Order this book
Department of Sociology, University of Leicester
One complicating factor in reviewing this book is that I have a lot of sympathy for any kind of sociology that attempts to be rigorous and avoid jargon-ridden speculation. My criticisms are not intended to undermine the objective of rigorous sociology (which I strongly support), only to question the extent to which AS really exemplifies (or defines) such rigour. Leaving aside their labeling as AS, nearly all contributions to this book are well ahead of far too much sociology in terms of clarity and intellectual value. Unfortunately, however, profound and contentious debates about the value (and nature) of rigour in sociology cannot be avoided merely by rebranding. Despite what Manzo says in his extensive and thought provoking introduction, it really isn’t clear that most of the papers here are rigorous in a distinctive way (rather than just being more rigorous than average) that would distinguish them from many older examples of statistical, experimental or mathematical sociology with their associated demerits (over simplified assumptions, lack of empirical relevance and so on.) As such, the fight about what “good sociology” is remains unavoidably institutional and normative. It isn’t merely that AS is misunderstood (and several contributors come across as rather defensive on this point.) The reasons why thoughtful sociologists mistrust mathematical formalization are as cogent as they ever were even if many others happen to reject any attempt at formalization for no good reason. (Being critical of the angels doesn’t necessarily make you a devil especially in science!) One reason for this concern (and one has to be realistic about the extent to which edited volumes really represent a shared vision rather than just claiming they do) is that while Manzo is at pains to stress that AS needs to be understood as a whole set of interlocking methodological claims to generate its intellectual force, in practice he is happy to allow individual chapters to “focus” on only some of these (or less charitably ignore others) with the result that they don’t look much different in practice from existing forms of research. This is particularly true of the experimental chapters (by Grossman and Baldassarri and Takács et al.) The point is not that these are weak or uninteresting chapters, very far from it, but that reading them it is hard to see why they are any more analytical than most competent experimental work not described as analytical.
It is here that the contrast with ABM is instructive. Even though it is seldom carried out in its entirety (though there are “gold standard” exceptions like Hägerstrand 1965 and Abdou and Gilbert 2009), there seem to be no serious alternatives to the generative methodology laid out by Gilbert and Troitzsch (2005) and Epstein and Axtell (1996). However sloppy an ABM may be in practice, there is rarely any danger of mistaking it for a piece of statistical analysis, experimentation or field research and it is usually clear what should have been done to make the ABM rigorous even if it managed to get published regardless. (Although I think ABM can be “analytical” in a sense that AS could not but approve, in practice it very rarely seems to be for institutional reasons. The bar is simply not set high enough for publication in my opinion.) By contrast, it seems to me that most of the articles in the present volume could easily have fitted quite happily into Rationality and Society, JASSS, Journal of Mathematical Sociology or Social Networks without standing out as a special thing called AS. Unlike even quite unimpressive ABM, I am not sure that AS can yet pass a “blind taste test” as a well-defined field with a definite associated methodology. (The fact that the methodology is often ignored is no more an argument against ABM than bad statistics would be against the value of good statistics. Its logic is clear nonetheless.)
ABM being a relatively clear-cut method and methodology also avoids other forms of awkwardness that AS (at least as presented in the present volume) seems prey to. For example, Manzo argues that social networks and methodological individualism are important elements of AS. In ABM one just has to decide if such features are appropriate to modeling a particular domain and if not, leave them out. It doesn’t seem useful to have to say that a model of a riot (for example) can’t be AS because, as it happens, social networks are not very important in riots and there really does turn out to be such a thing as “collective hysteria” (which is not reducible to the individual level). It seems unwise (and more importantly unnecessary) for AS to shackle itself to positions on contentious and possibly irresolvable issues like methodological individualism. Interestingly, stripping away aspects of AS which seem to be more contingent than Manzo tries to argue the irreducible core of the methodology looks really rather like the generative strategy in ABM! (As stated much more concisely by Gilbert and Troitzsch and Epstein and Axtell back in the last century. More generally, though I don’t have space to develop the argument here, the present volume perpetuates the inaccurate view that ABM was developed almost exclusively by Americans and ignores the considerable independent role of European ABM research in developing many of the ideas presented here. Given that AS has now apparently decided that ABM is a key part of its approach, it seems surprisingly uninformed about what was happening in the field before it received the AS seal of approval.)
