The Collapse of Civilization in Southern Mesopotamia

Southern Mesopotamia was the cradle of civilization. The first great city, Uruk, was founded around 5000 BC, and the region remained the centre of the Akkadian, Assyrian, Babylonian, several Persian and the first Arab Empires. The region’s importance was due to its extensive irrigation networks, which sustained agriculture and population (Adams 1981). Six millennia of success ended in the second half of the ninth century AD, when Adams’ archaeological data reveal the almost complete collapse of settled population centres in this relatively short time frame (Adams, 1965, 1981).^{Footnote 1} Even today, this part of Iraq is completely deserted (Fig. 1).

Economic activity in southern Iraq in 2000. Note: The canals shown in yellow were the principal canals of the Islamic period. Many are now abandoned. The area around the former Grand Sarat Canal and the Nahr Nars is virtually deserted today, but it was densely settled in the ninth century. Compare the area around the Nahr Nars here with the corresponding area in Maps 2 and 3

Full size image

In this paper, we study the causes of this collapse, and test the hypothesis that the immediate cause of collapse was political, and involved factors—state failure, civil war, and social rebellion—that were disruptive across the globe (Blattman and Miguel 2010). While political collapse is common throughout world history, it was particularly destructive in Iraq because of its geography: Iraqi agriculture can be very productive, but only after heavy public investments in irrigation works. These investments—and their subsequent maintenance—require large-scale collective action or public good provision, both most easily accomplished through a centralized government. In absence of a coordinating authority, collapse of this infrastructure results in collapse of agriculture and depopulation (see, e.g. Wittfogel 1957).^{Footnote 2}

To be more precise, southern Iraq is a desert crossed by the Tigris and Euphrates rivers. When irrigated by their waters, the plain is very fertile. This geography raises three issues that are central to this paper. First, the irrigation system had to be constructed and maintained. Small-scale irrigation could be organized by local groups (Mabry 1996). As perfected during the Persian Sassanian Empire (224–621 AD), however, the system featured five giant transverse canals running from the Euphrates to the Tigris, a sixth running from the Euphrates into the Great Swamp, plus the Nahrawan canal which ran parallel to the Tigris on its eastern side. These canals allowed the full water flow of the two rivers to be applied to the land. The system was so large that it required the state to build it and maintain it. Second, the waters of the Tigris and Euphrates, while fresh, still contained small quantities of mineral salts. Modern irrigation systems have drains as well as water supply channels to remove salts from the land, but ancient Mesopotamia had no drainage system. Instead, the water supplied to fields evaporated unless it was taken up by the plants (Potts 1997, pp. 12–3). Over time the salt content of the soil increased. Farmers responded by planting barley, the more saline-resistant crop, instead of wheat, but agriculture suffered none the less. Even today, after the introduction of drainage, the ‘Shura soils [of southern Iraq] are often [covered] with a white crust’ of salt (Quereshi and al-Falahi 2015, p. 87). The question is whether this issue came to a head in the ninth century forcing the abandonment of cultivation and settlement (Charles 1988, Jacobson 1982, Potts 1997, Campopiano 2017). Finally, a complex system of irrigation requires a level of compliance from the population. The conflict of interest between peasants and the rulers who taxed them was mitigated by the need that each party had for the other—the ruler could not rule without an agricultural base to tax, and the peasants could not survive without the irrigation infrastructure provided by the ruler (Heldring et al. 2020). By the end of the ninth century, excessive taxation and collapsing irrigation led to peasant revolt and tax boycotts, which may separately have been responsible for the collapse of the irrigation system.

All three competing explanations require some form of political coordination to be resolved. Therefore, we argue that the immediate causes of civilization collapse were failures on the institutional plane. Adam Smith (1776, Book V, Chapter 1) listed the essential functions of the state—protecting the society from foreign aggressors, maintaining internal order and a system of justice to protect the weak from the strong, and ‘erecting and maintaining those public…works, which though they may be in the highest degree advantageous to a great society, are, however, of such a nature, that the profit could never repay the expense to any individual, or small number of individuals’. In the case of southern Iraq, the irrigation system was the pre-eminent example of such a ‘public work’. The Caliphate did not succumb to foreign invasion, but it did fail spectacularly in performing the other functions. We argue that it was the failure to perform the second function that led to the failure to perform the third.^{Footnote 3}

To support these claims, we present two new sources of evidence that, to our knowledge, have not been applied to the study of the collapse of Iraqi civilization before: A calibrated model of the ‘hydraulic’ economy and panel data regressions using data on tax returns. In a three-period panel straddling the collapse of 27 tax districts, we test for the presence of the collapse and its causes. We find a near complete collapse between 846 and 918CE, consistent with the qualitative literature. We then measure state capacity by the presence of weirs, a particularly important public good. We also measure salination and proximity to the source of most important uprising, the Zanj rebellion. We find that the collapse of civilization (measured by tax receipts on agricultural output) was much less pronounced in those districts where irrigation investment was reduced comparatively less. Salination or rebellion had no impact on tax revenues. Our model provides a clear interpretation of this effect. Faced with scarce resources, rulers whose power was threatened became more short-term oriented (in order to stay in power) and diverted resources away from long-term irrigation investments to short-term military investments. Our contribution is, therefore, in the introduction of these new sources of evidence, and the relation of the conclusions they facilitate to several hypotheses advanced for the collapse of the Iraqi countryside.

What precipitated the collapse and what threatened the ruler’s power? The timing of the collapse results from a change in politics. In fact, from the early ninth century, failures to maintain internal order and administer justice were manifest at the top and bottom of society. In particular, a shift in the organization of the military to the use of slave armies precipitated a crisis of political stability.^{Footnote 4} The first ‘slave army’ in the Islamic world was purchased in this period by Caliph Ma’tasim (833–842). At the same time, political turmoil erupted. There were twelve caliphs in the ninth century—and ‘depositions’ were common—six of the caliphs were murdered.^{Footnote 5} Why so many? The answer has nothing to do with revolts against the arbitrary exercise of unchecked power. Rather the explanation lies in the rule governing succession to the caliphate. In the Abbasid Caliphate, possible successors included brothers of the deceased caliph, as well as sons and grandsons. The two sons of Harun al-Rashid fought a civil war in the early ninth century without too much damage to the caliphate. However, when the three sons of Mutawakkil fought each other between 861 and 870, the result was anarchy that eventually did in the caliphate. The top-level institutional failure was applying the succession rule of a desert band to a great empire.

Settlement and Irrigation, 637–883. Note: The red dots denote villages occupied in the early Islamic period according to Adam’s The Land Behind Baghdad, while the green dots denote early Islamic settlements listed in Adam’s Heartland of Cities. These villages lie in his two ‘intensively surveyed’ regions of Mesopotamia

Full size image

The failure to maintain order in the upper reaches of society led to disorder at the bottom that also undermined the economic system. During the anarchy of the 860 s, the various contending factions were in desperate need of cash to pay troops. They resorted to various expedients to raise funds quickly. These included assessing the annual tax more than once a year, and tax farming where there were no limits on what the tax farmers could, in practice, take from the agricultural population. On occasion, the treasury was bypassed altogether, and the army was allowed to pay itself by pillaging the countryside. In 869 rural revolution broke out, and it lasted until 883. Peasants in southern Iraq no longer paid taxes, and that refusal reduced the economic incentive of the caliph to maintain the irrigation system, as well as raising the cost of doing so.

