I came back duly, and soon gathered an audience around me. A few compliments in pidgin-English, some tobacco changing hands, induced an atmosphere of mutual amiability. I tried then to proceed to business. First, to begin with subjects which might arouse no suspicion, I started to “do” technology. A few natives were engaged in manufacturing some object or other. It was easy to look at it and obtain the names of the tools, and even some technical expressions about the proceedings, but there the matter ended. (...) I took a village census, wrote down genealogies, drew up plans, and collected the terms of kinship. But all this remained dead material, which led no further into the understanding of real native mentality or behaviour, since I could neither procure a good native interpretation of any of these items, nor get what could be called the hang of tribal life.
In the opening paragraphs of
This version of the meaning problem is arguably peculiar to ethnography. Indeed, ethnographers have developed their own vocabulary for it, such as the distinction between
To illustrate the nature of this problem, I draw on the example of a web corpus (a collection of websites connected by their hyperlinks) that I curated in 2012 as part of a project that aimed to map the so-called New Nordic Food movement and trace its development across the Scandinavian countries. The New Nordic Food movement formally traces its origins to the publication of the Retrieved from
Eight years later, when I was building my web corpus, that situation had dramatically changed. Following up on the Manifesto, the Nordic Council of Ministers had launched its New Nordic Food programme to promote awareness of the Nordic region as a gastronomic destination and encourage cross-sectoral innovation and business development. In 2009, the Faculty of Life Sciences at the University of Copenhagen had received a research grant of ∈13.3 m for a project on “optimal well-being, development and health for Danish children through a New Nordic Diet” (OPUS). In addition, in 2010, Noma had dethroned El Bulli in Catalunya as the World’s Best Restaurant on the annual top fifty list published by the British Restaurant Magazine, a feat that it repeated three years in a row, bringing a veritable surge of young chefs and fine-dining lovers to Copenhagen in search of the new Nordic revolution. Similar restaurants had opened up in Sweden and Norway and were climbing quickly up the charts of the international gastocracy. Importantly, it was not only the chefs but also their growers and suppliers that attained star status. Lifestyle magazines were writing stories about a former investment banker from London who was handpicking sea urchins in Arctic Norway, an eccentric asparagus grower by a fjord on Zealand and a wildly bearded forager from Scania. Such stories, and many others, seemed to cement the impression that something had changed fundamentally in the way in which Scandinavia perceived itself (and was being perceived by the rest of the world) as a gastronomic hot spot.
From an ethnographic point of view, what made the New Nordic Food movement interesting and worthy of inquiry was the obvious differences in the way in which it had been translated into everyday practice across Scandinavia. Even from a position in Copenhagen, where it was common to be presented with the same rather heroic epic about Nordic chefs conquering the world, there were differences to be observed, for example between the intervention research taking place at the Faculty of Life Sciences and the fermentation experiments happening in the kitchen at Noma (see also Munk & Ellern, 2015). Nevertheless, high gastronomy and public health had somehow become related. It was clear that changes were happening to the social fabric and that these changes were likely to take on very different forms across Scandinavia. Figure 1 shows the beginning of one of my attempts to map these differences. I used the Navicrawler plug-in for Firefox to collect websites that are in one way or another dedicated to foraging, wild foods and gastronomy in the Nordic region. The figure shows the resulting network of connections between Danish mushroom and wild herb enthusiasts.
The Navicrawler browses through the pages of each website and indexes the hyper-links. From this neighbourhood of linked websites, the researcher can make choices about where to visit next. The plug-in is specifically designed to keep the crawling process on a tight manual leash (Figure 4 gives a sense of the interface). Every time a site is crawled, the user must decide which of the neighbouring sites discovered through outlinks to include in the corpus and in turn crawl. I gradually developed the idea that I would continue this iterative process as long as I was discovering new food-related websites from Scandinavia. Eventually, after months of work, I ended up with a data set of 2007 websites that I had manually vetted as being both on the topic and located in the region and from which I had scraped all the hyperlinks (as shown in Figure 6) and was able to search for the presence of particular indicator words (as shown in Figure 2).
