Origins and Organization

While slave-raiding in Eastern Europe dates back to antiquity, it remained limited and sporadic until the late medieval period. A series of devastating wars, culminating in Mongol invasions and the establishment of the Ulus of Jochi in the thirteenth century, led to a pervasive state of insecurity and deprivation in which the abduction and sale of children became common (Roşu Reference Roşu2022, 9).Footnote ⁴ The supply of slaves dramatically expanded in the mid-fifteenth century with the disintegration of the Ulus of Jochi and the fall of Constantinople, which reoriented Black Sea commerce toward the “economically thriving”—and thus labor-hungry—Ottoman Empire (Witzenrath Reference Witzenrath2022, 29). The Crimean Khanate, a powerful Tatar successor state to the Ulus of Jochi, was acquainted with agriculture but found slave-raiding more profitable for two reasons. First, only coastal districts of the Crimean peninsula were suitable for intense cultivation, and their yields “were insufficient to support the multitudinous warring layers” (Ivanics Reference Ivanics, Fodor and David2007, 193). Second, Crimean Tatars retained traditional nomadic skills and military know-how that enabled them to conduct rapid and destructive raids across the steppe. Ottoman control of the Black Sea and, from 1475, Crimea itself created a vast international market for Christian slaves—Muslims were prohibited from enslaving coreligionists—who came to play a central role in the empire’s economy, military, and bureaucracy (Inalcik and Quataert Reference Inalcik and Quataert1994). The slave trade became a “cornerstone” of the Crimean economy (Kizilov Reference Kizilov2007, 2), with some contemporaries estimating that slaves outnumbered natives by more than 2:1 (Davies Reference Davies2007, 25).Footnote ⁵

While seemingly chaotic, slave raids were highly organized. Most raiding expeditions were conducted either in winter—when frozen rivers and grassland could be more easily traversed on horseback—or after the spring thaw and were planned 3–4 weeks in advance (Ivanics Reference Ivanics, Fodor and David2007; Kizilov Reference Kizilov2007).Footnote ⁶ Raiding parties, ranging in size from several hundred to tens of thousands, typically followed one of four trails stretching from the northern edge of the Crimean peninsula deep into the Black Sea steppe.Footnote ⁷ By following these routes, raiders “always traveled between two major rivers, staying on the highest ground” (Beauplan Reference Beauplan1660, 47) and thus avoiding natural barriers and detection by enemy settlements on river banks. Raiders would approach their target area furtively, often traveling on moonless nights and switching between trails to confuse enemy watchmen, while undertaking continuous reconnaissance patrols (Gliwa Reference Gliwa and Roşu2021, 206–9). A fortified field camp would then be constructed, from which raiders fanned outward as far as 140 kilometers, setting buildings alight and conducting demonstrative executions to arouse panic and fear (Gliwa Reference Gliwa2016). Renowned for their speed and mobility, Tatar cavalrymen commanded multiple horses each and deployed a combination of bows and arrows, sabers, spears, and rope to seize captives (up to 7 per soldier) (Kizilov Reference Kizilov2020, 253).Footnote ⁸ Finally, the raiding party would reconvene and return to camp to divide up the spoils.

Captives were marched to Crimea in chains, with the ill and wounded frequently killed to avoid slowing down progress (Fisher Reference Fisher1999, 35). Upon arrival, a small number were retained for ransom—generally nobility and high-ranking military officers—or domestic and agricultural work, while the rest were distributed to one of Crimea’s many slave markets, the largest being the port of Caffa (modern Feodosia).Footnote ⁹ There, an assortment of handlers, gatekeepers, watchmen, and brokers categorized captives according to sex, age, and skill; assigned them to a storage facility; and, within a few days, sold them to a local merchant (Fisher Reference Fisher1999, 35).

