[Authors and titles are listed at the end of the review.]
This is the eighteenth volume in the series “Oxford Studies on the Roman Economy” that Alan Bowman and Andrew Wilson launched in 2009, and arguably the most innovative to date. It sends a strong message that computational modeling has now sufficiently matured to make a substantial contribution to our understanding of Roman economic history, and even more importantly to our understanding of how to study that history.
Tom Brughmans’ opening chapter is a model of clarity, addressing the question “Why simulate Roman economies?” He makes a powerful case for the value of computational simulation, defined as “a formal model representing the operation or behaviour of a system or process where the functioning of the model over time is made explicit” (p.5). This approach, he argues, forces us to make our assumptions explicit, enables us to test them in a controlled manner and generates explanations and predictions that illuminate key dynamics of historical processes and offer guidance for future data collection.
Brughmans’ confident emphasis on the importance of explanatory hypotheses and the heuristic simplification of complex systems serves as a much-needed antidote to the self-defeating affectations and anxieties that bedevil much of the academic humanities, with their commitment to complicating narratives, reverence for supposedly irreducible complexities, denunciation of reductive reasoning, skepticism about causal relationships, and distrust of research strategies tainted by association with the social sciences. He reminds us that formal modeling and simulation impose accountability by allowing the falsification of claims that might otherwise go unchallenged. All of this is of relevance for scholarly practice well beyond the themes covered in this volume: Brughmans’ chapter ought to become required reading in any graduate course on the means and methods of studying the ancient world.
Ten case studies follow, eight of them concerned with issues of trade and transport and the other two with demography. Among the former, two chapters deal with transportation networks as such. Pascal Warnking, in a convenient summary of some of the points made in his much richer yet less accessible book on Roman maritime trade, demonstrates how a sophisticated commercial software tool for modeling sea voyages can be put to use for analyzing Roman-era connectivity.[1] His observations about the extent to which modern views regarding the ability of Roman ships to sail against the wind determine simulation outcomes are well taken.[2] However, his attempt to derive a new cost model from the pricing of sea travel in Diocletian’s Price Edict is called into question by the fact that his simulation tool of choice could handle only 20 of the 49 reported routes.[3] More generally, his contribution raises the question of how to harness the capabilities of proprietary software priced at $1,295 per license for public-facing scholarship without somehow replicating it from the bottom up (cf. p.63), a question that points to larger issues of funding and IT support for simulation projects in a relatively resource-poor field such as Roman studies.
The other chapter on transportation, by Pau de Soto and César Carreras, draws on their extensive labors in constructing an admirably fine-grained transportation network for the Iberian peninsula that encompasses more than 41,000 kilometers of Roman roads alongside navigable rivers and coastal sea routes. Their chapter is a bit of an outlier, light on simulation but heavy on chronological narrative and description. The present payoff may seem somewhat modest compared to the efforts that went into this project. Then again, by throwing light on the connectivity advantages of regional centers and on the relationship between the distribution of traded goods and transportation costs, their model gives us a glimpse of future promise. Its principal value lies in opening up new avenues for future research, which makes us wonder how accessible this and comparable tools will be to the academic community. It would have been helpful if the authors of this chapter as well as others had addressed this question head-on.
Exchange of material objects dominates the other six chapters on connectivity. Hanson and Brughmans explore the effect of settlement size on trade and economic integration in the eastern Mediterranean with the help of an agent-based model, modified from previous work and employed to make sense of the dissemination of ceramic tablewares. What seems to me the most intriguing result is the fact that their simulation matches observed outcomes reasonably well only for the period from 50-100 CE but fails to do so for 25 BCE-50 CE and 100-150 CE (p.126-33). The authors defer explanation of this divergence to future work. In another chapter Brughmans and associates continue their analysis of tableware from the same region by testing the hypothesis that merchants were influenced in their choices by the strategies of other traders, which, if true, would suggest that they enjoyed access to reliable commercial information, which in turn might be considered a sign of economic integration. The simulation results speak against this notion, producing the best match for a scenario of independent learning by agents and the poorest fit for copying others. While these dynamics await further investigation, for now this is bad news for proponents of a highly integrated Roman market economy.
But that is only part of the story. In a complementary chapter, Shawn Graham describes a model he and Scott Weingart had developed to test a particular vision of the functioning of the Roman imperial economy, Peter Bang’s concept of the “Roman bazaar,” which foregrounds information uncertainty and market segmentation.[4] Their agent-based model seeks to operationalize these and related premises in a simplified manner. The results are ambivalent, suggesting that these features are at the very least insufficient to account for observed outcomes. At the same time—and therein lies the main value of this study—Graham enthusiastically confronts thorny questions about how rich or parsimonious models need to be to test Bang’s ideas and how to go about quantifying the assumptions of a verbal model. Those are key issues that must be resolved if we are to bridge the gap between the latter, which represents the default mode of argument in the field, and computational simulation.