In conclusion, civil war at the top and revolt at the bottom meant that the state stopped maintaining the irrigation system. When caliphs were desperate to pay troops, dredging canals seemed of little consequence. When peasants revolted, they stopped paying taxes, and their raids and non-compliance increased the costs and lowered the benefits of maintaining irrigation. By the end of the ninth century, the system had collapsed because the political institutions proved inadequate to the environmental challenges. We do not claim to introduce the hypothesis that the collapse of the state is at the core of the collapse of civilization, a point made by, for example, Adams (1965). We similarly do not aim to add to the qualitative historiography on southern Mesopotamia. We instead bring together data on the agricultural economy in our model to provide a quantitative benchmark of the explanatory power of the collapse of the state, and we construct a new data set to be able to test the most common alternative hypotheses for explaining the observed collapse in settlement and irrigation against one another. Therefore, our contribution lies in distinguishing between competing hypotheses in a disciplined manner, rather than in generating new hypotheses or presenting new primary evidence.

An important alternative explanation is cultural. The cultural approach maintains that features of Islam militated against modern economic growth. Whatever truth there might be in this view, it cannot explain what happened in Iraq in the ninth century. First, this is a very long run theory, and its main concern is with the decline of the ‘scientific spirit’ in medieval Islam and the failure of the Islamic Golden Age to lead into the Scientific Revolution. Second, the philosopher al-Ghazali is the usual bête noire in cultural explanations, and he was born more than a century after the events discussed in this paper. Third, while the enthusiasm for philosophy among the Caliphs had its ups and downs, it was in the ascendancy at the end of the period we are concerned with. The Golden Age is said to have begun with the Caliph al-Mansur (754–75), who initiated the translation movement, and was pressed forward by Harun al-Rashid (786–809), who patronized philosophers and founded the House of Wisdom in Baghdad. This support was continued by al-Ma’mun (813–833) and al-Mu’tasim (833–842). Enlightenment lost out to religious conservatism under the reign of al-Mutawakkil (847–61), and all official support for any cultural activities disappeared in the years of anarchy (861–70) when one caliph after another was murdered, and the state barely functioned. However, stable government resumed with al-Mu’tamid (870–92). Caliph Mu’tadid (892–902) began again to support scholarship and science but at a reduced level reflecting his strained circumstances (Kennedy 2005, pp. 245–6). Unfortunately, he was the last effective caliph. Under al-Muqtadir (908–929), the caliphate was effectively bankrupt. Baghdad declined as the centre of Islamic scholarship, as other centres emerged in Persia, Central Asia, Egypt, and Andalusia whose rulers could afford to support it. The Golden Age ended in Baghdad not because it was overwhelmed by Islamic obscurantism, but because the Caliphs and their associates ran out of the money to pay for it. We explain why in this paper.

The paper proceeds as follows: The next section is a short history of the Abbasid Caliphate emphasizing fiscal, military, and agrarian issues, since these are the events that define the issues. Section 2 presents a model of hydraulic society in the short run that captures the key features of the Caliphate (Wittfogel 1957). In Sect. 3, that model is calibrated, and in Sect. 4, its implications for a macro-view of Iraq in the Golden Age are elaborated. Section 5 uses the model to measure the decline in output and population during the collapse of the late ninth century. Section 6 extends the model of Sect. 2 to include investments, and that extension underpins Sect. 7, which undertakes an econometric investigation of the causes of the decline. Section 8 concludes.

Soon after the death of the Prophet Mohammed in 632, Arab armies swept out of Arabia and conquered much of the Middle East, North Africa, and Persia. Iraq fell in 636–8. The Arabs did not follow the common practice of dividing conquered land into fiefs that were parcelled out among the nobility. Instead, the Caliph ‘Umar order that new cities—Kufa, Basra and Wasit—be founded for the Arabs tribesmen to live in. The Sassanian taxation policies were continued, and most of the money that was collected was paid to the Arab settlers as military salaries. The register of those who received payments and the amounts they were entitled to was known at the time as the diwan. Only 7% of the tax revenue went to the Caliph’s central treasury, 7% was spent on local administration, and the rest went to those on the diwan (Kennedy 2001, p. 71). This way of dividing up the agricultural surplus was in keeping with the egalitarianism of the nomadic Bedouin who had conquered Iraq. Nevertheless, Iraq became a tribute state in which the peasantry supported the militia of invading Arabs.

As the settlers grew older, they became less effective fighters, and as they died the question became whether their heirs could inherit a place on the diwan. The Caliphs resisted their claims because they wanted to hire professional soldiers. By the end of the rule of the Umayyad Caliph ‘Abd al-Malik in 685, the caliphate had won and controlled the diwan, which became synonymous with its treasury.

The Umayyad Caliphate with its capital in Damascus was overthrown by the first Abbasid Caliph al-Saffah in 750. The Umayyad Caliphate was dominated by Arabs and alienated many non-Arabs. The Abbasid revolt began in Khorasan, the vast eastern province of Persia where resentment against Umayyad taxation and administration was high. On the ideological plane, the Abbasids were descended from the family of the Prophet, which, they claimed, made them legitimate rulers. They constructed a regime that was more inclusive ethnically than the Umayyads had been.

The second Abbasid Caliph al-Mansur moved the capital of the caliphate to Baghdad, which he established in 762, and began to finance the translation of Greek philosophical and scientific texts into Arabic. The Golden Age of Islam was in full bloom during the reign of Harun al-Rashid, who was certainly the most glamorous Caliph, due to his presence in The Thousand and One Nights. He led armies of 100 thousand against the Byzantine Empire, patronized the arts and sciences, and founded cultural institutions like the House of Wisdom in Baghdad.

Harun al-Rashid could afford all this since his income was so high. Table 1 summarizes the tax receipts of the Abbasid Caliphate between 780 and 918. Generally, these figures are based on official sources and were compiled by state officials. There is some uncertainty as to the exact years to which they apply. The list we date to 812 was compiled by Kodama, who was a high civil servant, and derives from the earliest return in the archives dating from immediately after the fire of 819. The list of 846 is from Ibn Khordadbeh’s book of post roads published in that year. Ibn Khordadbeh was postmaster general and rapporteur for the province of Gabal. The return of 918 was drawn up by the vizier Ali ibn Isa.^{Footnote 6} The returns may indicate target collections based on averages of earlier years (so-called ‘ibra) rather than amounts actually received (El-Sammaraie 1972, pp. 85, 140–6, 195). This makes conclusions drawn from them less exact. Nevertheless, they are the best basis we have for assessing the income of the Abbasid state, and they have been relied on by all historians of the period.

Nominal income was greatest in the eighth century, and so was real income. Our knowledge of wages and prices is extremely fragmentary for these years (Ashtor 1969; Beg 1972; Pamuk and Shatzmiller 2014). We have some idea of the wages of ordinary soldiers and craftsmen in Baghdad, and compensation was similar for the two (Ashtor 1969, pp. 64–72, Kennedy 2001, pp. 78–9). Higher status people had much higher incomes (Ashtor 1975, p. 154). By deflating the nominal income by the wage of a craftsman/soldier, we can get an idea of how many people the Caliph could employ (Table 2). In the case of Harun al-Rashid, the answer is over half a million on a full-time basis. This was perhaps 7% of the adult male labour force of his empire.^{Footnote 7} Harun could field 100,000 troops, build Baghdad, and have money to spare for poets, philosophers, and mathematicians. Later Caliphs were not so blessed.

