Theory and method in anthropology: an historical speculation

[John McCreery]

The following remarks are an edited version of something I just wrote on Savage Minds. The topic is method or, from my perspective, the lack thereof in interpretive anthropology.

------------

@Rex

I’ve never thought this was a problem related to cultural data or to anthropology’s method of interpreting it.

In this respect you are, I suspect, typical. You are quite correct to point to a

whole cottage industry in anthropology that worries over interpretive excesses

but can you say that its worries have been taken seriously? My impression that they haven’t been is admittedly based on personal impressions, not systematically collected data. But since our conclusions can be no stronger than the data on which we base them, I’d like to see some if you have it to offer.

Meanwhile, let me offer a few frankly speculative conjectures, based on my reading of classic and later, mostly British, social anthropology.

(1) Malinowski and other authors of classic monographs were acutely aware of the difficulties of interpretation. Their standard critique of Tylor and Frazer, et al, was that they imposed conjecture on fragmentary data torn out of interpretive context, resulting in what Evans-Pritchard labeled “If I were a horse” stories.

(2) The structural-functionalist embrace of Durkheim’s “social facts” was, with important implications for fieldwork practice, a way of distinguishing the social, what everyone seemed to say and do and the visible structuring of spaces (homes, palaces, temples, villages, cities) within which they said and did it, from the personal (the private feelings of the unformed child, the village idiot, the women, or others), discovered in behavior or comments that appeared to be idiosyncratic. The social was defined as the domain of anthropology; that other stuff left to the psychologists and psychoanalysts.

(3) The result was a practice in which cross-checking with multiple informants and sifting through data to separate the essential from the accident were standard procedures. All perfectly consistent with humanistic scholarship articulated and argued in terms of Aristotlean/Thomist logic.

(4) The world changes. The grand stereotypes embodied in classic theory are challenged on both moral and empirical grounds: morally because it is no longer seen as proper to exclude the voices of the children, the women and others regarded as minors and dependents and lacking the authority to make definitive pronouncements; empirically because growing concern for subjectivity and agency shift the focus of analysis from social facts (the stuff that everyone is supposed to take for granted) to individual opinions and choices and, thus,  from generic similarities to the range and distribution of individual variation.

(5) Here, however, anthropology confronts a problem shared with other humanistic disciplines. A solid majority of its practitioners hated math in school and have found all sorts of snarky reasons for denigrating statistics to provide excuses for their ignorance. As a result, we lack the technical know-how to address individual variation in anything but an impressionistic and moralistic manner in which—now I am being snarky too—what informant X happened to tell me on a day I was paying attention is conflated with an interpretation of “CULTURE,” of which informant X, for no demonstrated reason, is taken to be typical.

(6) The result can sometimes be the production of deeply moving stories. Take, for example, Ruth Behar’s Translated Woman. The writing is gripping and thought-provoking. But in what sense is Behar’s Mexican friend typical of any category to which we might assign her? To see her as typical of women oppressed by poverty and the macho habits of the men in her life is probably not wrong. But her toughness? The initiative she demonstrates in forming and maintaining a relationship with Behar? Are these generic features of her social type or exemplary features of a rare and idiosyncratic response to the hardships that shape her life? We shall never know until someone does a study of a population of individuals like her, to see if she lies close to the mean or represents an outlier in the distributions of the qualities we attribute to her.

In sum, the breakdown of the classic division between social fact and personal idiosyncrasy that enabled anthropologists of earlier generations to sort through their data and focus their attention only on the common features of “society” or “culture” breaks down when subjectivity and agency are concerns. It is time to stop whinging and moaning or simply asserting the superiority of our trained intuitions and develop methods appropriate for addressing our new concerns in a serious manner.

That’s my take on where we are. Others closer to the game should be able to show me where I’m wrong. In a scientific spirit of openness to falsification, I await your comments.

[John McCreery]

The best description of anthropology I know is one I steal from Lévi-Strauss, "the science of the concrete." The best anthropology is never abstract theory alone. It lives in the concrete details of the lives we study and the projects we undertake. Here is how I describe my current research in an introduction (still at draft stage) for the book I hope to write one day.

