline dealers advertising on websites such as Craigslist (Tofighi et al., 2016), and a number of smartphone apps are also known to facilitate drug transactions (Borromeo, 2016; EMCDDA, 2016; McGurk, 2016; Ferguson, 2016). As increasing numbers of researchers and students become interested in ethnographic research on digital drug transactions, we need to more formally address how we do this. Can we formalize specific ethnographic methods for this space? How would those look? And how do we handle ethics and data concerns, particularly given the anonymity of participants in these transactions and therefore complications verifying what we observe and are told. And how can researchers protect themselves from legal risks and questions of personal safety?

Darknet markets research

There has been extensive ethnographic work on the world of offline extralegal enterprise. Even if we restrict ourselves to studies of the drug trade, the literature is vast (Levitt and Venkatesh, 2000; Mohamed and Fritzvold, 2011; Williams, 1990, 1992; Zaitch, 2002, 2005). Digital ethnography (also referred to as virtual- and online-ethnography) has been examined in more general contexts not specific to DNM transactions (Barratt and Maddox, 2016; Hine, 2008; Lupton, 2015; Pink et al., 2015). There is a growing literature on observational methods in online chatrooms and message boards, communities located in social rather than physical spaces, where ‘community members cannot be visually or aurally monitored through face-to-face co-presence but through their textual representations of self’ (Fine and Hancock, 2016: 4). But, as Fine and Hancock caution, we ‘must carefully consider this new frontier of field labor: How do we identify our subjects? What counts as data when we are following links, post, and threads? Indeed, sitting at our computer, are we even “in the field?”’ (2016: 4).

The foci of research on darknet markets are wide-ranging: marketplace size (Christin, 2012); risk and international transactions (Décary-Hétu et al., 2016); vendor branding behavior (Broséus et al., 2016); participant motives (Van Hout and Bingham, 2013a); trust and reputation (Tzanetakis et al., 2016); violence (Tzanetakis et al., 2016; Décary-Hétu et al., 2016); harm reduction (Barratt 2012; Van Hout and Bingham, 2013a); and the different drug types (Munksgaard et al., 2016). Research into location-based smartphone app drug transactions facilitated by apps such as Grindr, Instagram, and Whisper, is limited due to this market’s newness, particularly to the research community. Indeed, EMCDDA (2016) makes only brief mention of app-based drug transactions, mostly citing mainstream journalism.

Some large surveys, such as the Global Drug Survey (GDS) ask about DNM participation, but should be handled with caution due to issues of self-selection and honesty in the survey population. ² As Barratt et al. (2014, 2016) point out, there is no way to test how representative the sample is of the larger population of DNM participants, and the GDS is ‘more likely to contain end-buyers and some retailers, rather than wholesalers’ (Barratt et al., 2016: 7).

A number of DNM studies have used data obtained from scrapes, or crawls, of marketplaces, otherwise referred to as ‘webometric’ data (e.g. Aldridge and Décary-Hétu, 2014; Christin, 2012; Dolliver 2015a,b; Soska and Christin, 2015). ³ There are challenges faced by researchers attempting scrapes of Tor sites: Connections are less stable and slower than clearnet connections, DNM operators might block scrape software or interrupt connections, and the scrape may only grab a non-representative snapshot of a day or two’s data from a DNM while appearing to have fully crawled the market. As Soska and Christin (2015) highlight, crawls only take into account listings existing at the time of the scrape; over the course of a year thousands of listings may be unaccounted for. Data obtained in this way can even miss entire vendor accounts.

The study of DNMs is particularly amenable to mixed methods. Tzanetakis et al. (2016) find that DNMs, as with offline drug markets, ‘seem to be shaped by social practices of trust, violence, and other dispute resolution techniques, and logistics.’ (Tzanetakis et al., 2016: 10). No matter if the market is on- or offline, ‘Law enforcement seems to be a trigger causing a special need for trust, an unpredictable and partly unknown occurrence of violence and theft, and certain forms of logistics employed’ (Tzanetakis et al., 2016: 10). This is consistent with Décary-Hétu et al. (2016) whose four point breakdown of ‘risk’ on DNMs – ‘arrests, violence, profits, and reputation’ – and their finding that the ‘perceived effectiveness of [local] law enforcement’ is a prime focus in vendors’ risk perception, conforms to what we know of offline drug marketplace risks.