The vast majority of purchased slaves were shipped via the Black Sea either to commercial centers across the Ottoman Empire, such as Bursa, Cairo, Constantinople, Damascus, and Edirne, or to smaller towns with slave markets, such as Haskovo, Nova Zagora, and Kazaluk in Bulgaria (Kołodziejczyk Reference Kołodziejczyk2006). Approximately 70% of slaves sold in Caffa made the 10-day journey to Constantinople, where several thousand people—including a guild of 2,000 merchants based in the Grand Bazaar—made a living off the slave trade (Fisher Reference Fisher1999, 584). Male slaves usually ended up working in agriculture, construction, small craft production, or the military; female slaves were used mainly in domestic service, either as concubines of their owner or as servants of his legal wives (Fisher Reference Fisher1999; Lavrov Reference Lavrov and Witzenrath2015).

Scale and Scope: New Geocoded Data

Slave raids in early modern Eastern Europe are known to have occurred frequently and widely while concentrating on the southern peripheries of Poland–Lithuania and Russia in the sixteenth and seventeenth centuries. The full extent of this complex system, however, remains surprisingly unclear. To gain a better empirical understanding, we collected data on the timing, location, and yield of Ottoman-era raids recorded in hundreds of primary and secondary sources, the most important of which are enumerated in Table A1 in the Supplementary Material.

Our data-gathering process proceeded in four steps. First, we attempted to compile a thorough list of raids mentioned in modern scholarship on the Black Sea slave trade in Polish, Russian, Ukrainian, Turkish, English, and Hungarian.Footnote ¹⁰ Second, where possible, we consulted the original historical sources referenced in this research to verify dates and figures, acquire additional information (e.g., more precise locations, raiding party size, collateral damage), and check for unreported raids. While most of these sources take the form of chronicles compiled by monastic or court scribes, they range from property registers and treasury accounts to diplomatic documents and military lists. Third, to err on the side of caution, we excluded raids whose source has been questioned by historians or lacks clarity on whether captives were actually taken. Finally, we geocoded the remaining raids, which in some instances required matching historical and contemporary location names using archival maps.Footnote ¹¹

In total, our dataset comprises 2,511 raids conducted between 1453 and 1774 across 882 geocoded locations—treated as distinct units of observation for raid-counting purposes—spanning 13 contemporary countries: Austria, Belarus, Croatia, the Czech Republic, Hungary, Lithuania, Moldova, Poland, Romania, Russia, Serbia, Slovakia, Slovenia, and Ukraine (data are available at Charnysh and Lall Reference Charnysh and Lall2026). A total of 3.64 million people were captured in these incursions, according to our sources.Footnote ¹² It is crucial to note, however, that information on captive numbers is missing for 56% of raids, implying a far higher true aggregate. Replacing missing values with the mean of observed ones (mean imputation) yields a total of 8.3 million captives. A more sophisticated multiple imputation model, which includes raid longitude, latitude, location type, year, and party size as predictors, provides a more conservative mean estimate of 5.23 million (with a range of 4.41–6.27 million).Footnote ¹³ We find the latter figure, which we discuss in detail in Section B of the Supplementary Material, considerably more plausible, not only because it leverages predictive information in a statistically principled manner but also because raids that lack data on captives are likely to be smaller, on average.

Based on demographic statistics from the History Database of the Global Environment (Klein Goldewijk et al. Reference Klein Goldewijk, Beusen, Doelman and Stehfest2017), our imputed captives total amounts to 27.5% of Eastern Europe’s population in 1400.Footnote ¹⁴ It is not possible to calculate this proportion in the case of Africa’s early modern slave trades, for which scholars have only estimated the aggregate number of exported slaves. The latter figure—approximately 18 million (Nunn Reference Nunn2008)—represents 32.8% of Africa’s population in 1400 (as per the History Database). As a high fraction of Eastern European slaves were exported to Ottoman lands (Kołodziejczyk Reference Kołodziejczyk2006, 151), it seems reasonable to view the Black Sea slave trade as roughly comparable to its African counterparts in terms of overall demographic burden.