Xavier Rubio-Campillo and María Coto-Sarmiento turn to the western provinces, not so much by running actual simulations but by subjecting the visualization results of exploratory data analysis to statistical significance testing. What stands out is their test of four competing hypotheses about the provision of olive oil to the city of Rome against distribution data of amphora stamp codes. Their findings are most consistent with a self-organizing free market, but also support the notion of strong concentration processes driven by the wealthiest landowners, two options that are of course perfectly compatible.
Brian Dermody and associates revisit an earlier model that marries climatic and connectivity data with landcover reconstructions and demographic estimates to assess environmental constraints on grain cultivation and trade across the empire, an exercise that turns out to be useful not least for highlighting shortcomings in some of the underlying datasets. Finally, Mark Groenhuijzen shows that, on the Lower Rhine frontier, routing supplies for military camps through intermediate distribution centers would often have increased efficiencies (p.264, 266). Readers are left wondering whether other criteria, such as a desire to track, manage and protect such goods, might also have shaped the transportation system.
But enough of trade and transport. As Marek Vlach observes in his highly ambitious and original paper on the Antonine Plague, “epidemics represent an ideal problem for computational modelling, as their dynamics have long been quantified and mathematically described in the natural sciences” (p.76). He devises a two-tiered cell-based simulation model of the entire Roman empire that takes account of population distribution, transportation infrastructure and environmental factors that likely affected the spread of smallpox, which he – in line with most modern scholarship – considers the most plausible pathogen behind the pandemic.[5] Simulating impact for different levels of transmission, his model generates a wide range of outcomes in terms of mortality and survivorship. We see that variation in population densities would have caused the pandemic to display considerable regionality, hitting the most densely populated areas the hardest. Because of this, the city of Rome was bound to experience massive attrition even if transmission rates were quite low whereas many regions would have gotten off much more lightly. Recurrences would have been common and likely to trigger economic and political problems that destabilized the empire.
On a much smaller scale, Philip Verhagen explores the ability of the Batavian territories on the Lower Rhine to supply all known Batavian military units with enough recruits. His demographic simulations regarding the effect of recruitment on population size suggest that this region was probably unable to shoulder this burden on its own.
Andrew Wilson concludes the volume with a survey of different types of computational modeling from network analysis to GIS and agent-based modeling, noting that advances in software quality and accessibility have brought down costs to entry. A key challenge that remains is to formulate questions and hypotheses that can be addressed with the data we have (p.311).
What are the next steps? This volume is without any doubt a milestone in our engagement with computational modeling in the context of Roman history. But, as befits a milestone, it is also just one marker on a road that stretches far ahead. What we need is not just more cases studies but also more integration. The contributions to this volume show why. While Chapter 3 features a relatively basic road map of the Iberian peninsula (p.84 fig. 3.7), Chapter 8 offers a far superior one (p.228 fig. 8.1; a discrepancy briefly noted in chapter 3 at p.83 n.78). Chapter 3 boasts a massive map of 8,640 Roman sites (p.80 fig. 3.3) whereas Chapter 4 uses a much smaller sample of better-documented sites (p.120 fig. 4.2). Both Chapters 3 and 4 offer estimates for population densities and numbers derived from their separate datasets but Chapter 7 is indifferent to these estimates and employs yet another dataset (p.205-6).
This is not meant as a criticism: In pursuing their projects, separate teams inevitably came up with their own samples, and a one-size-fits-all approach might not even have been optimal for all these different simulations. Even so, I agree with Wilson (p.318-9) that going forward, integration and standardization will be of the essence, rescuing researchers from the need to reinvent the wheel every time they launch a new venture and giving the audience a clearer sense of what is under the hood. This process needs to be cumulative, building firmer and broader foundations and expanding scope and capabilities.