This collection of hyperlinks and indicator words of course consists of onlife traces. They tell us something about who these actors are associating with (from Figure 1 we can already begin to discern two distinct communities), what they are talking about and when they do so (as shown in Figure 2). However, they do not reveal the meaning of those actions to us. They constitute something akin to what Malinowski began drawing up when he first arrived in the Trobriands, namely the readily observable stuff. They present us with a similar problem to that facing Malinowski, namely how to achieve a native interpretation of that stuff. The barriers that prevent us from achieving this present themselves in different ways. For the ethnographer committed to long-term fieldwork in an unfamiliar setting, they are typically manifested as a frustrating inability to become accepted by the group and understand the cultural codes that govern social interaction. For the digital sociologist, they are felt rather as a lack of context. Engaging with onlife traces involves an unsettling absence from the social situation of the informants.
Conventionally, in ethnography at least, it is the presence in the field that affords the researcher an opportunity to gauge whether he or she is making sense of the world on the same terms as the informants. That affordance is somewhat lost when working with onlife traces. We do not feel the inability to make sense directly as a problem in the same way as, for instance, Clifford Geertz described it when, together with his wife, he tried to establish rapport with their Balinese informants: “As we wandered around, uncertain, wistful, eager to please, people seemed to look right through us with a gaze focussed several yards behind us on some more actual stone or tree” (Geertz, 2005: 56). The fact that researchers working with onlife traces are not, in this way, faced with the consequences of not making sense arguably compounds the meaning problem. This is because there are no immediate social repercussions to remind the researcher that the problem exists. One could easily (in fact is highly likely to) gaze at the tightly connected cluster of Danish mushroom pickers on the left of Figure 1 and imbue it with meaning, for example that mushroom pickers respect each other’s advice and thus associate intensely, without checking such an interpretation with the mushroom pickers themselves.
There is a striking similarity, however, in the disconnect that both ethnographers and digital sociologists experience between the ease with which they can chart and observe traces of social life and the difficulties involved in achieving a native perspective on whatever has been charted and observed. Indeed, the similarity extends to the ways in which we attempt to overcome those difficulties in our analyses. By taking us back to Malinowski, I deliberately gesture towards an inflection point in the disciplinary history of social anthropology at which in-depth, qualitative work came to be seen as the solution to the meaning problem. Ethnographers like Malinowski were brought up as part of a larger and far more quantitative collection effort organized, among others, by the Cambridge school of anthropologists under the leadership of Alfred Cort Haddon, Charles Seligman and William Halse Rivers. In the late nineteenth and early twentieth centuries, considerable scientific effort was devoted to the systematic collection of ethnographic field data, for example through standardized anthropological questionnaires (Munk & Jensen, 2015; Urry, 1972). We could thus say that Malinowski embodied the moment at which quantitative survey methods and qualitative participant observation intersect in anthropology. Digital methods are arguably in the middle of a similar moment, when a combination of qualitative and quantitative work is broadly construed as the solution to the meaning problem. However, whereas in social anthropology this moment came to signify an almost all-encompassing shift from surveys and questionnaires to in-depth interviews and participant observation, the study of onlife phenomena seems to offer new and interesting opportunities for fusing and merging the qualitative and the quantitative. What makes onlife phenomena fascinating in this respect is their potential for being both data traces and venues for qualitative work at the same time. One of the widely cited promises of digital sociology is thus the capacity to bridge the quali-quantitative divide (e.g. Latour et al., 2012).