Qualitative research investigating the population of users on darknet marketplaces, particularly ethnographies and open-ended interviews, can provide understanding of participants’ self-perceptions (Barratt, 2012), and their reasons for transacting for drugs on DNMs (Van Hout and Bingham, 2014). Foremost in reported reasons are promises of dose reliability and safety, lower risk of arrest, convenience, ease of purchase, product quality and the promise of anonymity (Barratt, 2012; Barratt et al., 2014, 2016; Orsolini et al., 2015; Van Hout and Bingham, 2013a, 2013b, 2014).

Because of the newness of DNM ethnography, and because of often inaccurate beliefs that these might be ‘easier’ and ‘safer’ fieldwork sites to access, we need to attend to the range of possible empirical foci and to methodological challenges and benefits of DNM ethnographic methods. Some DNM researchers have already made inroads outlining a methodological program, most notably Barratt and Maddox whose methodological ‘workflow’ confronts ‘consequences of particular choices of action upon knowledge production.’ (2016: 1). They argue powerfully for the contribution of digital ethnography to DNM research, referring to it as ‘…a contextual counter-balance to the increasing use of unobtrusive analyses of digital traces in the study of stigmatized populations online.’ (Barratt and Maddox, 2016: 3).

While the online location of darknet markets to some extent resolves the first challenge of accessing drug buying and selling population – the question of locating and ‘physically’ entering the scene – gaining access nonetheless requires preparation before attempting to directly engage with participants (Barratt and Maddox, 2016; Knox, 2001). Van Hout and Bingham (2013a, 2014) took 12 months preparing to begin data collection. Despite this, they report ‘recruitment of site users was hampered by negative and suspicious reactions’ (Van Hout and Bingham, 2013a: 386). ⁴ Earning trust over time was similarly a feature for Taylor and Kearney (2005).

My experience differed from Van Hout and Bingham’s – I did not have the benefit of site operator support as they did, but by not publically declaring my interest I also did not face the same degree of public kickback. I did, however, face distrust from those to whom I explained my interest in private chats and emails; some warned me to stay away. But, unlike other researchers, I was never threatened or harassed: Barratt and Maddox (2016: 7) describe negative responses to their presence ‘characterized by accusations of the researchers being “community outsiders”’.

Indeed, the trust challenge is not unique the study of darknet markets, but is universal to the study of populations where there exist ‘legitimate concerns about disclosing information to researchers that, if known more widely, could have serious negative repercussions for their lives’ (Barratt and Maddox, 2016: 2). My offline experience with drug selling and using populations was indispensable in my DNM fieldwork. Coupled with the time I spent observing the DNMs this positioned me as a ‘Privileged Access Interviewer’ (PAI), one who has ‘access to the population under investigation, usually from within’ (Barratt and Maddox, 2016; Taylor and Kearney, 2005; Van Hout and Bingham 2013a, 2013b, 2014).

In the case of smartphone apps, these are used as geolocated advertising by drug dealers and buyers seeking to perform an offline transaction as well as by those seeking to perform a digital transaction. In the latter, the connection is made on one app (e.g. Instagram), contact information is exchanged to connect parties on another app (e.g. Whatsapp), and payment is made in person or via a payment app such as Venmo (Ferguson, 2016). Smartphone app drug deals emerged on the scene after the DNMs, and while the geolocation aspect of app-based online-offline transactions reduces the privacy of participants – one of the strongest features of DNMs – the ubiquity of smartphones and the ease with which apps are adopted makes these more appealing to a great many consumers and dealers. Indeed, a dealer no longer has to stand around on the street waiting for business, or rely on a static customer base for deliveries. Dealers can now combine those two business models – reducing the risk in the streetside salesman model while increasing their visibility versus the closed market delivery dealer model, all via a smartphone, on the go, without necessitating a Tor connection or the clunkiness and delays of escrow payments and shipping logistics.