My first encounter with social network analysis was in the late 1960s, while still a graduate student at Cornell. I no longer remember precisely whose work I was reading. It does seem likely, however, that it had to do with the Rhodes-Livingstone Institute research in Central Africa directed by J. Clyde Mitchell. Mitchell’s students were all involved, in one way or another, in studies of individuals who had migrated from tribal homelands to cities in search of employment opportunities. In this new setting, their behavior could no longer be attributed solely to the implications of tribal institutions. How they formed ties and interacted with co-workers from other tribes as well as their own had to be considered. According to Lin Freeman’s history of social network analysis, it was Mitchell who recognized that a series of studies on superficially different topics shared this structural core (Freeman, 2004:4).

At this point, however, I was unaware of social psychologist Jacob Moreno’s work in the early 1930s on small group dynamics and his invention of Sociometry, the identification of social stars and later key opinion leaders whose interest becomes the tipping point at which new trends take off. I was also unaware of developments in graph theory, the branch of mathematics on which network analysis tools are based. As I then imgined it, social network analysis was, in effect, synonymous with the study of what sociologists call “informal organization,” social relationships that escape the boundaries of formal organizational structures. It may have been some notion of the importance of informal relationships that led to my seeing Chinese ritual as more about establishing and manipulating relationships than simply affirming corporate group or territorial boundaries. At this stage, however, my use of the concept of social network analysis was only metaphorical. It was neither empirically grounded nor solidly quantified.

In any case, interest in social network analysis receded into the background and became a path not taken until 2007, when casting about for a new research project, I noticed that, while teaching a seminar on the making and meaning of advertising at Sophia University, I had acquired a small collection of recent volumes of the TCC広告コピー年鑑 (TCC koukokukopii nenkan, Tokyo Copywriters Club Advertising Copy Annual). The annual, published every year since 1963, contains the output from the club’s annual ad contest. Every year, several thousand pieces of advertising are submitted as entries to the contest. About ten percent of the entries make it into the annual, divided into roughly a dozen industrial categories (the exact number of categories has changed a bit over the years). Each and every ad in the annual comes with a set of credits that list not only the medium in which the ad appeared, its sponsor and the agency and/or production houses that paid for and produced the ad but also the names of the individuals who made up the team that created the ad and the roles for which they are given credits. It occurred to me that if the credits were put in a database, I could use social network analysis to see which individuals worked together in winning teams, identify key figures and trace their careers over time.  I could also map how how their relationships changed over time and explore how these changes reflected the development and overall structure of the industry. I would have a solid framework on which to hang historical and ethnographic data.

Thus it was that an overall plan for the project took shape. It would start by creating a database for the credits data,  studying social network analysis and learning available software. Meanwhile, on a separate track, loomed another task. The key figures in the networks examined in this are often authors in their own right. They have written or contributed chapters to books in which they expound their views of how advertising works, what makes a good idea, what goes into a good presentation, how the work should be produced, team-building or leadership. They are also frequently interviewed in an active trade press, asked to comment on current topics of interest to the industry. The publications in question produce hundreds of pages each month and are available in series that go back for half a century or more.  For this part of the task, the anthropologist would have to be an historian and had a lot of work to do. Finally, there was the holy grail. Having done the social network and historical analysis, the ethnographer might be able to interview some of the people in question, to hear what they had to say, to correct or add to his findings.

[John McCreery]

Thanks for clearing this up, Jacob.

[Jacob Lee]

I might be wrong, but Pajek, the very popular software for social network analysis, was originally developed with genealogical analysis in mind.