Two caveats about our dataset should be mentioned. First, it is unlikely to be complete. Some raids may not have been recorded by contemporaries (for instance, due to their small size or remote location), and not all archival material relating to the Black Sea slave trade may have been accessed by historians.Footnote ¹⁵ While information about raids surely varies with location size and prominence, localities with fewer than 5,000 inhabitants constitute the bulk of our dataset, suggesting that our sources provide good coverage of minor settlements.Footnote ¹⁶ Second, neither raid nor captive numbers should be taken as exact. In addition to the problem of missing data, there is no guarantee that every historical source is accurate (though we have seen no evidence of systematic exaggeration or underreporting). For these reasons, it is prudent to treat our data as indicating approximate orders of magnitude rather than precise quantities.

With these qualifications in mind, the dataset opens a rich window on the scale, scope, and dynamics of slave-raiding in Eastern Europe. Figure 1 plots the annual and cumulative number of raids and (non-imputed) captives between 1400 and 1800. Raids were conducted fairly consistently throughout the period, particularly in the sixteenth and seventeenth centuries. Nevertheless, there are discernible peaks in the first two decades of the 1500s (338 raids), 1610–1630 (414), and 1660–1680 (297). While more sparse due to missing data, captive numbers exhibit similar crests in the first half of the sixteenth century (1.41 million captives) and in the mid-seventeenth century (0.63 million between 1640 and 1660). According to our imputation model, 5 million captives—95% of the total—were taken by 1700. The number of Africans exported in the transatlantic slave trade by this date is estimated at 1.28 million (Slave Voyages 2021).

Figure 1. Frequency of Slave Raids in Eastern Europe, 1453–1774

Note: The left panels display the annual number of raids (top) and captives (bottom, in thousands); the right panels show the cumulative number of raids (top) and captives (bottom, in millions).

Despite the overall magnitude of the Black Sea slave trade, the dataset indicates that raids were typically modest in size. Among raids with information on yield, the average number of captives is 1,450, with 42% producing less than 1,000 captives and only 16% generating more than 5,000 captives.Footnote ¹⁷ In general, the largest raids were conducted during military campaigns led by the Crimean khan, the most famous example being Mehmed I Giray’s 1521 invasion of southern Russia, when a sizable army ravaged towns and villages up to the outskirts of Moscow, forcing Grand Prince Vasilii III to flee to Volokolamsk. Contemporaries estimated that hundreds of thousands of Russians were enslaved during the campaign.Footnote ¹⁸

Figure 2 displays the geographical distribution of slave raids with state borders from the late fifteenth century.Footnote ¹⁹ Raids spanned the full extent of Eastern Europe, ranging longitudinally from the Caucasus Mountains to the Vyatka River basin (roughly 1,300 kilometers) and latitudinally from Lower Austria to the Ural Mountains (roughly 3,000 kilometers). Even so, they were heavily clustered in two areas: the stretch of western Ukraine and southeastern Poland ruled by the Crown of Poland and the Grand Duchy of Lithuania until 1569 and by the Polish-Lithuanian Commonwealth thereafter (65% of all raids); and southern Russia, which was ruled by Muscovy, later known as the Tsardom of Russia (24%).Footnote ²⁰ Not by coincidence, these areas were directly intersected by the four major trails followed by raiding expeditions (indicated by thick lines).Footnote ²¹ In terms of contemporary borders (see Figure A2 in the Supplementary Material), 51% of raids took place in Ukraine, 26% in Russia, 13% in Poland, 3% in Belarus, 3% in Romania, 1% in Hungary, 1% in Moldova, and less than 0.5% in the remaining six countries listed above.

Figure 2. Geographical Distribution of Slave Raids

Note: This map displays the locations of 2,511 slave raids in Eastern Europe between 1453 and 1774.