We must also examine questions of research ethics and how these interplay with methodology in the digital realm. In the context of DNMs I frequently hear people assume digital ethnography is a ‘safer’ or ‘easier’ way to conduct this type of ethnographic research into drug markets. Indeed, in many ways it might be, but the ethical considerations regarding our handling of subject confidentiality remain, and we also find – often unexpectedly – risks to researchers themselves. In recent years we have learned more about the risks to offline fieldworkers (Bahn, 2012; Barratt and Maddox 2016; Perrone, 2010; Poulton, 2012). Similarly, some darknet market researchers have felt at greater risk than others in the course of their research.

Some researchers operate on darknet fieldwork sites under false identities (e.g. Gehl, 2016). On DNMs as well as on the clearnet, researchers have operated dual identities – one ‘real’, one pseudonymous (e.g. Paechter [2013] who observed a ‘UK-based divorce wiki and support site’ as both insider and outsider, having established herself as a pseudonymous insider before entering the site as a researcher). Other researchers choose to operate under their real identity (e.g. Barratt and Maddox, 2016). Citing Gehl (2016), Barratt and Maddox explain their choice to use their real names on DNMs was based upon a belief that this would ‘allow for independent verification of our identities by the community and to establish the shared trust between researchers and participants’ (2016: 6) and ‘allay participant concerns about our identity as researchers, given the reality of undercover agents operating in this space’ (Barratt and Maddox, 2016: 7). Citing Paechter (2013), Barratt and Maddox say that while their choice to operate one pseudonymous identity and one real world identity on DNMs might provide ‘greater access as an insider, ethical concerns arising from dual roles became harder to manage as material obtained as an ‘insider’ could not become ‘unknown’ under the researcher identity’ (2016: 7).

There exist legal risks faced by DNM researchers, which vary according to jurisdiction. As Barratt and Maddox (2016: 13) point out, in Australia – where they are located – it is more likely for researchers to face a subpoena in a law enforcement investigation of drug transactions than would be the case in the United States. ⁵ And DNM researchers, particularly women, face the issue of sexist comments and creepy harassment directed at them on these sites. This occurred with Barratt and Maddox in their fieldwork (Barratt and Maddox 2016: 12), but this was not the worst incident of harassment in the DNM research scene. In conversation with a DNM ethnographer overseas a few years ago, I was informed that others in the field had been receiving threats they believed to be from supposedly organized-crime-linked DNM vendors, insisting they stop their research. And Décary-Hétu and Aldridge (2015: 11-12) report that while they were gathering data the distrust and paranoia of a subset of market participants led to a threat against their lives. These are serious concerns underscoring the call to formalize our methodological and ethical approach to the DNM fieldwork site.

Doing darknet market ethnography

My DNM work occurred on a number of marketplaces, starting with the original Silk Road in 2012. My initial research examined trust: how does someone trust an anonymous stranger such that they ingest drugs sent to them in the mail? This expanded into an examination of the operation of identity, trust, and reputation, how vendors and buyers assess one another under conditions of anonymity and in the absence of access to direct sanctions. I found these marketplaces represent cases where trust is grounded in the reputation capital generated by the feedback systems and open discussion on the forums, which establish or deny perceptions of authenticity and legitimacy that vendors attempt to convey on their vendor profile pages (Ferguson, working paper).

Access and entry

I initially learned of Silk Road from an undergraduate student who knew of my work in offline drug markets. I downloaded the Tor browser, signed up for a Silk Road account and then spent several weeks on the site, for hours every day – from 1-3 to upwards of 12-14. I tried to get up to speed with what this was, not just how the transactions worked but to familiarize myself with how interactions worked in this space, how participants operated socially, how they categorized one another, and how everyday norms operated in this scene. I endeavored, essentially, to become ‘local’ to the community. I accomplished this by reading profiles and observing marketplace forums and comments on reddit and Twitter, while I observed the market rapidly growing as reports in magazines and newspapers increased the visibility of darknet markets. ⁶

I began chatting with people on the forums after a few months spent getting a sense of ‘the scene’. Conversations occurred over private messaging functions on marketplaces and via encrypted emails and, later, the Signal messenger app. When I first messaged people – or responded if they contacted me first – I revealed I was there for research and directed them to a website describing my past and ongoing projects. Later, when the forums-only site known as ‘The Hub’ was launched, I informed the population and the moderators that I was there for research purposes.