Another difficulty which illustrates an earlier point is that in several contributions, the contributors don’t end up being quite analytical enough in ways that actually seem to undermine the stated logic of AS. For example, Wikstrom’s contribution looks extremely promising, being based on extensive data (including uncommon data on space-time budgets) but it is odd to read at the start of his paper that unemployment cannot cause crime because variables cannot cause anything (although an ABM would say that unemployment was genuinely causal at the individual level perhaps arising from an income goal without legal means to realise it) and at the end of the paper to find a fairly traditional path model! (The present volume doesn’t really seem to reach a conclusion on whether “traditional” statistics can or can’t be AS: Manzo suggests not but that’s what several contributors actually do and the apparent inconsistency is not addressed.) Similarly, the article by González-Bailon et al. contains excellent data and intriguing analysis but can it really be true in a threshold model that the number of users with particular threshold values (as shown in Figure 10.2 on page 269) is so profoundly discontinuous? How would that come about? (It seems to me that sociological statistics can still fairly be criticised for thinking that lots of data can compensate for it not being the data actually required. Both chapters seem to illustrate this difficulty.) Finally, and this is a slightly different aspect of AS potentially undermining itself, but why is Mitschele’s carefully argued article on witch trials AS at all? Despite the quality of its argument, it not only seems to deal with a typically “historical” topic but to use exactly the style of verbal causal reasoning that seems to be traditionally favoured by historians. (Nevertheless, an “analytical” concern can also be raised about this chapter. If other areas also have surges of witch trials then there need to be other mechanisms than the quest for office at work, since it is claimed that this doesn’t seem to have operated elsewhere. But if this is so, how do we know which of these mechanisms actually caused the Scottish trials? On the other hand if trial patterns are different elsewhere then this article really only explains Scottish witchcraft. As such, is it really a mechanism, or just a standard historical explanation of a case? From an ABM perspective of course, it seems odd for Mitschele necessarily to be looking for single dominant causes at all.) AS may not make itself popular with other disciplines if it simply absorbs anything it happens to approve of (regardless of subject area or method) or condones the same weaknesses in its adherents that it appears to criticize in others.
Apart from the direct interest of the excellent simulation contributions to JASSS readers (by Rolfe, Gabbriellini and Fountain and Stovel, with Gabriellini getting closer than many ABM publications to the full generative methodology and arguably closer than any other contribution in the book), there are also some interesting missed opportunities for simulation. In a classic rhetorical move, Kroneberg proposes an interesting general theory of decision-making but then has to admit that to operationalize it with available data, he has to throw most of the interesting bits away. ABM would operationalize the theory and then, ideally, go looking for data that matched (or didn’t match) the aggregate outcomes. Similarly, the chapter by Abell might make a better ABM (with less restrictive assumptions which could be more empirically grounded) than a mathematical model.
Taken simply as a set of chapters, this collection is considerably better than average (with JASSS readers being likely to benefit from reading many chapters and at least skimming most). It is also well presented and organized and the editor has clearly taken his job very seriously with good results. In terms of presenting AS as a coherent research programme, I found it a lot less convincing. I’m not sure we can delimit “good research” in general terms and trying to do so risks creating a self-regarding club rather than a self-critical field. Rather we need to devote ourselves to the greatest possible rigour in statistics, ethnography, ABM or whatever we do and debate earnestly with the most thoughtful practitioners in other areas to see if we can bridge the gaps that exist between the different approaches (organizing around actual research problems rather than insubstantial epistemological or philosophical positions). Even given its many weaknesses (and my obvious bias) I have to say that (based on reading the present volume) ABM now seems to have a sharper and less cluttered vision of what it is up to and where its challenges lie than AS does.
EPSTEIN, J. and Axtell, R. (1996). Growing Artificial Societies: Social Science from the Bottom Up. Cambridge, MA: MIT Press.
GILBERT, N. and Troitzsch, K. (2005). Simulation for the Social Scientist, second edition. Maidenhead: Open University Press.
HÄGERSTRAND, T. (1965). A monte carlo approach to diffusion. European Journal of Sociology, 6(1), 43-67.
Return to Contents of this issue
© Copyright JASSS, 2014