My purpose in this article is to explicate some of the rather different ways in which that might happen. We know from mixed-methods research that the marriage of qualitative and quantitative methods raises questions about how we collect data, what we do with those data once collected and how we can think more philosophically about those processes (Johnson et al., 2007; Tashakkori & Creswell, 2007). Is the quali-quantitative potential to be found in the fact that onlife traces are often both quantifiable (counting shares and likes or developing network metrics for interaction) and qualitatively rich (comprising unstructured text, user tags, profile data and the ability to situate within online communities)? Is it to be found in the way in which we engage with code and numerical data in more ethnographic or exploratory ways? Does it lie in our newfound capacity to make datascapes (i.e. interfaces that make messy and unstructured data navigable) available for exploration by our readers or our informants (Girard et al., 2016; Venturini et al., 2017)? Alternatively, is it to be found in the questions that we ask about the media environments that constitute “the field”?
Drawing on my example of the New Nordic web corpus, I identify four styles of quali-quantitative analysis that could all be said to address the meaning problem in their own specific way. The first, which I dub
Four styles of quali-quantitative analysis
Interpreting insights from a quantitative analysis of onlife traces by situating these insights qualitatively in the everyday environments that they claim to reflect. | |
Tracing how quantitative patterns on the macro level emerge directly from qualitatively rich interactions on the micro level, demonstrating how onlife traces simultaneously embody both qualitative richness and quantifiability. | |
Critically reappropriating (and thus manually curating) onlife traces to speak on behalf of certain phenomena or address certain questions. | |
Attempting to reveal emic modes of ordering the world through quantitative techniques, for example pattern recognition. |
The second style, which I dub
Finally, as a matter of discussion, I propose a fourth style of analysis that I call
The most obvious way to understand the hyphen in the quali-quantitative is to let it signify a complementary relationship (Blok & Pedersen, 2014). If that is the case, then the quantitative and the qualitative can be allowed to play their separate roles, performing work that is already familiar to us, and leaving it to the researchers to argue from situation to situation how they can enrich each other. The gist of such an argument is often that in-depth ethnographic sensemaking is necessary to complete the picture in a world where machine learning and social physics have become ever more omnipresent. The most straightforward way to approach the quali-quantitative as a question of complementarity is thus through a division of labour. As Tricia Wang put it in a now much-cited blog post:
Big Data reveals insights with a particular range of data points, while Thick Data reveals the social context of and connections between data points. Big Data delivers numbers; thick data delivers stories. Big data relies on machine learning; thick data relies on human learning.
To demonstrate how big and thick data can work in mutual support of each other, I take my mapping of the New Nordic Food movement as an example. One of the first explorations of the web corpus consisted of a query. I simply searched the data set for mentions of terms relating to New Nordic Food. In this way, it was possible for me to show not only how the usage of the concept had migrated into different contexts but also how it compared to tangential concepts, like organic, biodynamic, locally produced, slow food or terroir. Figure 2 shows a graphic elaboration of the full corpus of 2007 websites clustered into different communities. The clusters in the network have been replaced by coloured polygons, and a semantic heatmap is overlaid on top. The heatmap shows where a given query term is found and how often (Figure 2 compares “new Nordic” with “locally produced”). It thus visualizes the results of a quantitative analysis (clustering by modularity and force vector layout, heatmaps by word count) of onlife traces (hyperlinks and text).
This quantitative analysis is interesting on its own merits, because it prompts questions. Notice for example that the Swedish cluster (coloured brown on the left side of both maps) not only has far more talk about local production (the map on the right) than anything new Nordic (map on the left) but that the left half of the cluster makes no mention of new Nordic at all. Websites in this part of the cluster are typically producers, local authorities and other local food actors; websites in the inner parts of the cluster, where the New Nordic Movement is cited, are typically governmental or otherwise national in scope. In the Norwegian cluster (coloured red at the top of both maps), the pattern is similar (there is far more talk about local production), although here the New Nordic Food movement is cited specifically by a group of actors in the region of Trønderlagen.
A clear pattern thus emerges when we map onlife traces at scale, and it seems to be telling a story about the emergence of a new Nordic agenda and about the pre-existence of various other tangential food agendas, but we cannot discern the plot of this story or begin to understand its characters without further work. We can of course continue to refine the mapping. There is still a range of more quantitative, data-related questions to pursue. We need to consider how to normalize the raw occurrence of a query term relative to the size of the website on which it is found, for instance, and we could triangulate the findings, through other online media and/or other query methods, to corroborate that there is indeed a pattern. Being able to do so, however, will not provide an interpretation of why the pattern is there.
