Duncan J. Watts. Six Degrees: The Science of a Connected Age. New York: W.W. Norton & Company, 2003, 368 pp., $27.95 paperback.
Review written by Martina Morris, firstname.lastname@example.org, University of Washington
This is an odd reviewing assignment. The book is not aimed at sociologists, or at scientists generally. So the usual reviewing criteria seem like Cinderella’s shoe on some hopelessly outsized foot: small, narrow and precious. The audience for this review, however, is sociologists with a smattering of scientists :-). What do you need to know about this book, that wasn’t written for you, in terms that matter to you?
First, that it is one of the only books I know that has ever attempted to popularize the core paradigm of the social sciences. Not the marginal, but immensely more popular “social problems” genre that end up sharing shelf space with self-help books. But the true heart of the social sciences: the aggregation problem. Without the aggregation problem, there is only psychology. The problem of creating a whole from some irreducible function of the parts--a social fact--is the raison d’etre of social science. But try explaining that to your friends. Or your undergraduates. I am forever thankful that my undergraduate sociology professor was up to that challenge. But it takes some guts to attempt this feat in the realm of vox populi. And to the extent that Duncan succeeds, we will all have something to be thankful for.
Second, that this book chronicles a remarkable amount of cumulative methodological progress that is being made on the hard problems induced by dependence and dynamics. It is chronicled in a way that is useful even for those steeped in network analysis, since so many of the new publications are in journals outside our field, and so many of the papers are beyond our grasp. It makes one long for an “advanced” version of the book, in which the methodology itself would be the central focus of explanation, and the assumptions would be revealed. The author has such a gift for clarity, it makes you want to learn more.
Third, that this is still the “great man” version of history being written. For those in the trenches, there are still too many lives unacknowledged, and too much revisionism in this book. The application of percolation theory to epidemics (pp. 175-7) has a reasonably long history--including the focus on shortcuts (e.g. Mollison 1977). The suggestion that the vast majority of network models ignore hierarchy (p. 274) ignores a fundamental theoretical root of network analysis--balance theory--and some of its most interesting branches (e.g., Chase 1982). The swipe at a centrality-obsessed social network crowd (pp. 51-2) ends up by re-discovering the various models of centrality, from degree centrality (scale free models) to betweenness centrality (the original “shortcut”). It’s good storytelling, but bad history, and one is left with the wish that this missive to the masses did not have to be quite so restricted to egonets.
And finally, the signature of the scientific paradigm--engagement with evidence--remains remarkably underdeveloped in this “new” field. Starting with the title (and pervasive theme) of six degrees--if only 18 of Milgram’s 96 starting letters were actually delivered to the target, what exactly *is* all the fuss about? The data speak unequivocally here: on the off chance you are linked to someone, the average path length is 6, but most of us (80%) are not effectively linked. So what is the point of validating simulation model against Milgram’s conditional distribution of links (p. 155)? Both the Kleinfeld revelation and the simulation happily coexist in these pages, unreflectively. In this, some ears will discern the classic split between mathematical and statistical perspectives. A disciplinary blind eye in what is otherwise a paean to interdisciplinary work. Others will note the explicit faith in top-down models--as more "interpretable" than local models that aggregate up (p. 82)--and ponder the importance of deductive reasoning.
Yet while the history may not be right, and the science may be pre-social, the voice is remarkably clear, and in the end the main message is more important than the details. This book poses a real challenge to the social sciences, to get their collective technical acts together. How to do this is an interesting question. But it will require, at minimum, either teaching real social science to undergraduates (so that they come to grad school armed with a theoretical background and we can focus on teaching them good methods) or enticing mathematics BAs into the social sciences (so we can focus on teaching them social theory in grad school). It will also require a pretty much wholesale revision of our graduate methodological training. We should be integrating the methodological study of dependence into our most basic courses. That will mean teaching simulation in all of its many forms, and recasting the traditional statistical canon (linear model, independent observations) as a set of tools most appropriate for analyzing standard micro-level survey data. Of course, none of these ideas are new either. Back in the 1950’s when Columbia reigned supreme with its Bureau of Applied Social Research, one of the members suggested that survey research was the equivalent of grinding meat into hamburger, then trying to reconstruct the cow (the quiet but perceptive Alan Barton). Perhaps this is one of the reasons for the self destructive opposition between theory and methodology in our disciplines: they are at cross purposes. A set of quantitative tools that begins with the assumption of independent observations is an odd choice for social science. That it takes a physicist to remind us of this is even odder.
So if you have an undergraduate course, see if you can find a place for this book. It is a great lever for prying open many closed doors. If we do not start to open these doors, our history will continue to be written through the filter of a physicist’s net. There’s no point in whining about this. We just need to ante up.
Mollison, D. (1977). "Spatial contact models for ecological and epidemic spread." J.Roy. Statist. Soc. Ser. B 39: 283-326.
Chase, I. D. (1982). "Dynamics of hierarchy formation: the sequential development of dominance relationships." Behaviour 80: 218-40.