Pathways to Economic Development

We posit that, while primarily motivated by security considerations, investments in defensive state-building stimulated long-run development in raided areas through three mutually reinforcing processes: (1) the reallocation of human and financial resources from less vulnerable regions; (2) the proliferation and expansion of urban settlements; and (3) the emergence of a more secure and predictable environment for economic activity.

First, the inward flow of labor and capital to construct defense lines, fortresses, and city walls boosted local demand for goods and services. In Poland–Lithuania’s southeastern provinces, an average of 12,000 people per year joined military camps during the first half of the sixteenth century (Łopatecki and Bołdyrew Reference Łopatecki and Bołdyrew2024, 17). This influx prompted an expansion of “urban infrastructure—primarily merchants, transport companies, and artisans’ workshops to provide supplies and luxury goods to the soldiers,” bringing about a “redistribution of wealth from taxes collected throughout the country” (Łopatecki and Bołdyrew Reference Łopatecki and Bołdyrew2024, 31–3). Muscovy’s defensive effort involved even larger population transfers. Some 50,000 infantrymen worked on the Izium Line alone, for instance, and as many as 180,000 combatants and support personnel were mobilized in campaigns against the Crimean Khanate later in the seventeenth century (Davies Reference Davies2007, 171–9). Gentry from western and northern Russia and cossacks from the Pontic-Caspian steppe were lured south with the promise of landed estates and grain, while large numbers of peasants and servitors were forcibly moved (Kollman Reference Kollman2017, ch. 7). Even serfs were encouraged to join garrisons, whose commanders were instructed to retain—not return—enrolling fugitives (Stevens Reference Stevens1995, 26). As a result of these measures, military and economic migration to raided areas more than compensated for losses caused by slave raids.

Second, defensive state-building encouraged the formation and growth of urban communities, expanding markets for agricultural products, facilitating specialization, and creating hubs for trade and investment. In Muscovy, 79 garrison towns were founded in Belgorod and Sevsk—the provinces most exposed to raids—by the late seventeenth century. In Red Ruthenia, a heavily raided region of southeastern Poland, the number of urban settlements doubled between 1500 and 1650 (Bogucka and Samsonowicz Reference Bogucka and Samsonowicz1986, 17). Although established for defensive purposes, many of these settlements soon acquired commercial significance.Footnote ²⁹ For instance, the fortress of Voronezh, constructed in 1585 to monitor the Oka-Don plain for incursions, became the largest city and trading post in southern Russia during the seventeenth century. It was not only new settlements that prospered. Tula, founded in 1381 and subsequently integrated into the Abatis Line, became a renowned center of metalworking and arms production for the Russian army, by the late eighteenth century boasting a population of 25,000 and also producing samovars, hats, gloves, silk, rope, and tiles (Kollman Reference Kollman2017, 383). A striking case is Kyiv, which was sacked by slave raiders so frequently and thoroughly in the fifteenth and sixteenth centuries that it remained “practically empty” for decades (Subtelny Reference Subtelny2009, 83). An intensive fortification effort led by Poland–Lithuania and (from the mid-seventeenth century) MuscovyFootnote ³⁰ dramatically reversed the city’s fortunes over the next 300 years, boosting its population from less than 10,000 to almost a quarter of a million as it morphed into an “administrative, military, commercial, and cultural center” (Subtelny Reference Subtelny2009, 185).Footnote ³¹

Finally, improvements in defensive capacity strengthened the state’s monopoly on violence, enhancing security—and hence the returns to economic activity—in vulnerable areas. In addition to raids, southern borderlands were plagued by recurrent robberies, extortion, and assaults—frequently by cossacks—hampering commerce, finance, and institution-building (Osipian Reference Osipian, Charters, Houllemare and Wilson2020). The state’s growing presence in these territories, combined with the recruitment of cossacks into military service, gradually curbed lawlessness and marginalized banditry and freebooting (Kumke Reference Kumke1993). In Muscovy, garrisoned troops not only protected settlements from raids but also actively monitored Don Cossacks to suppress illicit activity (Davies Reference Davies2007, 198–9). In Poland–Lithuania, the crown army and Registered Cossacks were tasked with preventing raids by independent cossacks as well as Tatars (Osipian Reference Osipian, Charters, Houllemare and Wilson2020). While cossacks continued to periodically challenge state authority, the reduced incidence and threat of violence in raided territories created a safer and more stable climate for trade and investment.