This is where complementary qualitative work becomes useful. Prompted by the questions raised by the maps in Figure 2, I decided to visit some of the regional food actors in Trondheim. In the offices where my first interviews took place, I noticed a framed version of the
Later that day, while visiting a monthly food fair in Trondheim, I talked to a range of local producers and experienced a very different story. Be they dairy farmers, reindeer herders or the woman selling smoked whale meat from the fjords outside the city, none of them were the least familiar with the concept of New Nordic Food. Rather, they talked about their products as small scale and locally sourced. It reminded me of something Aslaug had told me earlier that day. Small-scale local production has always been a key part of the Norwegian agricultural policy. From Kristiansand in the South to Krikenæs in the North, the Norwegian Government works continuously to ensure that food can be produced locally, even if it would be cheaper to import it from abroad. This is a policy that is about national and local self-sufficiency in a country that stretches more than 25,000 km, all the way north of the Arctic Circle. It has kept local production at scales that would have been unimaginable in Denmark, where most small-scale producers were merged into larger and larger cooperatives in the latter half of the twentieth century. Trondheim’s local production has a completely different history and a different set of connotations from that in Copenhagen. Whereas the New Nordic Food movement in Denmark has come to signify a transition back to local food production, in Norway it is really more a question of rebranding small-scale producers and encouraging them to grow into viable export businesses that would help to develop the economy. New Nordic is seen as a development policy rather than as a culinary resuscitation project. This provides an explanation for why the concept does not (yet) occur in the discourse of the Norwegian producers on the maps in Figure 2.
To think about the quali-quantitative as a complementary relationship thus reflects, in the case above, a way of interpreting the findings from a quantitative analysis of onlife traces through additional qualitative work in the field. Although much mixed-methods research has generally adhered to this idea of existing methods complementing each other’s strengths and weaknesses, it is worth noting that the sequence in which it happens is typically reversed. In what Creswell called “exploratory sequential” research designs (Cresswell, 2014: 6), it is an initial qualitative analysis that produces questions and hypotheses, which can then be tested and affirmed through, for instance, a quantitative survey. The meaning problem, as it arises specifically in relation to the study of onlife traces and historically in much ethnographic work, tends to (or at least can very easily) reverse that sequence. The issue is not (only) how qualitative findings can be made more robust through additional quantitative work but how quantifiable signals were intended and how emic sense can be made of them.
When the data set is a corpus of websites, opportunities also arise for performing qualitative work directly on that data set. Even though the complementary division of labour that I outlined above is clearer when the qualitative work takes place in the field and the quantitative operations take place on data, the fact is that many onlife traces are both deep and big at the same time. A website can simultaneously be a place to conduct discourse or image analysis, count words or hyperlinks and, if it hosts a forum, a place to hang out and interact with users. This could be leveraged simply as another way to let qualitative and quantitative approaches complement each other. Qualitative work on websites provides depth to patterns detected in data from these sites. Many virtual anthropologists thus make use of automatic data capture tools to support their otherwise qualitative exploits online (Boellstorff et al., 2012).
Arguably, however, the simultaneous depth and breadth that characterize onlife traces also offer an opportunity to overcome the quali-quantitative divide. In making this argument, Bruno Latour (2012) echoed a decade-old actor–network theoretical dream of an analysis that does not a priori separate the world into a micro and a macro level (compare Callon & Latour, 1981 with Latour et al., 2012). He argued that onlife traces are rapidly making such a dream possible:
The best proof that those two levels do not correspond to any real ontological domains is that they begin to disappear, to be literally redistributed, every time one modifies or enhances the quality of access to the datasets, thereby allowing the observer to define any actor by its network and vice versa. This is exactly what the striking extension of digital tools is doing to the very notions of “individual” and “wholes”. The experience (more and more common nowadays) of navigating on a screen from elements to aggregates may lead researchers to grant less importance to those two provisional end points. Instead of having to choose and thus to jump from individuals to wholes, from micro to macro, you occupy all sorts of other positions, constantly rearranging the way profiles are interconnected and overlapping.
