john@oac

Profound Contributions

In my previous post I wrote,

The major difference between Sunbelt and OAC is, of course, that most of the participants in the former are comfortable using network and other forms of quantitative analysis, in a subfield of social science to which mathematicians and physicists have and continue to make profound contributions.

What did I mean, however, by "profound contributions"? Until the rise of network science (and chaos and complexity defined mathematically), social science analysis was, in my experience, confined to categorical or statistical reasoning. Categorical reasoning employed a deductive logic essentially unchanged since Aristotle's Organon. Societies, groups, statuses, roles—all were conceived categorically, as entitles defined by necessary and sufficient conditions, sharply bounded and uniform. In statistical reasoning, categories are replaced by populations and samples and necessary and sufficient conditions by measures of distribution and inferences based on probabilities. Statisticians might note the usefulness of other distributions (Poisson and T, for example), but the normal distribution became the paradigm assumed by informal debate about probabilities. Linear regression, based on the assumption that effects result from the sum of causes identified by least-squares reduction, became the standard formal model to which quantitative social science aspired.

Network science, together with chaos, complexity, fractals and recursion (of which fractals are one example) profoundly transforms the assumptions on which the standard quantitative (normal distribution and recursion) model of social science is based. Independent cases are replaced by interconnected nodes, which, by definition, do not act independently. Power law distributions, e.g., the Pareto curve, and the hierarchical, hub-and-spoke network topologies they imply have been identified across virtually the whole spectrum of natural and social phenomena, from protein cascades in cell biology to transportation and power grids, the Internet and World Wide Web, and a host of social networks.

In my own data, on winners of a Japanese ad contest, I have already seen how reliable the new mathematics of networks is. Large networks tend to have a single giant component (a component is a subnetwork in which there is at least one path from each node to every other) with a scattering of, relatively speaking, tiny components as outliers. The larger the network the more likely it is that the giant component is also a giant bicomponent, in which there are at least two paths connecting every node to every other (a matter of concern to those who worry about the robustness of networks). The degree distributions in my data (degree being the number of immediate neighbors of a given node) are, indeed, power laws. The fit is uncannily close, far closer than the .05 or .001 significance levels commonly referred to in statistical studies. This is a social physics that works--and works with amazing accuracy, just as mathematical theorems imply that it should.

Are we forced, then, to imagine a world in which human behavior is totally controlled by the mathematical laws that network science reveals? Not at all. I recall a high school physics class in which the teacher remarked that the laws of electromagnetism apply to large numbers of electrons, not to predictions about individual electrons. The fact that, other things being equal, the law of gravity will cause objects to fall from sky to earth does not prevent humans from designing balloons, airplanes and helicopters that fly. The trick is to find and employ countervailing forces.

In a similar spirit, the discovery that, left to themselves, totally free markets will inevitably result in sharp polarizations of wealth and opportunity does not mean that this result is inevitable. Here, too, countervailing forces may be identified and used to ensure different outcomes. Progressive tax schedules and estate taxes may, for example, be used to prevent inequalities in wealth and opportunity from reaching extremes. Ditto for rules governing executive compensation. Working out the specifics may be as difficult as designing or improving the design of an airplane's wings. It is not, however, impossible.

Views: 13

Comment

You need to be a member of Open Anthropology Cooperative to add comments!

Comment by John McCreery on July 8, 2010 at 2:35am
Finally, and speaking more directly to Huon's question, the passage that he cites refers to what I think of as the tool-testing phase of my research. Much of the first two years I invested in the project were devoted to creating the Filemaker Pro database in which I transferred the credits data from the TCC Copy Annual and to learning the basic concepts and methods of network analysis that might apply to them. In articles like Newman and Ghospal (cited in the presentation attached to the previous message), I discovered strong claims based on graph theory, the branch of mathematics on which much of network analysis is based. A reasonable first step was to see if predictable patterns like giant components and bicomponents and power law degree distributions, validated with other data, also applied to my own. If they did, I thought to myself, the results would have a direct bearing on the way I thought about the industry. The alternative hypothesis to the giant components and bicomponents was the possibility, derived from the keiretsu view of Japanese industrial organization, that the data would include at least two large components organized around Dentsu and Hakuhodo, the two largest Japanese agencies, firms that dominate an oligopolistic market. It was also possible that winning ads and creators would be normally distributed and thus a random outcome. This was, in the final analysis, clearly not the case; the power law distributions predicted by basic network theory were clearly visible. Together these exercises convinced me that I am working with very sharp tools, indeed. They demonstrated the presence of structures that ethnographic analysis would have to explain or incorporate.
Comment by John McCreery on July 7, 2010 at 8:39pm
After looking at the presentation uploaded with the previous message, you may still be wondering how qualitative analysis/interpretation fits into this project. HowJapaneseAdvertisingHasChanged.rtfd.zip is an unpublished essay that illustrates how I am grappling with this issue.
Comment by John McCreery on July 7, 2010 at 7:18pm
Huon, you ask a question of "deceptive simplicity." The answer is rather complex. At one level the answer might be a desire to better understand an industry with which I have been closely involved for three decades but always involved in positions that offered only small glimpses of a larger whole. At another the answer is a desire to better understand how ethnography can be extended to include the large amounts of readily available historical and quantitative information in relation to which ethnographic insight must be tested in dealing with large and complex social phenomena. I am aiming, in other words, to come come up with a plausible answer to Clifford Geertz's observation in Islam Observed that the value of insights gained through microscopic experiences can only be demonstrated in larger conversations, conversations now complicated, as Marcus and Fischer note, by the fact that we now work in settings where, “We step into a stream of already existing representations produced by journalists, prior anthropologists, historians, creative writers, and of course the subjects of study themselves.” At yet another level, I am exploring the frontier that divides the sciences and humanities, developing a demonstration of how both inform a fuller understanding of a complex human reality. Like most anthropologists, I imagine, I enter the field with broad interests that are only preliminary orientations, knowing full well that new questions will emerge and new ways to answer them have to be found. The file I have just uploaded is from a presentation given a year ago at Zhengzhi University in Taipei, which illustrates the first steps in the journey: ZhengZhiUPresentation.pdf
Comment by Huon Wardle on July 7, 2010 at 6:08pm
What was it you were trying to explain about the contest winners, John? In other words, what was the question to which this is the answer?*

The larger the network the more likely it is that the giant component is also a giant bicomponent, in which there are at least two paths connecting every node to every other (a matter of concern to those who worry about the robustness of networks). The degree distributions in my data (degree being the number of immediate neighbors of a given node) are, indeed, power laws. The fit is uncannily close, far closer than the .05 or .001 significance levels commonly referred to in statistical studies. This is a social physics that works--and works with amazing accuracy, just as mathematical theorems imply that it should.

*I stole that formulation from Keith, by the way: Its deceptive simplicity comes in very handy in post-graduate teaching.

Events

© 2019   Created by Keith Hart.   Powered by