Harun al-Rashid drew income from across the Muslim world, but the fiscal core of his realm was the old Sassanian Empire. Most of the income came from the Sawad (southern Iraq) and western and central Persia. They presented different administrative problems because of differences in geography. Irrigation was critical across the empire, but in Persia, it was generally supplied by qanats. These were privately constructed, and the state played no direct role in their operation (Bulliet 2009, pp. 1–68, Lambton 1989). In contrast, southern Iraq was irrigated through a large-scale system constructed and maintained by the state. Five principal transverse canals ran from the Euphrates to the Tigris—the Isa, Sarsar, Malik, Kutha, and Grand Sarat—the Nahr Nars ran into the Great Swamp, and the Nahrawan ran parallel to the Tigris on its eastern shore (Figs. 1, 2, 3). These canals, as well as many smaller ones, brought water to otherwise arid districts. The water was distributed from these giant canals to smaller canals leading to the fields. The state built the major canals, kept them clear and dredged, rebuilt the system when it was damaged by floods, and kept it in good repair. This work was managed by an irrigation ministry, the diwan al-Kharaj. It employed engineers, land surveyors, supervisors of water levels, construction labourers, reed assemblers, water carriers, and controllers to supervise the works. The diwan also sent delegations to villages to mediate disputes about water resources, and the Caliph participated in these adjudications (Cahen 1949–51, El-Samarraie 1972, pp. 105–9, 173–80).

Settlement and Irrigation, 883–1258. Note: Compare this map to Map 2 and observe that settlement dropped sharply in the area between the Grand Sarat canal and the Great Swamp, which was the stronghold of the Zanj rebels. The map inset in the upper right shows the location of al-Mukhtarah, the Zanj capital. It lays south-east of the region mapped here. Settlement decline was much less extreme east of the Tigris and further from al-Mukhtarah

Full size image

These activities were financed out of tax receipts, and ‘user charges’ were also imposed on landowners when private interests were served. The Sassanians imposed fixed taxes per unit of land on the main crops. This practice was continued by the Arabs after the conquest, but the rates were substantially increased and the range of crops taxed was extended (Morony 1982, pp. 99–106, 1984; Campopiano 2012). In 786, this system was replaced by one designed by the vizier Mu’awia ibn Yasar, and his system continued with some modification throughout the Abbasid period. Three main taxes fell on the agricultural population (El-Samarraie 1972, pp. 146–155). The kharaj was imposed on land held by non-Muslims at the time of the conquest and included much of Iraq. Under Harun al-Rashid the state took 50% of the production of the main field crops. Fruit trees were taxed as well. The rate was cut to 40% in 819 under Caliph Ma’mun and remained at that rate into the tenth century. The jizya was a poll tax imposed on non-Muslims. Its collection was combined with that of the kharaj and was of declining importance as the population converted to Islam. Finally, ‘ushr (tithe) was imposed on the private property of Muslims at rates varying from 10 to 25%. This tax was of considerable importance in south-eastern Iraq. In 628–9, the Tigris burst its banks and shifted its course. Much of the region was flooded and rendered uncultivable (Verkinderen 2015, pp. 50–60). Under Sharia law, anyone who brought wasteland into production acquired title to it, and that protection induced rich Muslims to reclaim land on which they were lightly taxed. The state tried to breakdown the distinction between kharaj and ‘ushr land both by legalistic changes to definitions that reclassified land as kharaj and by raising the rate on ‘ushr land. By the mid-ninth century, the latter had reached 50%. However, land grants to Turkish generals worked in the opposite direction, for they were only taxed at 10% (El-Samarraie 1972, p. 153).^{Footnote 8}

The Abbasid Caliphate crumbled during the ninth century. The first shock was the civil war between Harun al-Rashid’s sons Amin and Ma’mun. Harun devised a complicated formula to regulate the succession, but it did not resolve differences between them. Amin was in Baghdad and Ma’mun in Khorasan, where he raised an army of 50 thousand to attack his brother. Baghdad was besieged for a year, Amin was killed, and Ma’mun became Caliph. During the siege, the irrigation and agriculture in Anbar province were wrecked, and that is one reason that income derived from the Sawad was lower in 846 than it had been in 812 before the war.

The Caliphate recovered after the civil war. Ma’mun proved to be a great patron of philosophy and the movement to translate Greek texts into Arabic. Members of the court also sponsored science and scholarship. Ma’mun’s brother, Mu’tasim (833–842), succeeded him and continued to promote science and philosophy, although his attentions were directed to the military.

Mu’tasim decided he needed forces who were personally loyal to him, and he bought Turkish slaves as soldiers. In 835, he began to construct a new capital at Samarra to house them away from the rest of society. The Turkish soldiers numbered only about 5000, but their generals were leading figures in Ma’tasim’s regime. Whether or not they were slaves, the Turkish soldiers were far from servile, and, in the long run, they proved to be a fatal mistake.

Mu’tasim was succeeded by Wathiq (842–847) and then by Mutawakkil (847–61). He was no friend of philosophy, reversing many of the enlightened policies of the Golden Age. He too spent fortunes erecting new palaces across the Tigris from Samarra. He sought to check the power of the Turkish generals by transferring administrative responsibilities to his sons through a complicated succession plan. It failed even more spectacularly than Harun al-Rashid’s. His sons soon conspired against each other, and the Turks were keen to make alliances to preserve their influence. In 861, Mutawakkil was murdered probably at the instigation of his son Muntasir (861–2), who was soon established as Caliph. At the behest of the Turkish military, he removed his brothers from the succession that his father had established. Muntasir died mysteriously the next year, and the Turkish generals decided together to appoint his grandson Musta’in as Caliph. He was completely dependent on the Turks and even appointed a Turk, Utamish, as vizier. Utamish did not pay the troops, and then looted the treasury, after which he was murdered. The troops mutinied and in 863 the Caliph left Samarra for Baghdad where he broke with the Turks. The Turkish generals decided they needed a more compliant Caliph and appointed Musta’in’s brother Mu’tazz to the post. The result was a war in which Mus’tazz and the Turkish army besieged Baghdad. Once again the agriculture of the region was devastated. By 866 Mu’tazz had won, and his brother was banished and later murdered.

Mu’tazz did not last long. The financial crisis continued, and the troops were not paid. They mutinied again and again the Caliph was murdered. He was succeeded by Muhtadi who was also murdered in 870.

The disastrous nine years 861–870 came to an end through a kind of joint rule between two brothers. Mu’tamid was caliph (870–892), and his brother Abu Ahmad al-Muwaffaq was governor of Iraq and Arabia. Al-Muwaffaq had been a commander in Samarra and had close relations with the Turkish military, which earned him their confidence. He could broker a peace with the Turks by guaranteeing their pay and their standing in the state. In 892, his son became Caliph Mu’tadid (892–902), the last effective Caliph.

The years 861–870 witnessed the financial collapse of the Caliphate. There was some decline in revenue between Kodama’s figure which we date to 812 AD and Ibn Khordadbeh’s, which we assign to 846–7, the date of the first edition of his book.^{Footnote 9} The revenue figures for 892 show the situation as it developed in the 860 s: income fell from 280 million dirhams to 38 million, and those came only from the Sawad. Even these figures may have been too optimistic. With the inflated wages of 892, the Caliph could afford less than 15 thousand craftsmen and soldiers. This was a huge decline in state capacity. Salaries had not inflated in the 860 s (Kennedy 2001, p. 131), so the Caliph could hire about 70 thousand employees with his 38 million dirhams. Armies of 100,000 men were no longer an option. The largest army of the period was the one besieging Baghdad in 865, and it numbered 19,000. Indeed, military engagements in the 860 s normally involved only a few thousand on each side (Kennedy 2001, pp. 126–7).