Data-intensive methods are the means to an end here, not a separate mode of analysis. They make it possible to aggregate and disaggregate interactions without introducing a simplified macro level (e.g. “the social”, Latour, 2005) that bears no direct empirical connection to these interactions (which are then confined to their own micro level of analysis). It was Gabriel Tarde (2015) who first proposed the idea of a so-called “monadic” sociology. As opposed to Émile Durkheim (who eventually supplanted him), Tarde refused to resort to abstract referrals to social laws that must be assumed to exist
The difference between a complementary approach and a single level of analysis approach can again be illustrated through the example of the New Nordic web corpus. One of the interesting features of this corpus is a particular group of tightly connected websites that display an intense interest in seaweed. A complementary approach could notice this feature on the macro level through quantitative techniques and then attempt to make sense of it on the micro level through in-depth qualitative work. Figure 4 shows a screenshot from one of the websites in question. It allows us to situate the seaweed talk in context, in this case on the profile page of a participant in the OPUS research project who nicknames himself “Seaweed Thorkil”. We can for example begin to build an interpretation that involves arguments about the health benefits of seaweed (which makes it particularly interesting to actors carrying out research on the New Nordic Diet) and its value as a sustainable food source.
For a single-level analysis, however, the profile page of Seaweed Thorkil is not so much an opportunity to interpret the meaning of a pattern as it is an opportunity to show how this pattern is wholly contingent on work that happens at the micro level (thus blurring the micro/macro distinction). The pattern that we detect at scale can be traced in part to how Thorkil shares hyperlinks, how he writes about seaweed and how the websites of projects like OPUS make these sharing practices possible.
The American anthropologist George Marcus (1995) reminded us that the ethnographer is not only in the business of providing thick descriptions but should also take responsibility for positing the specific logics of association that define the space of research (the context within which cultural analysis takes place and makes sense of the world). For too long, he argued, this task of delimiting the field has been allocated to third parties in the academy, for instance historians, geographers or economists, conveniently allowing ethnography to maintain its self-image as the branch of social science that provides depth and meaning rather than overviews (maintaining the quali-quantitative divide, we could add). Marcus was drawing on, and tuning into, a growing body of anthropological scholarship that was questioning the isomorphic conflations of “place”, “culture” and “field” that would more or less automatically prescribe a geographically bounded space of research, even in a world where flows and interconnectivity had become the order of the day (e.g. Appadurai, 1988; Olwig & Hastrup, 1997). Without attempting to cover the wide range of interconnective circumstances under which this isomorphism was found to be failing, it suffices to say that the advent of electronic mediation, and the internet in particular, was definitely among them.
Marcus foresaw that a set of methodological anxieties would be likely to result from anthropologists abandoning their spatially confined, physically locatable, single-site notions of the field. They were, in order of appearance, 1) that the limits of ethnography would be tested, because the attempt to grasp some wider system or context than a specific cultural locale situated within it might turn out to be better served by methods imported from other domains of the social sciences; 2) that the power of fieldwork would be attenuated, because the intensity of the ethnographic experience and the commitment to classic ethnographic virtues would suffer from the researcher’s newfound mobility and ensuing lack of embeddedness; and 3) that the subaltern would be lost, that is, that the traditionally privileged position in ethnographic accounts of those placed outside the hegemonic power structures of the world system would gradually be eroded by attempts to “study up” or otherwise “complete the picture” (ibid.).