Several big issues are at stake – technical, thematic, organizational, and political. In the first category, we must ask how AI tools may be incorporated into these kinds of simulations. As for the second, we need to find ways to break free from the notion that such models might be suitable only for certain questions that are intimately tied to particular types of archaeological datasets, most notably ceramic remains (which feature prominently in several of the chapters), or other features that are readily susceptible to quantification, such as road networks, or other “quant-heavy” niches such as demography. While I understand Wilson’s emphasis on the usual suspects of agriculture, trade and population (p.312-6), simulation approaches cannot fulfill their true potential unless we accept and employ them more widely. To give just a single example, it has already become possible to replicate with an impressive degree of fidelity the spatial dynamics of large-scale state formation across much of Afroeurasia over a period of 3,000 years by means of simulations derived from a mere handful of parameters.[6]
The third and fourth dimensions – organizational and political – closely intersect. It should go without saying that computational modeling cannot simply be added to the ever-growing list of skills that students of the ancient world are expected to master. Brughmans briefly addresses this issue in the introduction, noting that one would like advanced students “to be familiar with the existence and basics” to the extent that they are able to assess and incorporate such findings (p.27-28), but he shies away from considering tradeoffs. One obvious and indeed perhaps the only real solution is to make collaboration more central to academic training and practice, especially when it comes to ancient history, which in this respect keeps trailing far behind archaeology.
This immediately leads us to the profoundly political question of how to evaluate this kind of scholarship. The current system of academic training, recruitment and promotion is not well equipped to recognize work that is routinely collaborative, may result in electronic outputs rather than traditional deliverables, and is not overtly focused on the monograph as the basic coin of the realm. All that makes it hard to reconcile with norms and expectations that are deeply entrenched in the academic humanities, most notably in the United States where institutionalized individualism and fetishization of the little-read book rule supreme. Academic incentive structures will need to be tweaked in favor of collaborative and non-traditional work to give simulation studies a chance to flourish. I am grateful that, at my suggestion, Andrew Wilson gives voice to these concerns at the end of his concluding chapter (p.319-20, with p.319 n.32). Yet any such calls can only be the beginning of what is bound to be a real struggle: no-one has ever lost by betting on academic inertia.
The objectives are clear: to expand computational simulations into a wider range of thematic areas, and to normalize their use as much as possible. The payoff promises to be considerable: novel pathways of analysis as well as consilience with social science fields in terms of research design and hypothesis testing. This is bound to be a long march. For now, the present volume serves as a model of how to move forward.
Authors and Titles
Introduction
- “Why Simulate Roman Economies?,” Tom Brughmans
Case Studies
- “Simulating Roman Maritime Trade: Modelling Sailing Times and Shipping Routes,” Pascal Warnking
- “The Antonine Plague: Evaluation of its Impact through Epidemiological Modelling,” Marek Vlach
- “Settlement Scale and Economic Networks in the Roman Empire,” J. W. Hanson and Tom Brughmans
- “Copying of Economic Strategies in Eastern Mediterranean Inter-regional Tableware Trade,” Simon Carrignon, Tom Brughmans, and Iza Romanowska
- “New Approaches to Old Questions: The Exploration of Large-scale Trade Dynamics Using Hypothesis-testing Frameworks,” Xavier Rubio-Campillo and María Coto-Sarmiento
- “A Model of Grain Production and Trade for the Roman World,” Brian Dermody, Alexander Chiu-Smit, and Rens (L. P. H.) van Beek
- “The Economic and Social Evolution of the Iberian Peninsula as Revealed through Analysis of Roman Transport Infrastructure,” Pau de Soto and César Carreras
- “Evaluating Hypotheses about Local Transport Systems through Spatial and Network Analysis: The Dutch Part of the Lower Rhine Limes and its Hinterland,” Mark R. Groenhuijzen
- “Modelling the Basics of Roman Demography: The Case of the Dutch Limes,” Philip Verhagen
Discussion
- “Mapping the Landscape of Ignorance,” Shawn Graham
- “Positioning Computational Modelling in Roman Studies,” Andrew Wilson
Notes
[1] Pascal Warnking, Der römische Seehandel in seiner Blütezeit (Rahden: Verlag Marie Leidorf, 2015), drawing on http://www.expeditionmarine.com.
[2] For the parameters of the “Orbis” connectivity simulation model (https://orbis.stanford.edu), which differ from Warnking’s, see most recently Scott Arcenas, “Mare ORBIS: A Network Model for Maritime Transportation in the Roman World,” Mediterranean Historical Review 36.2 (2021): 1-30.
[3] This constraint is not mentioned in the paper, only in Warnking (n.1) 278.
[4] Peter F. Bang, The Roman Bazaar: A Comparative Study of Trade and Markets in a Tributary Empire (Cambridge: Cambridge University Press, 2008).
[5] P.72. For doubts, see now Timothy P. Newfield, Ana T. Duggan, and Hendrik Poinar, “Smallpox’s Antiquity in Doubt,” JRA 35.2 (2022): 897-913.
[6] James S. Bennett, “Retrodicting the Rise, Spread, and Fall of Large-scale States in the Old World.” PLoS ONE 17.1 (2022): e0261816.