The causes of the collapse were twofold. First, as civil war broke out, troops were called to Baghdad from across the Caliphate. Regional leaders asserted their independence. The Caliphate had not provided public goods that made membership valuable—the bonds of unity were solely the ideological claim that the rulers were legitimate since they were descendants of the Prophet’s family—so taxes were sent to Baghdad for no good reason. Very quickly in the 860 s, all outlying provinces declared their independence or were taken over by local warlords or rulers who stopped remitting funds to Baghdad. This shift reduced the capacity of the Caliph in Baghdad, which is our main concerned, while transferring state capacity to the newly independent polities. Second, a confounding factor was the large size of the military, which was said to cost 200 million dirhams per year. This was most of the budget in 846, and that is the reason that troops could not be paid once income started to fall. It was safer for the Caliphs to stopping paying troops that were far from Baghdad rather than ones that were near, but unpaid troops in the provinces became the armies of separatist rulers who continued to collect taxes but spent them locally. For instance, the inhabitants of Rayy in Persia offered to pay the commander of troops recalled to Samarra more than the troops would receive in Samarra on the condition that they would stay in Rayy and guard the city. Kennedy (2001, p. 130) suggests that deals like this were common once the Baghdad government ceased to function in the 860 s.

The result was that fiscal pressure was increased on the farmers of the Sawad. First, annual taxes were collected more than once per year. Second, tax farming became general as the need for funds became desperate. Contracts normally lasted only one lunar year, and tax farmers frequently extorted more than the taxes due. The state failed to provide justice to the peasants (El-Samarraie 1972, pp. 171–3). Third, in the 860 s when there was no money to pay troops, the Caliph allowed the army to extort pay directly from farmers in the Sawad. This greatly oppressed the peasantry, and the irrigation system went into decline as there were no resources to maintain it. Finally, state officials and military leaders were compensated by granting them landed estates as iqtas. The grantees were allowed to tax the peasants and keep the proceeds. The iqtas also carried administrative responsibility for the area. As the grants were short term, however, the holders had no incentive to maintain the irrigation system.^{Footnote 10}

Revolution in the countryside was a consequence of the exploitation of the peasantry in the 860 s. The centre of rebellion was the Revolt of the Zanj (869–883). The Zanj were African slaves brought to Iraq to clear the land that had been flooded by the Tigris in 628–9. They were harshly treated. Ali bin Muhammad was their leader. He had previously tried his hand as a court poet in Samarra, as a revolutionary in Bahrain, and then as a prophet. When he discovered the slaves in southern Iraq, he recognized them as potential followers and proclaimed revolution in 869, preaching the egalitarian doctrines of the Kharijites. They established themselves in the Great Swamp in southern Iraq where they built a capital et al.-Mukhtarah. Support for the rebellion was widespread among the peasantry in the irrigated plain (El-Samarraie 1972, pp. 117–8), but the marshes remained the rebellion’s military strong point. The rebels successfully defeated the Caliph’s troops for many years, but their capital was finally destroyed in 883. Rebellion lingered on in widespread support for the Carmathians (Popovic 1975).

The Caliphate enjoyed a brief revival in the late ninth century under al-Mutamid (870–892), al-Mu’tadid (892–902), and al-Muktafi (902–908). They never restored the prosperity of the Sawad, but they managed to restore suzerainty over western and central Persia, northern Mesopotamia, and even Syria and Egypt. This led to an increase in revenue but a smaller increase in state capacity since wages rose sharply in the late ninth century. The budget of 918 indicates the Caliph could employ 70 thousand soldiers or craftsmen. This was far below the half million at Harun al-Rashid’s command, and it did not last long either. The Sawad supplied only one-fifth of the revenue. A state mainly dependent on outlying regions for its funding could not contest regional separatism. The ideology of Abbasid legitimacy was long discredited, and there were not enough troops to hold the empire together by force, so Caliphate dissolved as regional war lords and local governments ignored the Caliph’s commands.

A narrative history highlights important elements of the story, but it does not elaborate all of the interrelationships among them nor establish their quantitative importance. To clarify the linkages between political structures and actions and their implications for economic activity, we need a model of Iraq. Our model has four objectives:

The model is based on the idea that Iraq was a hydraulic state. The theory of the hydraulic state was first glimpsed in 1853 by Engels and Marx (Husain 2006, pp. 13, 263) and was catapulted into prominence by Wittfogel (1957). It has since been widely discussed and debated.^{Footnote 11} There is little consensus as to the theory’s content. Some scholars think that it entails that every irrigation system should bring forth a state (and then show that that is false); others think that irrigation must precede that state (and then show that states often preceded irrigation).^{Footnote 12} Our first objective is to define what we mean by the theory and show that it is consistent with standard economics. In our view, the construction of irrigation requires some existing administrative capacity. If that is a small state or proto-state and there are possibilities for expanding the irrigation system, then the state can invest in irrigation, increase the farmed area, and increase state power by taxing the enlarged area. The chance of a large state emerging is greatest when the irrigation system is large and presents problems of indivisibility and coordination. These preclude the piecemeal extension of small-scale irrigation by local communities. We have argued that the conditions for a large, centralized state emerged in Mesopotamia once the Tigris and Euphrates rivers had separated from each other and reached their present courses (Heldring et al. 2020). The Sassanian empire united the region and expanded the irrigation system by digging the great transverse canals. This is the system that the Arabs maintained—or not. The model we develop here elaborates this view.

Scholars disagree widely over important questions like the size of the Mesopotamian population. McEvedy and Jones’ (1977, pp 151) put it at 2.5 million in 800, Issawi (1981) raised the total to 5–6 million, while Russell (1958, p. 89) estimated it at 9.1 million. In view of this span, Goldsmith (1987, p. 61) settled on a ‘synthetic estimate’ of 4 million. All of these estimates rest on very weak foundations. Our model is a coherent framework that links agricultural productivity, food supply, taxation, wages, prices, and population. By requiring consistency among variables, our model pins down the population by showing how many people could have been fed by the agricultural base. Values of other variables like the income of the aristocracy are also implied by the model. Working out these implications is the second objective of the model.

Our third objective follows from the second—namely, to work out the implications of the collapse in irrigation during the ninth century for the size of the population, the agricultural surplus, and the division of people between farming and other activities. By requiring that estimates be consistent with the model, we require them to be consistent with each other and that leads to better founded conclusions than those of other investigators.

The fourth objective is to provide a coherent framework to nest the explanations for the collapse that we test empirically. The empirical application of the model allows us to compare the effects of political conflict at the top of society, peasant revolution at the bottom, and saline accumulation in the agricultural base.

Figure 4 shows the short-run equilibrium of our version of a hydraulic state. The model allocates Iraq’s workforce (population) AB between agriculture AL and the rest of the economy LB. The agricultural sector is assumed to consist of identical peasant farms in which one unit of labour cultivates one unit of land and produces one unit of output, which we will call grain. Since there are fixed proportions and the farms are identical, EA is the average and marginal product of labour. The size of the farm population AL is determined by the area of cultivatable land, which depends on the capacity of the irrigation system.

The hydraulic state: short-run equilibrium

Full size image

Labour income in agriculture, the agricultural wage w, is determined outside of the model. In the long run, it is assumed to be determined by Malthusian population dynamics, so it is the subsistence income that equates births to deaths. This system is assumed to have obtained generally in the Middle East including the herding population in the waste land adjacent to Mesopotamia. People are assumed to have moved into or out of southern Iraq as the agricultural income in the district deviated from the equilibrium Malthusian wage in the Middle East generally.

Population movements between herding on the steppe and farming near the rivers are widely observed in modern times. In the north Arabian dessert, sheep herders often planted millet and returned later to harvest it, so they had some knowledge of crops and sympathy for agriculture. Some tribesmen ‘cultivated and irrigated patches of land by the [Euphrates] river and spent the summer there in insubstantial villages of tents, huts and shelters made of branches and reeds. Most of them went in winter, with their animals, to the steppe north or south of the river’. More generally, ‘the way of life of some of the tribesmen also changed several times, from nomadism to tentative and temporary settlement, back to nomadism and then again to sedentarization’ (Lewis 1987, pp. 15, 70) Wilkinson (1994, pp. 190–1) describes analogous shifts in Oman, and we assume they were also common in ancient times.