Marcus reasoned that multi-sited ethnography would nonetheless be necessary – inevitable, in fact – even if there was ground for methodological concern. He pragmatically asserted that much contemporary anthropological research was already
If we decide on a strategy of following hyperlinks and use that strategy to trace the contours of an ethnographic field, then we quickly run into questions about the nature of such links. Indeed, hyperlinks themselves, as social phenomena, have not escaped ethnographic attention in the past (Beaulieu, 2005; Beaulieu & Simakova, 2006). Figure 3 shows a subset of the web corpus, namely a network of hyperlinked sites crawled from the profile page of Seaweed Thorkil. It makes it clear why careful consideration must still be given to the context in which links are provided and why the analysis of hyperlinks for this purpose cannot simply be automated but must retain a qualitative component. Besides the links shown on the screenshot of the page in Figure 4, we have to deal with links to ad services and web infrastructure and the fact that our tool, Navicrawler, is not capable of subdividing web entities like Twitter or Picasa into individual accounts (a feature that is now common in most crawlers).
The concept of web crawling stems from the early days of the internet when search engines were not yet available and the only means of online navigation was to browse from one website to another. A crawler is essentially an algorithm that automates this way of exploring and mapping the web. It works from a seed list of known starting points from which it follows all the available hyperlinks and draws up a list of neigh-bouring sites. From this new list of starting points, it then reiterates the process until, in theory, it ends up having indexed every connected site on the web. In practice, however, crawlers cannot possibly keep up with the pace at which the web is evolving, so they are continuously at work (re-)indexing websites to make possible, for example, the now taken-for-granted operation of a search engine. However, web exploration with a crawler cannot be reduced to this perpetual, indiscriminate and automatic undertaking. It can be restricted, put on a leash, to focus deliberately on a region of the web, typically found in a specified link distance from a meaningful seed list. This has proven useful for websphere analysis (Schneider & Foot, 2005) or issue mapping (Marres 2015) and hinges on what Rogers (2009) called the “re-purposing” of online media, in this case websites and hyperlinks, for social research.
In this version of quali-quantitative work, the quantitative mapping of onlife traces needs qualitative curation to be meaningful. We are, as Rogers (2018) put it in his call for “critical analytics”, obliged to think critically and reflexively about the way in which we appropriate natively digital traces. Tools such as crawlers or scrapers can be unleashed and run automatically, programmed to follow the shortest paths or the most-cited nodes, but these purely quantitative means of curation would not have been able to generate a New Nordic web corpus. This much is evident from Figure 3. By following hyperlinks indiscriminately and without further qualification from the profile page of Seaweed Thorkil, we would, in one iteration of the crawl, have been in the realm of Google, Picassa, Twitter, Blogspot and shortly thereafter the entire top layer of the web. Qualitative curation is necessary, and Navicrawler was designed specifically for this task.
If we intend to use hyperlinks (or any other digital artefacts) as a metric for meaningful associations in relation to a theme like New Nordic Food, we cannot appropriate them uncritically. Here we can draw on the work undertaken at the interface between media studies and STS (e.g. Marres & Gerlitz, 2016). The quali-quantitative as critical metrics is about understanding that seemingly technical choices about algorithms and data tools are themselves worthy of social inquiry. What is a hyperlink, for instance? In what context? At what point in time? Those questions must be pursued in tandem with any effort to reappropriate hyperlinks to perform social research. In mixed-method research, this is often thought of in terms of sampling strategies. Probabilistic approaches to sampling, which in our case could be a decision to construct the web corpus by letting the crawler proceed
Finally, as a point of discussion, I would like to suggest a fourth possible interpretation of the hyphen in quali-quantitative work. It takes a radical approach to the computational turn and claims that some data-intensive methods, such as network analysis and especially algorithmic forms of community detection and pattern recognition, can themselves be understood as a form of qualitative reasoning that has noteworthy affinities with how ethnographers work in the field. Ethnography, at its core, is an explorative undertaking. It is, as James Spradley noted, as much about discovering relevant questions from informants as it is about discovering the answers to them (Spradley, 2016). Big data methods have often been critiqued for their lack of theoretical commitment. Similar critiques have been raised about ethnography, which can rarely claim to be testing hypotheses and is rather in the business of revealing how native theories order the world.
