Please don’t link ancient shipwrecks and the economy
Debunking everyone’s favourite graph about the ancient Mediterranean
There was a lot of humming and drumming a few months ago thanks to a tweet by everyone’s favourite self-proclaimed history understander, Noah Smith. The ifunny quality meme is best left to the medieval historians to lampoon, but the closing of ranks by Smith’s peers around the economic downturn experienced in the Early Middle Ages has inadvertently thrown light on an often misunderstood set of data. Quantian’s quote tweet includes a whole raft of graphs and the caption “noooo actually there was literature, the concurrent collapse in population, wages, trade, and consumption doesn’t stop the early medieval period from being a rich and dynamic culture you can’t say it’s a dark age.” However, the majority of the graphs attached are decidedly Late Antique in focus. One even ends in 380 AD, some thirty years before the sack of Rome by the Visigoths.
Each graph hails from the same source, ‘Health and Wealth in the Roman Empire’ by Jongman et al. in Economics & Human Biology. Published in 2018, the ambitious article presents a massive database of skeletal remains and interprets trends in relation to material wealth and personal health. The words “Dark Age” don’t appear at all, nor, indeed, does the article argue the Early Middle Ages wasn’t a dynamic or rich culture. In fact, the article’s conclusion links the poor health of Romans to levels of urbanisation. If nothing else, the collapse of the Roman empire in the West improved health outcomes for Europeans. Giving Quantian the benefit of the doubt, I’d say they don’t understand the nuances of the archaeological data they’re looking at and amplifying.
What really ticked me off, though, was the inclusion of the shipwreck graph (below).
This graph appears everywhere and is often blindly accepted as evidencing any number of different things. It’s clear that when Quantian mentions the “collapse” of trade, this is their primary evidence. And — again giving them the benefit of the doubt — this is not an uncommon view. James Lacy’s Rome: Strategy of Empire (2022) — a monograph all about the military strategy employed along Rome’s borders — argues it attests to the “enormous growth of the Roman economy” at the turn of the millennium (p.37). John Brooke in his monumental Climate Change and the Course of Global History (2014) argues that “the number of recovered shipwrecks” reflected a “surge of shipping” (p.331). Jongman isn’t innocent either, stating in his article that “shipwrecks per period [function] as a proxy for long-distance shipping” (p.147).
A Brief History of the ‘Shipwreck Graph’
Rather than blindly rehashing data, it’s important to understand where all this information came from and how it has been treated and updated over the past few decades. As we shall see, cataloguing shipwrecks is a relatively recent endeavour (one that continues to this day!), but the methodology behind these graphs and their interpretation is problematic, to say the least.
The very first incarnation of anything resembling the shipwreck graph appeared in a seminal article by Keith Hopkins on Taxes and Trade in the Roman Empire (1980). Hopkin’s main argument relates to how the imposition of taxes in money on Roman provinces stimulated inter-provincial trade. Proving such a causal link inevitably meant attempting to estimate how trade in the Mediterranean changed over time. The primary evidence Hopkins presents is a simplified graph of 545 dated shipwrecks grouped into 200-year intervals. The lack of resolution in the data is obviously frustrating and Hopkins’ appeals to the prevalence of piracy and warfare in the last two centuries BC as an explanation for the unexpectedly large ‘200–1 BC’ category rings with special pleading. Nevertheless, his assertion that “there was more sea-borne trade in the Mediterranean than ever before, and more than there was for the next thousand years” became a commonly held view.
While Hopkins’ specific graph hasn’t survived the test of time, the source of his shipwreck data certainly has. Hopkins was “most grateful” to a certain A. J. Parker for allowing him to borrow and publish his data ahead of its official publication in the British Archaeological Reports series. Parker’s catalogue had since grown to some 1,189 wrecks by the time of the publication of his latest supplement in 1992 (below).
The catalogue featured wrecks from over 30 countries and was divided more finely along single-century columns. The 1st century BC remains the zenith of Mediterranean shipwrecks but it is joined closely by the first two centuries AD. The extension of the X-axis to include both Bronze Age and Early Modern wrecks increases the visual impact of the numbers recorded during the Classical period. The wrecks are plotted according to the mid-point of their respective date ranges. For instance, a wreck dated between 50 BC and 25 AD is placed in the 1st century BC column. What’s more, there’s no separate column for “Roman of unknown date” as on Hopkins’ graph. Instead, for these 76 wrecks, Parker takes the mid-point of the broad “Roman period” and uses that to place the wrecks.