The preceding discussion suggests that the developmental consequences of slave-raiding in early modern Eastern Europe may be better understood by analogy with interstate warfare in contemporary Western Europe than with reference to the transatlantic slave trade. Wars inflicted immediate economic damage on Western European states, reducing populations, fueling epidemics, and destroying capital stock. Over a longer time span, however, they set in motion a sustained process of state-building—Tilly’s (Reference Tilly1990) famous “war made the state” thesis—that is widely viewed as instrumental in the region’s subsequent economic ascent.Footnote ³² Faced with cutthroat military competition, bellicist theories contend, rulers centralized coercive structures, consolidated fiscal systems, and professionalized bureaucratic institutions. Enhanced state capacity, in turn, accelerated economic growth by expanding the provision of public goods critical to the effective functioning of markets, such as the rule of law, security, and basic infrastructure (Besley and Persson Reference Besley and Persson2010; Johnson and Koyama Reference Johnson and Koyama2017). Before wars built states, moreover, they built cities. Dincecco and Onorato (Reference Dincecco and Onorato2017) highlight how the vulnerability of rural populations to conflict drove migration to increasingly fortified urban settlements from the medieval period onward, which fueled development by strengthening self-governance and property rights protection and encouraging technological innovation, human capital accumulation, and economic agglomeration.

At the same time, it is important to recognize the limits of the “war made the state and city” analogy. First, early modern Eastern Europe was dominated by a small number of sizable polities whose existence was not threatened by slave-raiding. Tatars sought to neither conquer nor occupy territory, lacking the numbers and firepower necessary to succeed in conventional battle (Gliwa Reference Gliwa2016). Bellicist theories suggest that “raiding wars” are unlikely to deliver a strong stimulus to state-building, though evidence for this implication has largely been confined to Sub-Saharan Africa (e.g., Dincecco, Fenske, and Onorato Reference Dincecco, Fenske and Onorato2019). Second, population mobility and urban autonomy were comparatively limited in Eastern Europe. The institutionalization of serfdom—eradicated in Western Europe following the Black Death—bound peasants to hereditary plots of land, and most cities were legally private entities owned by noblemen, the church, or the monarch, to whose needs they were subordinated (Bogucka Reference Bogucka1996, 325; Kollman Reference Kollman2017, 238).Footnote ³³ Rather than fostering urban self-governance, slave raids strengthened the state’s incentives to dominate and extract tax revenues from cities, ends to which the deployment of military forces served as a key means. As a result of these differences, the stimulus to state-building, urbanization, and long-run development delivered by slave-raiding in Eastern Europe exhibited a more localized and “top-down” character than that provided by interstate warfare in Western Europe.

Data and Empirical Strategy

Our analysis combines the compendium of slave raids introduced earlier with the European Urban Population, 700–2,000 database compiled by Buringh (Reference Buringh2021). The latter source, which updates and expands Bairoch, Batou, and Pierre’s (Reference Bairoch, Batou and Chèvre1988) seminal population tables using recent archaeological and demographic research, records the number of inhabitants (in thousands) in 550 urban settlements across 23 Eastern European countries—24% of which were raided at least once—at one-century intervals from 700 to 1500 and at half-century intervals from 1500 to 2000.Footnote ³⁴ Bairoch, Batou, and Pierre include all settlements with 5,000 or more inhabitants at some point between 800 and 1800; Buringh adds those with at least 1,000 inhabitants in 700, at least 100,000 inhabitants in 2000, and capital city status in 2000.Footnote ³⁵ While its relatively infrequent reporting intervals inhibit our ability to detect short-run population losses caused by raids, the database paints a reasonably comprehensive and precise picture of European urban demographic trends over the long run.