Regarding kinship studies: not to nitpick, but I would want to emphasize that Dwight Read's contribution to kinship studies is not especially a contribution to genealogical mathematics, which had already been well developed. Instead, his contribution has been in the development of a formal framework for the analysis of kinship terminologies which does not depend upon an a priori set of genealogical relations of dubious cross-cultural validity. The principal subject of analysis is the set of relationships between terms, not genealogical categories. For example, the relationship between the terms 'brother' 'father' and 'uncle' in a sentence like,'The brother of my father is my uncle.' can be analyzed without recourse to genealogy, kin-types, etc. at all, e.g. as expressing a relation between symbols 'uncle' = 'brother' of 'father' of 'self', where 'of' is an operation on kinship terms. How these terms map onto, or fail to map onto, sets of genealogical categories is a distinct problem. Hence, Dwight Read's program can be viewed as a positive (and structuralist) response to the critique of kinship of the 80s initiated by David Schneider's A Critique of the Study of Kinship (1984).

[John McCreery]

@Huon

In my experience, people involved in social network analysis are pretty generous about acknowledging contributions like those made by Rivers. Anthropologists like Kim Romney and Dwight Read keep the flame of kinship studies alive by working on the mathematics of genealogies and, while now largely neglected by other anthropologists, are highly regarded in the network analysis community.

@M. Izabel

The hopelessly intricate tangles in question are called "hairballs" in the jargon of network analysis. Their complexity arises from multiple and overlapping paths connecting nodes that are, in most cases, of only one or two types. They are complex as a tangle of barbed wire or an intricate macrame are complex—not because the connected nodes or the lines that connect them are diverse.

[M Izabel]

I find this application of Network Theory very interesting.

"Over the past two decades, Barabasi and other researchers have developed a sophisticated theory of networks that helps make sense of what look at first like hopelessly intricate tangles. Network theory is allowing scientists to understand how networks produce unexpected kinds of behavior you wouldn’t be able to predict from looking at individual parts, from the remarkable robustness of the Internet to the sudden crash of financial markets."

http://e360.yale.edu/feature/network_theory_a_key_to_unraveling_how...

[Huon Wardle]

best to say Rivers formalised the genealogical method around 1910 because his use of the basic idea goes back as far as the Torres Strait expedition in 1898. He uses it extensively in The Todas where he managed to map the relationships of every member of that Indian community. Perhaps it is time to feed his Toda data in a network map. We can add though that Rivers emphasis on genealogy has been accused of reflecting the class assumptions he brought to the field.

[Huon Wardle]

First, I agree with Keith that John has done a fine job here. Sorry for conflating NT and NA. I should say that anthropologists have been using some methods akin to the ones described here for a long time. WHR Rivers developed his genealogical method around 1910; it consisted of asking participants 'give me a list of your relatives' , mapping the terms used and the persons indicated, then cross referencing to create a kinship diagram.  Some societies (to use Sahlins case since that springs to mind) such as 'big man' centred groups show something like the power law and cascading failure too. Others like hereditary Polynesian chiefships, do not since the hereditary rule itself and other principles limit it. In other cases the mechanisms for preventing build up of debt are highly wrought; so Gibson's study of the Buid shows how they strictly avoid public gift giving and instead leave anonymous gifts. So, we do have in anthropology models to think about equality and inequality not to mention developmental cycles in which equality and inequality feature in different ways.

The point I wanted to explore with John is whether equality is somehow only a matter for humanistic inquiry. It seems odd if it is; at the same time it would nonsensical luddism to say that using the new tools should be avoided on that basis (though equally, based on Rick's list, the uncontestable findings of NA may need some further thought).

We have been led to believe for more than a century that the bell-curve is preponderant in the physical world and this has helped to make prevailing social ideologies ‘natural’. European societies still largely hold to these ideologies. But the Americans have long held that income inequality is inevitable and today even the radical democratic wing of internet society, the bloggers and the peer-to-peer activists, tend to accept the fact of power-law distributions, claiming that as long as choices can be made freely (equal opportunity), this inequality is acceptable, one might say ‘natural’ or even ‘normal’.

Arguably, the bell curve fits with a carefully orchestrated class society - as in the British grammar school adoption from Plato whereby some people were 'bronze' some 'silver' and some 'gold' and that each contributed in their own important way. Likewise, the power law does seem prima facie to have an affinity with the politics of the Tea Party, the raging debate over healthcare reforms etc. where at issue is the idea of self-sufficiency and the right of every individual to succeed or fail without collective interference.