Previous offline fieldwork with drug gangs, independent drug dealers, and traffickers, and a personal connection with a number of drug sellers and recreational drug users, located me in a privileged position regarding access: I could talk the talk in various drug sub-communities from day one, there was no learning curve as far as terminology or the characteristics of different drug types, users, or sellers, and I had a reasonably sophisticated understanding of variations characterizing different types of seller and user.

This knowledge accelerated entry, enabling me to effectively communicate with individuals much sooner, and to more easily understand subcultural nuances. My previous fieldwork eased my entry and acceptance, and permitted my observations without requiring that I lose status by asking participants to explain basic terminologies and scene-specific forms of talk. Anyone seeking to perform ethnographic work on DNMs would be best prepared if they first built such an understanding and an interactional ease with drug populations. My understanding of drugs and of drug users and sellers allowed me to hang back, to talk as an insider, to adopt the role of the Simmelian Stranger (Simmel, 1950).

I did not spend as much time in the preliminary stages as Van Hout and Bingham (2013a, 2013b, 2014), but was on Silk Road for six months before I began talking directly with participants. Then, in August 2013, site operators began an overhaul of the feedback system, an area I had decided would be a focus for my examination of trust and identity. With a lot of posts being made regarding the feedback system overhaul, I streamlined my engagement to focus on this issue.

My research methods on smartphone app drug transactions took a slightly different approach, with less time between my first observations and making contact with participants. I sent out messages to people with drugs advertised for sale on Whisper and Instagram. I stated I was a sociologist at Princeton University with prior experience in offline drug markets as well as on DNMs, asking if they would be willing to chat. I provided my real name. I received a handful of responses, most intended to put me off, some of which offered to talk only in return for sexual favors. But a few people responded positively and I followed up, trading messages about their motivations and experiences, and their methods for selling and buying drugs on these apps. I conducted phone calls with two of these individuals.

Lurking

Following the example of other online ethnographers, I initially adopted the role of ‘lurker.’ In so doing, I could be certain my presence in no way altered the behavior of those I observed. For some this may raise ethical questions. As others have asked of clearnet (e.g. Brotsky and Giles, 2007: 95-96) and DNM ethnographic ethics: is it deceptive to observe without informing the observed, should they be given the option to alter their behavior accordingly, how might we obtain their consent? Since the onus is on researchers in this situation to justify breaking standard IRB protocols in this way, I suggest the following three justifications for my approach:

This is one way that darknet marketplace observation is easier and safer than observations would be in offline drug markets: it is possible to observe behavior on DNMs without expressly announcing one’s presence. This would be all but impossible in offline drug marketplaces where the presence of a frequent observer, particularly one who lurks, would be quickly noticed and would almost surely cause the subjects to abandon their location or activity, potentially putting the subjects’ as well as researcher’s safety at risk.

Ethical considerations

There are a number of ethical issues involved in ethnographic work with subjects who break the law, online or offline, and specific protections we need to consider as a result. The key issues center on how we protect subject confidentiality, broadly:

The permanence of digital footprints – theirs and ours – intensifies the challenge of identity protection, and we must remain aware that in the future these will be even easier to trace and de-anonymize. Consistent with Barratt and Maddox (2016), I did not ask about or observe specific transactions, instead I focused on self-presentation, alias identity, reputation, trust and authenticity.

I handled identity and confidentiality issues in two ways. The subjects themselves adopt DNM user names masking their true identities; for the purposes of transparency – and replicability – these IDs remain the same in any analysis. To change the identities of the authors of public posts I quoted would be redundant from a ‘protect-your-subjects’ standpoint. However, any private messages and non-public statements regarding activity in DNMs or in offline illegal transactions need to be treated differently. My private messages were not publicly available and subjects’ identities from these were changed in my notes and not associated with the DNM identities of the individuals. This presents analytical challenges and I accept it this is a flaw in my method, one which I hope can be resolved in attempts to formalize DNM ethnographic methods.

I encrypted messages and phone transcripts, which I stored offline. Early on I was advised by a colleague that cloud storage services would not be a wise choice, an issue addressed by other darknet market researchers (e.g. Barratt and Maddox, 2016: 13). Private messages were initially stored in their original encrypted form, but after the first Silk Road bust I realized the corruption of encryption keys was a distinct possibility. In early October 2012, I re-encrypted all messages I had stored offline, using a separate key only ever used for this purpose. I exclusively used external hard drives and a dedicated desktop for all of my DNM data, a strategy that presented no challenges because this was not a collaborative project. I was less concerned with the encryption of my publicly available data – public forum posts and vendor profiles – given that any public materials I had pulled from DNMs was surely also in the possession of other researchers, journalists, and law enforcement agencies worldwide.