Hopefully, you can spot the methodological pitfall here — Andrew Wilson certainly did. By using mid-points Parker’s graph misrepresents the probable dates of many wrecks. Most notably, the “Roman period” wrecks (with a range of 150 BC to 400 AD) are given a midpoint of 125 AD and are all placed in the 2nd century AD column despite having an equal probability of wrecking in any year within the date range. The 2nd century AD column is unduly inflated as a result. Wilson’s new approach assumed an equal probability of wrecking in any year of the given date range and allocates shipwrecks accordingly. As such, instead of a whole wreck counting for the 2nd century AD column, the wreck is split into chunks and allocated to each relevant century (0.09 for the 2nd century BC and 0.18 for every other century to the 4th century AD).
In the final iteration of the shipwreck graph (above), Wilson not only corrected the distribution of ambiguously dated wrecks but increased the total catalogued to 1,646 shipwrecks. This graph, Wilson’s 2011 rejig, became the graph for anyone interested in measuring ancient maritime trade and is the graph referenced by Brooke, Lacy, Jongman, and (now) Quantian.
Linking Shipwrecks and Economic Activity
So Wilson fixed the graph? Well, yes — but also no. Wilson corrected the representation of the data but he also articulates the most substantial criticism of the graph’s interpretation to date. Namely, directly equating the fluctuations in the number of shipwrecks with trends in Mediterranean trade rests on two a priori assumptions.
1. “The probability of a shipwrecking is the same in every period.”
This first point can be attacked from multiple angles. First off, it doesn’t take into account the effects of innovations in sailing technology and practices over time. The bilge pump, for example, we believe was invented sometime in the 1st century BC and was a crucial piece of Roman technology that allowed water to be bailed out far more efficiently than before. The pump had massive implications not only for maritime safety but also for the tonnage of Roman ships going forward. In theory, ships without bilge pumps are more likely to sink than those that do, thus distorting the graph and warping our interpretations.
Moreover, in terms of sailing practice, it should also be noted that the pre-Roman norm of coastal sailing (cabotage) was over time replaced by more direct open-water sailing thanks to innovations in sails, navigation, and tonnage. While it is impossible to quantify the differences in wreck probability between open-water and coastal sailing, it is unwise to assume they are the same.
Second, the growing field of palaeoclimatology has much to say about the late Roman Empire and the centuries that followed. Not only was the Mediterranean subject to yearly climatic variability, but recent studies agree that the Roman Climatic Optimum began winding down in the 3rd and 4th centuries AD, leading to lower temperatures and deteriorating weather. What’s more, the Roman propensity to sail during the winter months again altars the risks and probabilities associated with wrecking relative to other periods. In lieu of these studies, it would be foolhardy to assume a perfect parity in sailing conditions across the entire period considered. Indeed, going as far as the Early Modern period would include a whole raft of climatic variables which are impossible to adequately integrate.
The final criticism of this assumption relates to human-based danger to ships. In this case, it is Hopkins, himself, who provides the ammunition. In his attempt to explain the anomalous shipwreck peak in the 1st and 2nd centuries BC, he claims that they “were more marked by piracy and warfare than the later period.” A threat, he goes on to argue, extinguished by the relative peace of the Pax Romana and the defeat of the pirate menace by Pompey. By his own logic, then, the likelihood of being sunk due to attack differed between centuries and must have influenced the probability of wrecking in a given period.
Each of these factors alters the probability of a shipwrecking in a given period. The difference in the number of shipwrecks could as much be a function of these unknowable factors than of the volume of Mediterranean trade.
2. “All wrecks are equally visible.”
This second assumption is demonstrably false.