The rectangle EDCw is the agricultural surplus, the difference between total farm production EDLA and the labour income of the farm population wCLA. We assume that the state taxes the entire agricultural surplus.^{Footnote 13}

The Caliph himself eats a minute fraction of the surplus: the bulk of it is used to build palaces and mosques, pay soldiers, hire craftsmen to make the luxuries of urban life, and to support poets and philosophers for amusement and enlightenment. The demand for labour outside of agriculture is the rectangular hyperbole rising from F asymptotically towards the vertical line BG: tax revenue = LB·w* where LB is employment outside agriculture and w* is the wage outside agriculture. In the simplest case, w* would equal w, but in reality, non-agricultural wages were higher, and Fig. 4 is so constructed.

The model highlights how the agricultural and urban sectors are inter-related. First, if the productivity of agriculture in the sense of output per worker EA increases, then there is no change in the agricultural population, but the urban population expands since all of the extra income accrues to the Caliph as tax revenue (Watson 1974, 2008). Conversely, rising salt levels in the soil lower agricultural productivity leading to a smaller agricultural surplus and a smaller urban population. Second, if the irrigation system is expanded, so that the agricultural population AL increases, then tax receipts rise, and the urban population also expands. Conversely, if the irrigation system contracts, then both the rural and urban populations fall. This is what happened in Iraq in the 860 s and later.

We can calculate the size of the population and the economy before and after the decline in the irrigation system, if we calibrate the model. Calibration means putting numerical values on all of the parameters in Fig. 4. We add a few extra details to make the model more realistic. We begin with life in the Golden Age when population and output were at their greatest. Note that to generate predictions, we will have to make assumptions that might not survive close scrutiny if detailed historical data were available. In the absence of such data, the merit of making these assumptions is that we get clear predictions about changes in surplus and population when parameters change, we can bring more information to bear on the problem than would otherwise be the case, and we can test the sensitivity of our results to alterations in our assumptions. We begin by describing the parameters and the values of the exogenous variables c. 800. We show the implications of those values for key endogenous variables at that time. Then, we consider what had changed c. 900.

4.1 Subsistence income in agriculture

Poor people consume some legumes, vegetables, cloth, and fuel, but grain is generally the largest share of their spending (Allen 2017, 2020). Our model has only one good—grain—so we set income at 250 kg per person per year, bearing in mind that in reality some of this supported the production of the additional goods that were also consumed by the poor. 250 kg is implied, for instance, but assuming that food amounted to 80% of a subsistence income, that average consumption across the population was 2000 cal per day, and that grain yields 3600 cal per kg.^{Footnote 14} In Fig. 4, subsistence income in agriculture, w = 250 kg per year.

4.2 Irrigated acreage

In the hydraulic state, the agricultural population is determined by the irrigated acreage. In the early twentieth century, there was no effective, modern irrigation system in Iraq. Proposals to remedy the situation were developed by Sir William Willcocks. Willcocks was a leading Victorian engineer, who had previously worked in India and in Egypt where he surveyed the Nile (Willcocks 1899) and built the low dam at Aswan as well as other important irrigation works. Willcocks’ (1917) proposals for Iraq formed the basis for that country’s irrigation policy in the twentieth century. Willcocks and his team were the first to measure the flow of the Euphrates and Tigris rivers and calculate their potential.^{Footnote 15} The following is his conclusion:

Assuming that the duty of water in Mesopotamia would be the same as that of the hottest and driest parts of the Nile valley near Luxor, we may say that one cubic metre per second of water will, in winter, irrigate 3,000 hectares, and in summer 400 hectares of rice, or 1,250 hectares of dry crops, such as millets, sesame, etc. Using the discharge tables of the two rivers given above, we may say that the water supply will permit of 3,000,000 hectares of winter crops, while the summer discharge will permit of 400,000 hectares of rice or 1,250,000 hectares of millet, sesame, cotton, etc. (Willcocks 1917, p. 9)

The logic behind these calculations is not entirely clear,^{Footnote 16} but similar results can be obtained by applying modern rules of thumb to Willcocks’ discharge tables. The rules of thumb are (1) that a cubic metre of water is required per square metre of land by the plants for their growth under middle eastern conditions, (2) that losses from evaporation and leakage mean that 2.5 cubic metres of water must be taken from the river to supply the plants with one cubic metre per hectare, and, in addition, (3) that 35% of the water withdrawn from the river finds its way back into the river (Kolars and Mitchell 1991, pp. 124–42, especially 134–5). The upshot is that a net withdrawal of 1.625 cubic metres is required for each square metre cultivated. 3 million hectares of winter cultivation would, therefore, require 48.75 billion cubic metres of water or an average flow of 3100 cubic metres per second over the winter half of the year. This is easily accommodated by the monthly flows recorded by Willcocks (1917, p. 9) and his team: the period December–May averaged 3050 cubic metres per second and the rate increased to over 3650 cubic metres per second if the start date of the six month period is pushed back to February. The exact timing of water requirements is not known, but the land was watered in the summer before planting and again as the crops approached maturity in the spring.

The calculation assumes that the full flow (at least on a seasonal basis)^{Footnote 17} of the Tigris and Euphrates was being used for productive purposes. This assumption is consistent with the calculations of Adams (1981, pp. 180, 210) using archaeological survey data. Adams traversed much of plain between the Tigris and Euphrates collecting pottery shards, which he dated based on ceramic style. He was thus able to reconstruct the settlement pattern of the region from the beginnings of settlement c. 4000 BC through the Islamic period. The region achieved its most complete development at the end of the Sassanian period, i.e. when the Arabs took over. His settlement map shows ‘an exceptional correspondence between the limits of this simulated zone and the limits of the surveyed area itself, indicating that virtually continuous cultivation within large parts of the latter would have been necessary’. Furthermore, ‘with a whole series of massive diversions upstream, it is not unlikely that in Sassanian times the Euphrates entered the swamps [at its lower end]…with very little if any residual flow’. All the water was being used for agriculture. The same was true of the Tigris as well, whose waters not only irrigated the east bank but were drawn west onto the central plain. Adams also analysed the tax data for the Sawad of Kufa, (the region around Kufa) and concluded that the tax receipts implied an area under cultivation that ‘was so slightly above the maximum limits of the Sawad of Kufa as calculated from Euphrates drainage patterns as to lie within an acceptable margin of error’. These arguments make it reasonable to assume that Willcocks’ estimates of the land that could be cultivated were actually in tillage.

Since the winter acreage is much greater than the summer acreage, we assume that the size of agriculture was determined by availability of irrigation for winter cultivation. When the system was at its peak in the days of Harun al-Rushid and Ma’mun, we assume that 3 million hectares of winter grain were cultivated. Since land was fallowed every other year, agriculture was practised over twice this area. Fixing the winter acreage at 3 million hectares pins down the size of the population and the economy before the collapse.

4.3 Agricultural population

How much labour was required to cultivate 3 million hectares? Peak labour demand occurred at the harvest. Experimental archaeologists have discovered that it took 20 man-days of labour to harvest a hectare of wheat using sickles with stone or early metal blades (Korobkova 1981; Russell 1988, p. 116–7). The harvest lasted two months, so one man could reap three hectares (Wilkinson 1994, p. 496). Harvesting 3 million hectares, therefore, required one million men.^{Footnote 18} Assuming four people per family implies an agricultural population of four million. This is line AL in Fig. 4.

4.4 Income of the agricultural population (peasant income)

The four million people in the countryside were each assumed to consume 250 kg of grain per year, so their total income was the product of the two numbers. This equals area wCLA in Fig. 4.