We must be extra cautious in the case of findings and data that may at some point in the future be used to trace subjects, and investigate or prosecute them. Some information provided in private conversations, while relevant to my foci, struck me as potentially identifying and I decided I could not use these materials. This was disappointing. But an individual who exposes too much personal information – geographic and demographic as well as health, family structure, and employment details, might be leaving enough out there to be unmasked. If they were to be later arrested for their DNM transactions, any research data on their participation could further the prosecution and might even add to a fine or prison sentence. It is also possible that what subjects told me privately about their businesses could have been exaggerated – a common issue in ethnographies of drug populations, and all but impossible to detect on DNMs. That exaggeration might land a subject with a longer prison sentence or greater fine than would be warranted for their real infractions and so certain data were omitted from my analysis.

To control the risks to subjects we should at the very least follow standard IRB protocols, maintain strict confidentiality and data protections, and I suggest we steer away from conversations with subjects that may lead to their self-incrimination. In my work, I did not want to know my subjects’ real identities or locations, it was irrelevant to my purpose and would have increased their risk. Where anyone gave me details that might be used to piece together an identity, no matter how unlikely this might have been, I deleted that information from my transcripts before encrypting these data for offline storage.

Researcher risks

The risks to participants in DNMs are largely similar to the risks faced by subjects in offline drug market research – they might face exposure, arrest, loss of social or professional status. For researchers the risks are somewhat different on DNMs versus in offline ethnographic work in these communities. There is less risk of physical harm or entanglements with law enforcement, but there is an elevated risk of ‘digital harm’ and unseen adversaries.

Similar to Gehl (2016) my public persona on these marketplaces was pseudonymous, however in private messages I freely revealed my identity to my subjects. Barratt and Maddox suggest they might have benefitted from deploying ‘dual identities, such that one remained the identified researcher and the other could blend in more effectively through use of a pseudonym and without having to account for the research in online encounters’ (Barratt and Maddox, 2016: 5). Instead, they were open, using their real-world identities online, but I am unsure of the utility of Barratt and Maddox’s approach in this regard. They were open with their identities in the belief this would inspire participants to trust their motivations, but DNM participants see this as exactly the sort of elaborate identity-camouflage undercover law enforcement operations might deploy. This being the case, I am not convinced Barratt and Maddox’s self-exposure would be of much benefit to their study. Moreover, their openness with their real identities created personal safety issues, and identifying as female led to some unpleasant online harassment (2016). For these reasons, I suggest we formulate other ways to handle questions of researcher identity in protocols for DNM ethnographic methods.

Empirical foci

My darknet marketplace work focused on the performance of pseudonymous identity in high risk transactions and how this relates to reputation, authenticity, and trust. My conversations with participants on DNMs and app-based markets involved thoughts on trust; explanations as to why they projected the identity they did; questions about the purpose of my research; questions relating to my experiences in offline drug markets; and tales of personal tragedy, often framed as justifications for buying or selling drugs online. With participants in app transactions, I also discussed the logistics of these deals (information a DNM vendor would post in their profile); transaction successes and failures; and their views on different transaction methods (mail or in person; preferred payment methods – again, information a DNM vendor would post in their profile) and I asked their reasons for buying or selling on apps rather than DNMs. Lastly, on DNMs, vendors make a point of highlighting their tenure in the business, both online and offline, and so I also asked about this with my app dealer subjects. As for buyers, as with DNM buyers, app drug buyers cited convenience, reputation, and a sense of safety as their main motivating factors.

Forum data

In contrast to the app-based study, the bulk of my DNM data was not interview-based, but came from observational work. Darknet marketplace forums provide a wealth of data that can be scraped and used for quantitative or qualitative content analysis, but which are useful to ethnographers beyond the early stages of a study during which the forums can be a resource for familiarizing oneself with local norms, rules, forms of talk, and how these shape the community. ⁷ The forums are also a good source for understanding immediate issues as they arise within the community, as when I began focusing my data-gathering on feedback issues as a result of Silk Road’s feedback system overhaul. Also immediate issues facing the community can be used by researchers as springboards to entirely new studies. The forums are public and as such it is possible to ‘lurk’ in the background reading all the posts, maybe following a few central participants, over time building an ethnographic impression of the scene, community-defined population subcategories, arguments, norms, and the application of rules: a picture of the social world and economic life of the DNM.