First of all, we must understand exactly how an ancient shipwreck is preserved and identified. In both stages, the cargo is the most important factor. As the ship sinks to the seafloor, it typically comes to rest on its side with one-half of the hull hugging the sea floor and the other half exposed to the environment. Over time, the exposed side of the hull deteriorates leaving behind the mound of cargo, under which lies the preserved second half of the hull. Should the cargo survive, the wreck becomes a habitat for marine life and sits undisturbed as an innocuous mound on the sea floor. These mounds on the sea floor can be identified by sea-floor mapping and corroborated by archaeological dives.
As such, in order for a wreck to be preserved and identifiable, the cargo must be non-perishable. As Wilson notes: warships, grain freighters, slave ships, and any other shipwreck carrying perishable cargo are not preserved except in very particular circumstances. In Parker’s original data, nearly 70% of identifiable wrecks had a cargo of amphorae or pottery while construction materials, metal ore, and marble make up the rest. As such, the method by which many wrecks are preserved leads to the massive overrepresentation of certain types of trade and the near-total elimination of others.
“But if the ratio of non-perishable to perishable cargo stays roughly the same throughout, won’t the overall trends be the same?” I hear you ask. And the theoretical answer is yes. However, storage techniques and cargo types obviously changed over the course of the entire period. Some goods, like wine and olive oil (two of the most widely traded in the Roman empire), moved from being stored in perishable containers, to non-perishable containers, and back to perishable containers again. More importantly, though, we don’t actually know the ratio between perishable and non-perishable cargo.
The mighty wooden barrel, a common staple for the transport of goods and ageing of wine, came into use sometime between the 2nd and 4th centuries AD in the western Mediterranean, replacing amphorae. However, unlike their ceramic predecessors, wooden barrels could not survive a shipwreck nor serve to protect the wooden hull of a ship. In other words, the widespread adoption of wooden barrels would be expected to lead to the preservation and discovery of far fewer wrecks. For this reason, Wilson suggests that the decline in shipwrecks is “at least partly explicable by a move from amphorae to wooden barrels” (p.220).
Secondly, while we have established quite clearly that not all wrecks are preserved, it is also the case that not all surviving wrecks have been discovered. In fact, the significant regional variation in the data is based more on the geographical limits of modern marine archaeology than on the prevalence of wrecks, themselves. For instance, while it is true that Parker’s data represents over 30 countries, just three of them (Italy, France, and Spain) account for more than 70% of the total wrecks discovered. While Italy being surrounded by Roman-era wrecks is no great surprise, the prevalence of France and Spain over other parts of the Mediterranean should raise some eyebrows.
Moreover, the relative weakness of wrecks discovered in the eastern Mediterranean runs contrary to our expectations with regard to preservation as amphorae remained in use there for several centuries after the western Mediterranean adopted barrels. It’s clear that touristic diving efforts, international funding, and local politics all affect the extent to which Mediterranean coastlines have been mapped and trawled for wrecks. And that’s not even mentioning the limited mapping of deep non-coastal waters missing potentially far more Roman-era wrecks.
Even Parker, who created the original data set, never claimed that the number of shipwrecks correlated with levels of maritime trade. Indeed, his analysis points to an overrepresentation of well-documented coastlines (p.548) and non-perishable cargo (p.20). He was duly ignored by most economic historians.
Inconsistency with Land-Based Archaeology
Even during the periods of peak shipwrecks, in the High Empire, there are surprising inconsistencies between trends in the data and land-based archaeology.
Wilson raises an important point in his analysis regarding the massive overrepresentation of the wine trade between Italy and Southern Gaul (France) in the wrecks of the second and first centuries BC. The trade subsequently fell away in the following two centuries, reducing the overall number of shipwrecks. This phenomenon, however, would be consistent with “import replacement” in Gaul, with the growth of local viticulture mitigating against the need for imported Italian wine. As he concludes, “an apparent decline in [shipwrecks] may nevertheless be consistent with a model of economic growth.”
Moreover, in the 2nd and 3rd centuries AD, there is ample evidence for the massive redevelopment of port facilities at Portus (the gateway to Rome) ordered by the emperor Trajan and the intensification of export-destined olive oil presses and port infrastructure in Roman North Africa and Tunisia. New presses were appearing not just along the Sahael coast, but far inland on the rocky slopes of the Aures Mountains. The use of such marginal land for oleoculture implies healthy profits were being made and speaks to the ludicrous nature of the olive oil export market. Indeed, the appetite for African oil is reflected in the literary writings of both Juvenal and Pliny the Younger, as well as the changing origins of pottery found in the Italian port city of Ostia (below).