 

[John McCreery]

Thanks, Keith. Very informative indeed. I wonder, though, if the following description of graph theory is entirely correct.

Specialized study of networks in social science arose in the 1950s as a result of the development of graph theory in mathematics. The assumptions of this theory are now revealed to be unrealistic. It described an inventory of nodes whose number is fixed and remains unchanged throughout the life of the network. All nodes are taken to be equivalent and are linked together randomly. These principles of randomness, stasis and equivalence were unquestioned for forty years. 

It is true that the first applications of graph theory to social networks involved small, static networks. This limitation was not, however, inherent in the mathematics, where such questions as what happens to network density as the number of nodes approaches infinity (it approaches zero as a limit) were being posed early on. From my perspective, the early focus on small, static networks reflects two factors: (1) model-builders' tendency to begin with small "toy" models, which are easier to think about, and (2) the practical difficulty of assembling and analyzing large data sets prior the spectacular advances in computing of the last couple of decades. The combination of fast, easy-to-use computing and big data have created possibilities for research that were, if not unthinkable, not at all feasible just a few decades ago.

I recall (I think I may have mentioned this before) reading a methods textbook back in the 1960s, which included a  four-cell table illustrating possible research strategies. It wryly observed that sometimes asking just one question of one subject can be very important, "Will you marry me?" for example. It then noted that, practically speaking, researchers were forced to choose between asking many questions of a few subjects or a few questions of many subjects. Only organizations like Gallup had the resources to conduct omnibus surveys with hundreds or thousands of subjects. The lots of questions of a few subjects strategy was best for exploratory research, e.g., ethnography, when what might become important questions was still unknown. The few questions and lot of subjects approach was best for hypothesis-testing, the assumption being that the relevant questions had already been refined to the point where testable hypotheses could be defined. With these as their only options, most scholars chose one or the other. Historians, art historians, literary scholars, and ethnographers tended to choose the exploratory strategy, trying to puzzle out what the questions should be, based on their use of particular, limited, and often highly fragmented data. Meanwhile, economists, sociologists and political scientists tended to choose the hypothesis-testing strategy, using what Andrew Abbott calls the Standard Causal Approach (normal curves and regression analysis). 

The developments of the last half-century, however, have made it possible to do exploratory research using big data and to test hypotheses (at least those for which big data provides evidence) on the fly. The result, or so it appears to me, is a world in which the humanist's argument that the hypothesis-tester's hypotheses are too stereotyped and simplistic is no longer as strong as it once was, and the hypothesis-tester's claim that the humanist's research produces nothing of positive value for hard-nosed realists who want to know how the world really works is also losing its grip. I fully expect that within what is left of my lifetime, smart students will come to regard arguments confined to this frame as hopelessly old-fashioned. They will be figuring out ways to combining big data and humanist insight to produce new understandings of human behavior that will no longer be confined to elite or average perspectives blinkered by purely individualist or attribute based analysis.

[Keith Hart]

I am very grateful to all participants in this thread, but especially to John, for returning one corner of the OAC to engaged free-form intellectual debate. I have been interested in network theory for all my adult life and posted a reflection on its history here. To John's recent summary I would add two comments: that the last decade has seen a new approach to the statistics of networks and that this entails a shift to greater emphasis on their inequality.

If you count the book sales on Amazon and plot them according to frequency, the curve hugs the vertical and horizontal axes, indicating a few very large numbers (the blockbusters) and many small ones (the ‘long tail’ of books like yours and mine). This is a typical manifestation of something called a ‘power-law’ distribution. This is a relationship between the size and frequency of a variable, where the frequency decreases faster than the size increases. If the data are plotted on a log-log scale, the result is a straight line sloping down from left to right. Thus an earthquake that is twice as strong will occur four times more rarely. If this pattern holds for earthquakes of all sizes, it is said to ‘scale’, meaning that there is no typical size that could be said to be representative of earthquakes as a class of phenomena, as is the case with normal distributions.