Décary-Hétu et al. point to the utility of forum discussions in their analysis of international shipping and of the necessary risk assessments performed by vendors in deciding where to ship and how: ‘Forum discussion confirmed that many vendors believed that smaller consignments were safer for international shipping.’ (2016: 74) This information would be very difficult to obtain from a scrape of item listings or feedback posts – the personal decisions of the vendors in choosing to whom they would sell and what they would list are however fully examinable in public forum postings.

Feedback data

Examining the meanings we confer upon feedback and ratings systems is an important aspect of transaction behavior (Blank, 2007; Brown, 2010; Danescu-Niculescu-Mizil et al., 2009). We know the importance of word of mouth recommendations in, for example, choosing a restaurant (Chevalier and Mayzlin, 2004; Watts, 2003). Feedback and forum reviews can be seen as an online extension of ‘word of mouth’ recommendations, providing potential transaction partners with ‘…a way of engaging with the social relationships around purchasing, thereby providing an alternative to concentrating on the producer-consumer relationship’ (Blank, 2007; Brown, 2010; Brown and Laurier, 2012).

The feedback systems on darknet markets also provide a wealth of data that can be scraped and used for quantitative or qualitative content analysis, as well as providing ethnographers with data for analyses of DNM life and culture, trust, reputation, and risk. Feedback posts are also useful to ethnographers for testing and confirming observations from forum posts and interviews: If a vendor makes claims about the success of their business or of particular items it is reasonably easy to verify that information from feedback and ratings provided by buyers, something that would be far more difficult to accomplish in offline drug markets. Décary-Hétu et al. also used feedback posted by buyers in order to operationalize ‘reputation’ which was central to their analysis of perceived risk in vendors’ decisions as to whether or not they would offer drugs for sale to international buyers. They explain that this methodological decision stemmed from their assumption that vendors with the strongest reputations – those with the highest ratings, ‘… may conclude that the loss of vendor reputation is not worth the risk that international sales may entail’ (2016: 74), an assumption that they point out was substantiated by posts made in the forums.

Studies of DNMs that have examined trust and professionalism find that building a professional reputation – whether as a vendor or as a buyer (Bartlett, 2014) – is crucial to ongoing successful transactions (Van Hout and Bingham, 2013b), and that trust in the digital interpersonal relationship begins with trust in the feedback system and the ways in which participants discuss one another in the forums (Aldridge and Décary-Hétu, 2014: 16; Van Hout and Bingham, 2013b). Buyers and sellers alike understand the importance of feedback in the DNM space. In the forums, buyers expressed interest in seeing recent feedback at the top of the list on a vendor’s page, rather than comments being distributed according to some status-based algorithm that weighted comments, pointing out that for drug purchases in particular, ‘recent feedback is crucial’ in establishing trust in a vendor and in their products.

As I learned, a typical darknet market customer in search of a vendor will combine feedback and profile data in their assessments of the various available sellers, which they handle by filtering vendors in ways similar to how we might filter sellers on eBay or Amazon. Vendors carefully positioned themselves as craftsmen, scientists, horticulturalists, whose skills and experience divorced them from the general ‘drug dealer’ label; buyers told me that vendors’ profile claims humanized them. Again, this all centered on trust, which is adapted for the DNM environment and would be very difficult to fully examine in the absence of ethnographic input.

Conclusion

An immense amount of data is available in vendor profiles, in feedback posts, and in forum discussions. This makes it possible to qualitatively analyze the general motives of a large population of buyers and sellers on DNMs, their concerns and approaches to the risks taken by participating in these marketplaces, and broader notions of trust, risk, reputation, identity, and community. For ethnographers, these digital fieldwork sites offer a breadth of data impossible to access in offline fieldwork sites.