4.5 Gross production of field crops

We have considerable information on yields, seed, and animal feed for farming in southern Mesopotamia in earlier periods, and we use that on the assumption that conditions were similar. Wheat and barley were the main field crops, and we treat all crops as though they were wheat. There is debate about the yield of wheat (Potts 1997, pp. 80–2). We settle on a yield of 900 kg per hectare. We apply this to all 4.625 million hectares that Willcocks believed could be planted in summer as well as winter. Total production of field crops was 4.1625 million tons.

4.6 Agricultural costs: seed

Potts (1997, p. 82) following Halstead (1990, p. 187) put the seeding rate at 32 kg per hectare, which implies a much higher yield–seed ratio than is normally achieved when seed is broadcast. Cuneiform texts frequently mention a rate of 240 sila per bur or 37 L (28 kg) per hectare (Powell 1990). Halstead attributed the high yield ratio to the use of a drill plough. We set the seed rate at 32 kg per hectare.

4.7 Agricultural costs: feed for oxen

There is no natural pasture in southern Iraq, so all oxen had to be fed grain. Using information from ancient texts on the distance between furrows, the speed of plough teams, and the number of times fields were ploughed, Potts (1997, p. 85) deduced that a square field of 1 bur (6.48 hectares) would have required 24.98 days of labour. If the plough team consisted of the normal two oxen and if each ox ate 6 L of grain per day, then the oxen in the plough team would have eaten 35 kg of grain per hectare. This result is consistent with Babylonian cost accounting in which it was reckoned that seed per hectare and fodder per hectare were equal (Potts 1997, p. 83, Powell 1984, Hruška, 1984, p. 154).

4.8 Agricultural costs: grain to meat

Meat was consumed in Iraq, particularly by the upper classes. Much of Baghdad’s meat must have come from animals grazing on the steppe of northern Iraq, but meat was also raised in the Sawad. We assumed that 20% of the gross output of grain was fed to animals. Cattle require approximately six kilograms of feed for each kilogram of meat they yield (Shike 2013),^{Footnote 19} so meat production in southern Iraq was one-sixth of 20% of gross grain production.

4.9 Agricultural costs: spoilage and losses

10% of gross output was assumed to have been eaten by vermin or otherwise lost or spoiled. This is a typical value for a poor country according to the UN Food and Agricultural Organization’s food balance sheets. Ratios in recent years ranged from 6% in India to 12% in Iraq to 15% in Afghanistan.^{Footnote 20}

4.10 Net agricultural output

Net output was computed as gross output of grain plus the production of meat minus grain used for seed, feeding to oxen, feeding to cattle for meat, and losses. This equals area EDLA in Fig. 4.

4.11 Total agricultural tax revenue

Total tax collections around 820 were 109 million dirhams (Table 1). This sum represents peak tax collections from the Sawad. Most of the taxes were in kind and converted to cash at the rate of 450 dirhams per korr of 2925 kg of grain (von Kremer 1875, pp. 362–3). Some tax was also taken in money. The total is expressed in kilograms of grain by dividing the total by the price.

4.12 Total aristocratic rental income

There was privately owned land in the Sawad that was leased and which supported an aristocracy. Indeed, the sum of tax revenue and peasant income is less than net agricultural output, and we take the difference to have been aristocratic income. The sum of tax revenues and aristocratic income equals area EDCW in Fig. 4, and that total gives rise to the non-agricultural labour demand curve FH. (Tax and aristocratic income are not distinguished in Fig. 4 since they affected labour allocation in the same way.)

We noted earlier that the typical tax rate on Kharaj land was 40% of production. Whether this was gross or net is not clear from the sources. Other land was taxed at a lower rate. Our calculation that taxes amounted to 28% of net production looks in line with these figures.

4.13 Average urban wage

Urban wages were much higher than rural incomes. The average agricultural family of 4 is assigned a subsistence income of 1000 kg of grain. This would have cost 154 dirhams per year—about 13 dirhams per month—if valued at the same price as tax collections. In the early ninth century, the average wage of a soldier was 60 dirhams per month, and that agrees with scattered information about the wages of craftsmen in Baghdad (Ashtor 1969, pp. 64–5, 70–1, Kennedy 2001, pp. 78, 80–1). Ashtor (1976, p. 154) shows a broad range of earnings in the urban economy but the highest earnings only accrued to a small number of high ranking officials. We assume the average urban wage was four times the rural wage. This is a substantial premium, but, of course, the urban population was mainly Arab, while the rural population was the descendants of the peasants under the Persians.

4.14 Urban population

Under the maintained assumptions, the non-agricultural population of Iraq was 1.8 million. It might, in reality, have been larger in view of the fiscal transfer to Baghdad from other parts of the Caliphate. Russell (1958, p. 89) conjectured that Baghdad had 300,000 residents. Other cities like Kufa, Basra, and Wasit were important, and the non-agricultural population included soldiers and people in smaller settlements, too.

4.15 Total population

The total population is the sum of the rural and urban components—5.8 million.

We have only fragmentary information about different economic, social, and geographical features of Iraq in the early Islamic period. Our model can be used to combine these into a macro-picture. While one can dispute the parameters, there is at least a consistent logic that combines them into a whole. Table 3 shows some key results for the Golden Age around 812 AD.

The top panel shows total (net) agricultural production measured in kilograms of grain and its division among the principal social classes. On the assumptions made here, peasant income was 36% of the total, taxes were 26%, and rents were 38%. Irrigated farming in Mesopotamia was so remarkably productive that almost two-thirds of net production could be taken from the farmers to support the Caliph, the upper classes, and, indeed, the urban population in general. Iraq was fertile enough to support a Golden Age.

The second panel shows the implications for the size and distribution of the population. Slightly more than two-thirds of the population were in agriculture, and the remainder were in the cities or in military installations, etc. The total population implied by the assumptions made here amounted to 5.8 million. This is in the middle of existing estimates that range from 2.5 to 9.1 million.

The final panel of the table shows output per worker in agriculture, income per worker in agriculture, and income per worker outside of agriculture. Income per worker in agriculture of 1000 kg per year corresponds to a family of 4 receiving 250 kg each on average—the subsistence income. The difference between output per worker and income per worker in agriculture is the agricultural surplus, which is the income received by the Caliph and the aristocracy and which supports the non-agricultural economy. Income per worker outside of agricultural is set at 4000 kg per year—four times the farm income. Had the incomes been the same, then the ratio of agricultural surplus to agricultural output would have equalled the ratio of the non-agricultural population to the total, i.e. 63%. In reality, it was much less because urban incomes were much higher.

We can use our model to simulate the effects of the economic collapse in the late ninth century. The best indicator is the decline in tax revenue from the Sawad, and numerous historians have pointed to it as a measure of decline (e.g. Ashtor 1976; Campopiano 2012; Christensen 1993; Waines 1977). Between about 846 and 918 tax receipts from the Sawad dropped by 48% (Table 3). The usual interpretation is that the 48% drop in tax revenue indicates a corresponding 48% drop in production on the assumption that the state took a constant share of the harvest. This inference, however, is off the mark for two reasons. First, agricultural prices increased several folds in the late ninth century, and that means that the decline in real tax revenue was much greater than the decline in nominal revenue. Pamuk and Shatzmiller (2014, p. 202) found that the price of wheat rose by a factor of 3.55 from the mid-9th to the mid-tenth centuries, in which case, the volume of the harvest in 918 was only 14% of what it had been in 846. Second, however, there was an offsetting factor that must also be taken into account, namely, the administration of the tax system shifted from direct collection by state officials to taxing farming. In 893, for instance, the entire Sawad was granted to Ahmad b. Muhammad at-Tai as a tax farm, and the 38.3 million dirhams in taxes shown in Table 1 was the sum that Ahmad remitted to the Caliph (Busse 1967). He would have collected more from the peasantry, and difference represents his costs and profit. It is the value collected from the peasants that must be compared to earlier values to calculate the decline in output, so the amount remitted to the state must be increased accordingly. It was the same situation in 918. We have no information about the premium collected by tax farmers.^{Footnote 21} If we presume that they increased tax collections by one-third, then real tax collections and the volume of production dropped to 18% = (0.48 /3.55) * 4/3 of the 846 level. We will work with this 82% decline.