The ‘new science of networks’, growing out of the physics of complexity, has been announced by authors such as Albert-Laszlo Barabasi (Linked 2002) and Duncan Watts (Six Degrees 2003). Just as, in the late nineteenth century, the normal distribution seemed to lend unity to statistical patterns emerging in a number of apparently unrelated fields, such as criminology, astronomy and plant genetics, now the power-law distribution appears in fields as disparate as the worldwide web, stock markets, air transport, Hollywood actors’ networks, electric power grids, urban hierarchies and molecular biology.

The very word normal says it all — conformity to a standard revealed by a central tendency, meaning that a population can be described in terms of an average type. The key assumption is randomness. This means that every member of a group has an equal chance of being selected. The democratic premise is obvious. This is an egalitarian as well as an atomistic model. Moreover, the quantities have to be measured on an interval scale, so that size is a continuous variable, not broken up into the separate classes of nominal or ordinal scales. Parametric statistics are cross-sectional data and fundamentally synchronic or static. Time-series are built on afterwards. Populations are expected to be bounded and knowable as such, much like the citizen body of a nation.

The power-law distribution is characterized by a few very large quantities and many small ones. In network science, it is commonly observed that networks consist of a few hubs with many links and a large number of weakly-connected nodes. The discovery of power laws is related to the physics of complexity, the attempt to study interconnectivity in a non-reductionist way (as opposed to the isolated atoms of the random universe). This science is mainly concerned with the edge between order and chaos and with critical moments of transition, as when chaotic water molecules assume the rigid pattern of ice. It is now thought that self-organization, including life, flourishes in this interstitial zone. Power laws thus describe open recursive processes without any of the bounded and synchronic assumptions built into parametric statistics.

Specialized study of networks in social science arose in the 1950s as a result of the development of graph theory in mathematics. The assumptions of this theory are now revealed to be unrealistic. It described an inventory of nodes whose number is fixed and remains unchanged throughout the life of the network. All nodes are taken to be equivalent and are linked together randomly. These principles of randomness, stasis and equivalence were unquestioned for forty years. Territorial states lent some credibility to networks configured in their own image. Thus road maps do not diverge markedly from the model, each centre having roughly the same number of links as the others.

People vary widely in their ability to make social connection and in this they resemble an air traffic grid, with a few O’Hares and many small airports. Networks were now seen as linking nodes of unequal size and depending on a few highly connected individuals. But what produces this effect? Barabasi established a fit between patterns of website links and the power-law distribution. These networks are ‘scale-free’ and lack the parameters of the normal distribution. There is no characteristic node in the continuous curve described by the power law which reflects the fact that networks grow over time. Skewed distribution of links may be accounted for by ‘preferential attachment’, so that growth with preferences accounts for the hub phenomenon (early-comers tend to attract more links) and undermines graph theory’s key assumptions of randomness, stasis and equivalence. There is an analogy with the market principle that ‘the rich get richer’. Indeed in the network economy ‘winner takes all’. The winner is often unpredictable until one node crosses a threshold and takes off. The trick is then to find the threshold. When hubs are weakened, the network as a whole may be visited by ‘cascading failure’.

The convergence of world markets and the internet has multiplied opportunities for scale-free networks. If corporate hierarchy was well-suited to the era of mass production for national markets (‘Fordism’), the rise of a web or network model of economy involves a shift from vertical integration to flat, virtual integration, as Castells (2001) has long insisted. Even if it can be shown to be regular, power law growth is unpredictable. Statisticians can only say that sometimes a variable crosses a threshold and then it takes off. We have been led to believe for more than a century that the bell-curve is preponderant in the physical world and this has helped to make prevailing social ideologies ‘natural’. European societies still largely hold to these ideologies. But the Americans have long held that income inequality is inevitable and today even the radical democratic wing of internet society, the bloggers and the peer-to-peer activists, tend to accept the fact of power-law distributions, claiming that as long as choices can be made freely (equal opportunity), this inequality is acceptable, one might say ‘natural’ or even ‘normal’.