Contrasting ethnographic methods in the study of DNMs (and indeed in the study of any digital community) with quantitative methods – particularly those involving digital scrapes – the ethnographic method benefits from interaction with participants. Barratt and Maddox mention the ability to follow participants across various marketplaces and digital spaces, especially in the context of participants ‘who engage in illicit or stigmatized practices’ (2016: 14). To this I would add the context that can be obtained from engaging with participants, and the ways this context can produce more informed data on users’ experiences and perceptions, motivations and behavioral responses. For ethnographers, these sites offer access to far larger data than we could obtain in offline fieldwork. In return, ethnographic work can produce deep insight into these communities, in ways that could not be obtained without ethnographic engagement and observation.

A serious issue with crawl data arises from the inconsistency in the language and terminology used by darknet marketplace participants –in profiles and listings, in forum discussions and feedback posts. This inconsistency is greater than in legal online spaces because a great number of participants deploy linguistic OpSec: Masking one’s identity by reducing the extent to which one uses one’s typical offline patterns of speech and writing when in an illicit online environment. So not only is there a standard lack of consistency in grammar, spelling, and terminology, but participants are in fact making a concerted effort to not use certain terms, or not to use (or only to use) them in certain ways. Basic word usage and flexibility in the use of terminology might be quite different from what we would otherwise expect. In addition, individuals might completely throw grammatical correctness out of the window for OpSec reasons if in their ‘real’ life they are a stickler for grammatical rules. Aldridge and Décary-Hétu (2016) discuss the issue in the context of how vendors vary in deciding which category to list their products posted for sale, and how this is further complicated by inconsistency in which categories are even offered across different DNMs. And as they point out, ‘Although there are multiple ways researchers may wish to categorize drug types sold on cryptomarkets, we strongly recommend, therefore, not relying on the categories into which vendors themselves place their listings to accomplish this’ (Aldridge and Décary-Hétu, 2016: 3-4).

Lastly, and related to the point about language variation and purposeful linguistic OpSec, there are cultural dynamics at play in the online environment in which participants are global, do not have a standard terminology or even a standard cultural context by which to frame their self-presentation and participation in these marketplaces. Barratt and Maddox point out that ‘…the meaning of the data may be lost or misinterpreted when taken out of the social and cultural context within which it was produced (boyd and Crawford, 2012; Lupton, 2015); and therefore ‘there are limits to what researchers can expect from these new digital artefacts of social behavior, both in terms of interpretation and representativeness’ (Barratt and Maddox, 2016: 3). Indeed, ethnographers deal with variations in interaction and communication as a matter of course, and we should not only be looking at this as an aspect of darknet marketplace research in which we can excel but we should be using these skills in collaboration with non-ethnographer researchers in these spaces, particularly in mixed methods studies with quantitative researchers basing their analyses on DNM crawl data.

Directions future digital ethnographies of DNMs could take include pairing transactions between vendor and buyer and examining how the drugs are used – of particular interest would be the extent to which buyers use DNM purchases for re-sale either online or offline, and the decision making processes behind such endeavors. Aldridge and Décary-Hétu (2016: 7) suggest self-report surveys for this, but I would instead suggest ethnographic and interview methods. In this way we could prevent the selection bias issues faced with survey data – particularly in sites such as DNMs – and we would benefit from greater depth in the responses from participants. Further studies on trust, risk, and the effects of reputation would contribute to a growing body of work in this area. And there is need for research that performs a close examination of linguistic OpSec techniques – one that could only be performed where trusting relationships exist between researcher and subjects, and which would again be fruitful for testing mixed methods approaches to DNM research.

We need continued efforts to devise and formalize a well-defined program and set of legal, ethical, and security-related protocols for digital ethnography where the fieldwork site involves risk for the subjects, researchers, or both. Beyond any benefits to the field and to the production of knowledge, there are additional benefits that would accrue to researchers themselves, including the ability to communicate ethnographic research findings based on much larger populations than has previously been possible. And our students would also benefit, as fortunate recipients of a formal ‘how-to’ that would better equip them in darknet market fieldwork, which is indeed in some ways easier, and certainly less expensive, than offline ethnographic work with these populations. Lastly, formal protocols for darknet marketplace ethnographic methods would increase researcher and subject safety, would enable non-ethnographers to better understand what we do in these spaces, and could inspire more mixed methods studies for which darknet markets – and indeed app based drug markets – are perfectly suited.