However, since the coastlines of Tunisia, Libya, and Algeria are less thoroughly researched than their north Mediterranean counterparts (mostly due to issues surrounding funding, tourism, and politics), there are far fewer shipwrecks to speak of, rendering the shipping invisible on the graph and prompting erroneous commentary about the “decline” of trade in the Mediterranean — at least in those centuries. This is especially surprising considering olive oil (specifically) remained an amphora-based commodity in this period. The conclusions initially drawn from the shipwreck graph are, therefore, inconsistent with the archaeological picture on the ground.
What’s more, there is again the spectre of barrels to contend with. Archaeology in the wine and oil-producing regions of Spain suggests production only really started to substantially drop off in the 5th and 6th centuries AD, however, despite this, wrecks containing amphorae from this region heavily declined by the 2nd century AD. What are we to make of this discrepancy? Provincial archaeology points to continued exportation while the shipwreck graph points to decline. The unfortunate lack of preservation for barrel-carrying wrecks is masking a potentially vast amount of Late Antique shipping in the Western Mediterranean.
Why do people care?
It’s more a question of why people don’t care, to be perfectly honest. Despite the myriad methodological and interpretive issues with the shipwreck graph, it still manages to wriggle its way into peer-reviewed journal articles, popular monographs, and innocuous Twitter spats. People like to claim it represents all kinds of things, though very few take the time to substantiate their views.
Why do people continue to use it? I can think of only two plausible reasons.
First, because there’s simply nothing better. Reconstructing the scale of Mediterranean trade two thousand years ago is a nigh impossible task. Can we really hope to do any better than using shipwrecks as a proxy? Not really, I don’t think. Indeed, when broken down into constituent regions, the graph can actually tell us a lot about particular routes favoured by ancient mariners or intervals of high-sea traffic, such as between Western Italy and Southern France at the turn of the second century BC. But does that mean we must accept the generalised conclusions drawn from the graph? No.
A second reason is that the graph just feels right. It makes sense. Based on all our evidence and literary sources, the peaks and troughs of the graph just make sense. It mirrors our expectations, prompting people to accept it outright as representative. This is the main reason I believe the graph is so widely cited, especially by individuals who haven’t engaged with the topic of the ancient economy before.
The Roman empire was a pan-Mediterranean political phenomenon. In Polybius’ words, the sea was “an organic whole” (1.3) linking events and people together in a way never seen before. The ecology of the Mediterranean already necessitated the constant redistribution of surpluses and people, but the Roman empire extended these lines of connection and made them archaeologically durable. Surpluses grew larger and needed transporting in durable containers on high-tonnage ships. The products went further, from the fields of Tunisia to the banks of the Rhine. It makes sense that trade peaked during the Roman empire and disappeared when the empire did too.
Similarly, it just makes sense that the Dark Ages would yield so few shipwrecks, because of the downturn in material living standards evidenced by our sources and archaeological finds.
Such a view, of course, betrays a highly Western Eurocentric undertone. The Roman empire and, later, Byzantium, continued on in the Eastern half of the Mediterranean for centuries. Constantinople was fed, much like the legions on the Rhine, by a constant stream of imported foodstuffs from Egypt and elsewhere. Trade continued and North Africa and parts of Italy were even brought briefly back into the fold of Justinian’s empire by the middle of the sixth century AD. This is hinted at in the graph, with the majority of those precious few late antique shipwrecks located in the Eastern Mediterranean. But these nuances are lost completely by those who choose to cite the graph from a Western European perspective.
Accepting the graph because it feels right is obviously no way to go about historical enquiry and I hope this blog post has adequately argued why.
Since writing this Noah Smith has come out with another mini-blogpost about the Dark Ages during which he commits the sin of citing the shipwreck graph (this time from Jongman 2007). Smith posits that those who do not believe in the Dark Ages ought to argue “that economic output is a poor measure of civilization,” but he does not make the case that the proxy data sets he uses are, in fact, indicative of economic contraction in the whole Mediterranean. All the more reason to write this post and encourage people to think twice before citing this misleading graph.