In the simulation of the economy of Iraq in the early ninth century, the assumption was that its size was constrained by the discharge of the Tigris and the Euphrates rivers. In the late ninth century, the economy shrank. In the next section, we argue that the immediate cause was the collapse of state maintenance of the irrigation system. For the moment, we consider its effects. We reduce land, labour, and output by 82% since fixed proportions obtained among them. Table 3 shows some macro-implications of the reduction. The population drops by 82% from 5.8 million to just over 1 million. The model implies that the agricultural and non-agricultural population drop by the same proportion reaching 0.7 million and 0.3 million, respectively. This is extreme depopulation.^{Footnote 22} We cannot say how many people died and how many fled to other parts of the Middle East. The population drop is the reason that the archaeological survey data show the central part of the Sawad to have been empty in the tenth century. This can be seen by comparing Figs. 2 and 3, which show the settlements in the two portions of Iraq that Adams (1965, 1981) ‘intensively surveyed’. Settlement fell dramatically between the ninth and tenth centuries in the southern Mesopotamian plain near the Zanj stronghold in the Great Swamp and much less so in the region east of Baghdad, which was harder for the Zanj to reach. Indeed, the southern Mesopotamian plain is still unpopulated today (Fig. 1).^{Footnote 23}

Why did Iraq collapse in the late ninth century? The key to prosperity in the Sawad was irrigation, and it depended on the actions of the aristocracy and, especially, the Caliph. He could spend his tax receipts on consumption or investment. The palace, the harem, and the army were all consumption expenditures. The irrigation system was the critical investment. The system deteriorated rapidly due to silting and flooding, so a minimum expenditure was required simply to keep it going. More expenditure led to expansion. Private individuals also invested in land reclamation, in particular, in south-eastern Iraq when reclaiming land wrecked by the floods of 628–9. Our emphasis, however, is on the state, for it was responsible for the major transverse canals in the Euphrates plain as well as the Nahrawan east of the Tigris. The private land reclamation on the lower Tigris was concerned with drainage rather than irrigation.

Figure 5 analyses the investment decision of the Caliph on the theoretical plane. C_t is consumption in some year t, and C_t+1 is consumption in the following year. It is important to appreciate that ‘consumption’ here means allocating labour, for the way in which the Caliph consumes the surplus is by using it to hire soldiers, poets, or, in this case, irrigation construction workers. Consuming less this year means employing more irrigation workers, which allows more consumption in future years. The straight line MQLN shows the rate of return to allocating labour to canal work. If no labour is assigned to the canals, the maximum consumption N is possible, and, in that case, consumption in the following year will be I. If JN labour is instead allocated to irrigation work, then current consumption is reduced to J but income the following year is increased to K. A horizontal line has been drawn at K for a reason. K = C_t. In other words, JN represents the labour that must be allocated to irrigation maintenance to keep the system in good repair. If that is done, then consumption remains the same from year to year.

The hydraulic state: long run equilibrium

Full size image

How much labour is allocated to canal work depends on how future or present oriented the Caliph is. The Caliph’s time preference is represented by the curved indifference curve PQR. There is a family of these curves, and the one shown is the highest that touches the rate of return line MQLN. As drawn, the Caliph would choose to consume only S in the present year. SN labour would be allocated to the irrigation system, and that would allow expansion as well as maintenance, and consumption would rise to T in the following year.

It is an important feature of the model that the rate of time preference could vary. If PQR, for instance, shifted down and to the right, then Q could fall in the segment LN. In that case, not enough would be spent on the canals to maintain them, and the system would deteriorate leading to lower future consumption. A shift like this would indicate a more present-oriented outlook.

The model highlights two types of shifts that could have influenced investment in the canals. One is changes in the time preference of the Caliph, as just noted. The second is changes in the rate of return to irrigation investment. A higher rate of return is indicated by sliding M up the axis to a higher value, while leaving N as it is. Sliding M down indicates a lower rate of return, for consumption tomorrow increases by a smaller amount for any given investment. Whatever the position of PQR (the rate of time preference) a higher rate of return induces a higher level of investment since the PQR indifference curves get steeper in moving left, so tangency will occur further left when the slope of MQLN increases.^{Footnote 24}

In summary, our model encompasses all of the important explanations of the collapse of civilization in Iraq in the ninth century. The salinization hypothesis implies that agricultural productivity (line ED in Fig. 4) would drop over the course of the century and with it tax revenues (EDCw). Civil war between rival rulers implies an increase in the rate of discounting the future (indifference curve PQR shifts to the right in Fig. 5 so the tangency is to the right of J), which would lead to inadequate investment in canal maintenance and declines in the irrigated area (L moves to the left in Fig. 1). Tax revenue would fall. Finally, the revolt of the peasants would lower the profitability of canal maintenance (MQLN in Fig. 5 is less steep) since the cost of maintenance would rise in contested areas and peasant non-compliance would mean lower tax collections. In the next section, we measure these effects.

The starting point of our econometric analysis is a data set consisting of 27 tax districts in southern Mesopotamia.^{Footnote 25} We focus on the districts serviced by the transverse canals and the Nahrawan east of the Tigris. In these districts, canal maintenance was critical due to silting up of the canals and the constant maintenance required for canal walls, weirs and other infrastructure. We exclude the reclamation districts along the lower Tigris where drainage rather than irrigation was the main issue. For each district, we collect data at three points in time, 812, 846 and 918CE. Because the collapse took place between 846 and 918, we measure each district before and after the collapse. All of the hypothesized explanations for collapse have implications for tax revenues, and we study those relationships to establish the importance of the hypotheses. We compare districts before and after the collapse, and study whether measures of our competing explanations mitigate (or potentially) exacerbate the collapse. We do so in a cross-sectional set-up, as well as in a difference-in-differences framework.

8.1 Measurement

Our measurements of tax collections in each district derive ultimately from the same Abbasid officials whose work underpins Table 1. Tax revenues for the districts are available for 812, 846 and 918, in millions of dirhams. From these data we construct a cross-sectional outcome of interest, the ratio of revenue in 918 relative to its value in 846. For our difference-in-differences regressions, we construct a three-period panel, and use tax revenue as the outcome.

We measure maintenance of the irrigation system by water regulators. On the transverse and Nahrawan canals there were qantaras (weirs)^{Footnote 26} that regulated the flow of water to control its discharge to smaller canals leading into the fields. The tenth century writer Ibn Serapion (Suhrab) wrote a book about the river and canal system of Iraq and in it listed the working water regulators towards the end of the ninth century (El-Samarraie 1972, pp. 31–2). We have mapped these into tax districts according to the canals they serviced and, in some cases, the names of the qantaras. Adams (1981, p 216) noticed that the qantaras were not evenly spaced across the Sawad and remarked on the importance of this information but did nothing systematic with it.

The main hypothesis of this paper is that the collapse of the economy was ultimately caused by a failure in the functioning of the state. We operationalize this hypothesis by testing whether the presence of a qantara in a district was positively related to its tax yield in 918 relative to 846. We expect that the decline was smallest where the state maintained the irrigation system the most. Our model suggests that the presence or absence of a qantara depends on both the profitability of canal maintenance (itself dependent on the severity of peasant revolt) and on rate of time preference (reflecting war among contenders for Caliph). In addition, the salinization hypothesis would hold that salt build-up simply reached a tipping point. We have argued that the rate of time preference changed at the top of society, and for our empirics is captured by time-effects, which we discuss below. We measure salinization and revolt as follows.