One of the most interesting outcomes of mapping the New Nordic Food phenomenon through hyperlinks was that it jeopardized my assumptions about relevant empirical differences in relation to gastronomy in Scandinavia. Since the topic was about geographical typicality, about landscapes and terroirs, it seemed reasonable that geography would also define the way in which actors associated online. Indeed, this turned out to be the case most of the time. I have for example explored above how a Norwegian cluster of websites talks about locally produced food. In other instances, however, it was not possible to label a cluster of websites by their nationality or regional affiliation. This was the case for beer brewers, mushroom pickers, wine producers, foragers and schnapps makers (some of the smaller clusters in Picture 2). The websites committed to these practices were more interested in associating with others that were interested in the same theme than they were in associating with others from the same area.
If I had discovered these differences in the association of food practices in the course of an interview or a stint of participant observation, I would have been very happy to count it as an ethnographic result. It is precisely these types of insights into native orderings of the world, which arise from following the actors on the ground, that constitute the objective of much ethnographic work. As Michael Agar put it: “Traditional social science is on the lookout for variables; ethnographers are on the lookout for patterns” (Agar, 2006: 109). However, in this particular instance, it was not the ethnographer who found those patterns. On the contrary, they were detected by running a spring-based layout algorithm on the network of websites comprised in the web corpus. ForceAtlas 2 (Jacomy et al., 2014) is designed to push nodes apart unless connected by edges, which then act as springs to hold the nodes together. The result is a visual landscape in which nodes are placed in relative proximity to other nodes to which they are densely connected and at a relative distance from nodes to which they are not connected. The algorithm is indeterminate, meaning that it never ends with exactly the same visual result. The same applies to the Louvain modularity algorithm used to colour the network in Picture 2 (Blondel et al., 2008; Lambiotte et al., 2008). This algorithm attempts to find the most optimal way of cutting the network into smaller components, severing as few edges as possible in the process. Again, there is no exact and reproducible result. It could be said that these types of algorithms work exploratively to find patterns in ways that are in fact rather similar to those that an ethnographer would use.
In mixed-methods research, Greene referred to explorative analysis as an area in which certain kinds of quantitative and qualitative analysis seem to have a lot of overlap: “in particular, across quantitative (postpositivist) and qualitative (interpretivist) studies. This strategy may yield ‘stories that converge’ or discrepancies that invoke fresh perspectives” (Greene et al., 1989: 257). Indeed, machine learning and data mining are, in general, ways of training algorithms to spot patterns. Although not compatible with conventional deductive statistics, it follows many of the principles laid down by, for example, John Tukey in his work on explorative data analysis (Tukey, 1977). Ethnography is in a similar predicament vis-à-vis its siblings in the social sciences. It is often in a situation in which it does not presume to know the theory of the world from which to generate hypotheses or ask questions and must thus first build a native worldview from the bottom up, exploratively and inductively.
I began this article with the frustrations of Bronislaw Malinowski, who had mapped all the observable aspects of culture in his Trobriand village without coming any closer to getting the “hang of tribal life”. I compared that frustration to the meaning problem in digital methods, in which it is often not clear how we are supposed to interpret onlife traces, even if we can relatively easily accrue such data traces in large volumes. I have proposed four ways of mobilizing quali-quantitative responses to that problem. In the complementary style of quali-quantitative analysis, the meaning problem is both cast and solved as a qualitative matter. Again, following Tricia Wang (2013), numbers need stories! That position leaves little doubt as to the division of labour between the quantitative and the qualitative components of a piece of research. If we instead adopt a single level of analysis, however, this division of labour is immediately compromised. The meaning problem is no longer about explaining the intentions behind onlife traces but about understanding how macro phenomena like web corpuses are enabled by interactions on the micro level. The way in which actors make associations can be followed in hand-held, qualitative ways or it can be tooled up and pursued quantitatively with a web crawler, but the basic