[John McCreery]

Huon, I wish I could say that Network Theory has something useful to say about egalitarian trends. To the best of my knowledge it doesn't. That could just be ignorance on my part. 

One thing I have notice about our conversation is that we are treating "Network Analysis" or "Network Theory" as if  these terms always referred to the same type of scholarly activity. According to Alexandra Marin and Barry Wellman's "Social Network Analysis: An Introduction," there are at least three, overlapping but distinguishable, types of network analysis. 

Formalist theories are concerned primarily with describing the mathematical form of social networks (see Scott, this volume). These theories study the effects of forms, insofar as they are effects on the form itself, and the causes of these forms, insofar as they are structural. For example, when networks are composed of clusters of densely connected nodes with many ties within clusters and just a small number of ties between clusters, the result is a network in which short paths are available between most pairs of nodes (Watts, 1999). Because these theories are concerned primarily with pure form – in the mathematical, platonic sense – of networks, they can be studied without the need for empirical data. Mathematical modelling and computer simulations can be used to create networks that allow researchers to observe unfolding patterns of relations that result from particular rules of tie formation or dissolution....

Structuralist theories are concerned with how patterns of relations can shed light on substantive topics within their disciplines. Structuralists study such diverse subjects as health (Lin and Ensel, 1989; Pescosolido, 1992; Cohen et al., 1997; S. Cohen et al., 2001), work (Burt, 1992; Podolny and Baron, 1997; Ibarra, 1993), and community (Fischer, 1982a; Wellman and Wortley, 1990)....

Network Explanations....network analysts base their explanations on how particular kinds of networks or network positions can cause particular outcomes. We follow Borgatti et al.’s (2009) classification of network arguments into four categories: transmission, adaptation, binding, and exclusion (see Borgatti, this volume).

Structuralists, the second type described above, engage in four subtypes of research.

1. Defining key concepts in network terms. Wellman's redefinition of community in terms of people who provide support for each other instead of geographical boundaries is one example.
2. Testing existing theories.
   
   

   For example, Wilson’s (1978, 1987) theory of the underclass suggests that as poor African Americans have come increasingly to live in high-poverty neighbourhoods, they have lost connections to people who provide ties to the labour market. Their social isolation contributes to difficulties in finding work, and it hinders social mobility. Although Wilson’s argument speaks of network connections, the evidence presented is still group-based, treating neighbourhoods as monoliths that are connected – or not connected – to the labour market by virtue of the neighbourhood’s class composition. Further, by focussing on within-neighbourhood ties, the theory neglects the possibility of out-group ties providing connections to the labour market. However, the story may be more complex. Fernandez (1992) finds that the urban poor do have out-group ties to people committed to labour market participation, while Smith (2005) further finds that what African-American urban poor lack are ties to people in the labour market who are willing to offer assistance in finding jobs. By looking at real patterns of relations rather than assuming a lack of relations based on a perceived lack of opportunity, such research creates a stronger link between theory and data. The original theory – like many social theories that are studied nonetheless from attribute-based or group-based perspectives – is about patterns of relations. Therefore, the theory can be more validly tested using data on relations than data on neighbourhood characteristics.

3. Looking at network causes of phenomena of interest. 
   
   Network-based explanations of substantive outcomes are fundamentally different from explanations that rely on individual-level or group attributes. Social network analysts often have little tolerance for norm-based explanations, norms being precisely the kind of content that Simmel argued was outside the domain of social explanations. Moreover, when causal forces are presumed to be internal or possessed by individuals, the mechanisms frequently are internalized norms or atomized rational actors (Granovetter, 1985). Social network analysts argue that internalized norms are inherently a-social mechanisms. Therefore, treating such norms as the primary causal mechanism provides a-social or psychological explanations.
   
   
4. Looking at network effects of phenomena of interest. 
   
   My own work falls into this category. I am particularly interested in how changes in media have affected the networks of  ad contest winners I study.