We measure the salt build-up hypothesis by the composition of in-kind tax collection. As soil salinity increased, the yield of wheat was reduced. Barley was more saline resistant and did not suffer to the same extent. Consequently, farmers cultivated more barley and less wheat in response to growing salinization. Much of the tax was collected in kind, and the returns distinguish between the wheat and barley collected. On the assumption that the tax collections reflected what was grown, we use the ratio of wheat to barley as an indicator of the severity of salinization. There was wide variation in this measure.

We measure the severity of the revolt by the distance of the tax district from the Zanj capital, al-Makhtarah, near modern Basra. The idea here is that the distance from a tax district to the Zanj stronghold in the Great Swamp measures exposure to the rebels and their ideology. To ensure comparability of coefficients, we binarize the wheat/barley ratio and the distance to the Zanj capital by a median split. Results do not change meaningfully if we keep continuous variables.

8.2 Results

We start with a simple OLS model. We run a linear regression relating the ratio of tax revenue in 918 to 846 to the presence of a qantara, measured by our indicator, which is equal to one if a district had a weir. We report the result in Table 4, column 1. At the bottom of the table, we report the mean of the outcome variable, which shows that, on average, tax revenue fell by about 64% between 846 and 918, consistent with our narrative so far.^{Footnote 27} The coefficient in row 1 measures the difference in this collapse between districts with a qantara and districts without. We find that the collapse was markedly less pronounced in districts with a weir. Our estimated coefficient, 0.3 (s.e. = 0.1, significant at the 1% level) shows that the collapse was about half as severe in qantara districts as in the others. This result is consistent with the hypothesis of this paper: where the state failed to maintain infrastructure the collapse was significantly more severe than where the state maintained public infrastructure.

In column 2, we present estimates from a second regression in which we also include our indicator for a high wheat to barley ratio, evidence of relative low salinization, and our indicator for districts that are far away from the Zanj capital. Because each of our measures are indicators, the coefficients are comparable. Importantly, the rebellion changed the profitability of maintaining a weir. Even if the state had little to do with the maintenance of public infrastructure, we might have therefore observed weirs disappearing, simply due to rebel attacks or local combat. We find, however, that both alternative measures enter with small and insignificant coefficients whereas the coefficient on the qantara indicator is unchanged. This suggests that the decision to maintain the weirs was plausibly determined by overarching political considerations, such as the ruler’s times preference, and not by local actions pursuant to the rebellion. Furthermore, these results are inconsistent with salinization reaching a tipping point in the late ninth century.

Clearly, there may be time-invariant differences between districts that confound these conclusions. Districts may also have been on different trends that correlate with qantara presence. Since we have observations on tax collection before and after the collapse for districts that were differentially affected by the collapse, a natural next analysis is to look at the difference in tax returns in districts that were more affected by the collapse versus districts that were less affected. In other words, we also approach the problem of explaining the fall in tax revenues with a difference-in-differences framework. We have three successive cross sections of tax revenues for the 27 districts in 812, 846 and 918. This allows us to combine the three cross sections and investigate whether the change in political reality between 846 and 918 led to the collapse in state capacity, as indicated by the absence of functioning weirs in 918, which in turn led to a drop in tax revenues.

The dependent variable in these regressions is tax revenue in millions of dirhams in period t, (t = 812, 846 or 918) and the independent variables are a full set of 26 dummies for each district (we drop one dummy for multicollinearity reasons), and a dummy for the 812 cross section and one for the 918 cross section. This means that 846 is the omitted category and we measure the time-effects relative to the last period before the collapse, 846. By comparing to 812 to 846, we can make progress on the identifying assumption for difference-in-differences analyses, namely, parallel pre-trends. In column 3, we just include only these fixed effects. We do not report the district fixed effects for space reasons, and report the two time-coefficients in rows 4 and 5. The first coefficient, for 812, measures whether average tax revenue in 812 is different from 846. We find no meaningful differences. The second coefficient measures whether revenue in 918 is lower than in 846. As before, we confirm that there was a large collapse between 846 and 918. Relative to the mean tax raised across all periods and all districts of about 1.6 million dirhams, tax revenue falls by about 1.4 million dirhams in 918 relative to 846. This last effect is highly significant, which we assess through clustering our standard errors at the district level.

In column (4), we reintroduce our qantara dummy, as well as our measures of salinization and rebellion, interacted with period indicators. Each indicator that is interacted with the 812 indicator measures pre-trends because this coefficient recovers whether districts with, say, a qantara look different in 812 than districts with a qantara in 846 (the non-interacted indicators measure the same for the districts without a qantara). All coefficients for indicators interacted with 812 are insignificant, suggesting parallel pre-trends. We then turn to our variables of interest, the indicators for the presence of a qantara, low salinization and being far away from the Zanj, interacted with the indicator for 918, in rows 7, 9 and 11. We find that, as before, salinization and rebellion did not affect tax revenue, but the presence of a qantara does. We find a coefficient of 1.3 (clustered s.e. 0.3 and highly significant). Note that the coefficient on the 918 indicator is equal to -1.7. This means that the reduction in tax revenue in districts with a qantara was about a quarter of the reduction in districts without a qantara. As in our OLS analyses, we find that a functioning qantara largely prevented the drop in tax revenue.

The main advantage of our difference-in-differences approach is that the use of district indicators rules out any influence of time-invariant factors and the period indicators capture any trends that affected the whole of southern Mesopotamia. Under the parallel pre-trends assumption, we conclude that in districts where the state still maintained the weir through the collapse, tax revenues were significantly higher in 918 than in districts where they did not.

In sum, our empirical results highlight the importance of state capacity in explaining the decline in tax revenues in the ninth century: the presence of a well-maintained qantara underpinned agricultural, settlement, and tax payments. The results also throw some light on why qantara were maintained. We found evidence in favour of a high discount rate in the late ninth century. We have no direct measure of the falling return to irrigating, but in many districts, it was below the rate of discounting. Beyond that, rural insurrection played no role in lowering tax collection. The results strongly reject the hypothesis that saline build-up played a role in the collapse of southern Iraq. This result is not surprising since salt accumulated over millennia, while the collapse took only a few decades.

The Arab armies that overwhelmed the Sassanian and Byzantine empires in the seventh century conquered some very rich territories. The Sawad of southern Iraq was the keystone of their empire. Our calibrated model of a hydraulic society shows why: Even with Stone Age tools, Mesopotamian agriculture was so productive that the support of the farming population only required 36% of the net output of food produced. The rest was available to support great cities, extensive commerce, and a vibrant culture. This prosperity underpinned the efflorescent of culture in Baghdad.

But this success required irrigation and the geography of the Sawad meant that state support was vital. Unlike Persia where irrigation with qanats was small scale and could be organized locally by private entrepreneurs, the Mesopotamian plain required giant long-distance canals to realize its full potential. These canals required public investment for their construction and maintenance. When Caliphs were rich and far sighted, the system worked. When money became tight and immediate needs dominated, maintenance of the canal system was threatened.

The contested successions of the ninth century led to civil wars that undermined the system. Civil war meant that the contenders for the Caliphate became extremely present oriented and sacrificed the irrigation system for victory in battle. In addition, it led to the exploitation of the peasantry and rural revolt. The Revolt of the Zanj contributed to the collapsed of agriculture in Iraq by making investment in the irrigated plains of Iraq closest to the rebel held marshlands riskier than otherwise. Investment suffered and by the end of the ninth century most of the Sawad was desolate and abandoned, and it remains like that now. This economic catastrophe meant that it was impossible for the Golden Age of Islam to continue as it had under earlier caliphs and was the first conjuncture in the reversal of fortune between the Middle East and Europe.