Figure 3 plots the average population of raided settlements and non-raided settlements between 1100 and 1900 (leaving at least a century before and after the slave trade). Over the three centuries preceding the slave trade, population evolved similarly in the two types of settlements, whose means are indistinguishable in a two-sample t-test. Once the slave trade begins, however, raided settlements exhibit consistently faster population growth than non-raided settlements, with the difference in means becoming statistically significant from 1550 onward. By 1900, the mean population of raided settlements (51,238) was more than 50% higher than that of non-raided settlements (33,411). Descriptive trends thus furnish initial plausibility for our expectations.

Figure 3. Mean Population of Raided and Non-Raided Settlements in Eastern Europe, 1100–1900

Note: Mean population (in thousands) of raided and non-raided urban settlements in Eastern Europe between 1100 and 1900; p-values are from a two-sample t-test of the difference in means between the two sets of settlements. The lower panel displays the total number of raids per year.

To analyze the impact of exposure to raids on settlement population, we adopt a difference-in-differences strategy that compares the average change in the population of raided and non-raided settlements after the onset of raids. Our baseline specification can be expressed as

(1) $$ \log \left({P}_{st}\right)=\alpha +\beta {R}_{st}+{\gamma}_s+{\delta}_t+{\varepsilon}_{st}, $$

where $ {P}_{st} $ , the outcome variable, is the population of settlement $ s $ (in thousands) in period $ t $ ;Footnote ³⁶ $ {R}_{st} $ , the treatment variable, is an indicator for whether $ s $ has been raided as of $ t $ ; and $ {\gamma}_s $ and $ {\delta}_t $ denote settlement and period fixed effects, respectively.Footnote ³⁷ We cluster heteroskedasticity-robust standard errors at the settlement level. To examine how the treatment effect evolves over time, we additionally estimate a “dynamic” version of Equation 1 that interacts $ {R}_{st} $ with $ {\delta}_t $ , including leads of $ {R}_{st} $ during the pretreatment phase (one lead for every period before 1500) as a placebo test (Sun and Abraham Reference Sun and Abraham2021).Footnote ³⁸

Provided that the population of raided and non-raided settlements would have followed the same trajectory in the absence of the slave trade, $ \beta $ in Equation 1 identifies the average effect of raid exposure on settlement population during the sample period. While Figure 3 is consistent with this pattern, a visual comparison of outcome means for treated and untreated units does not shed direct light on the parallel trends assumption because it fails to account for potentially confounding variables.Footnote ³⁹ Building on a growing methodological literature (Roth et al. Reference Roth, Pedro, Bilinski and Poe2023), we instead assess the plausibility of this assumption through an event study specification that analyzes the treatment effect’s evolution before and after initial exposure to raids:

(2) $$ \log \left({P}_{st}\right)=\alpha +\sum \limits_{j=-8}^8{\beta}_j{D}_{s,t-j}+{\gamma}_s+{\delta}_t+{\varepsilon}_{st}, $$

where $ {D}_{s,t-j} $ is an indicator for period $ j $ relative to the first raid on settlement $ s $ as of period $ t $ (meaning that this raid occurred $ j $ periods before $ t $ ).Footnote ⁴⁰ Following common practice, we specify the period before the first raid as the reference for estimating treatment effects (by setting $ {\beta}_{-1} $ to 0). To avoid inferential problems arising from treatment effect heterogeneity, we supplement this specification with three alternative event study estimators that exclude potentially problematic comparisons between already-treated units: Sun and Abraham’s (Reference Sun and Abraham2021) interaction-weighted estimator, which compares treated units with never-treated or last-to-be-treated units; Liu, Wang, and Xu’s (Reference Liu, Wang and Xu2024) imputation-based estimator, which imputes counterfactual outcomes for treated units; and Callaway and Sant’Anna’s (Reference Callaway and Sant’Anna2021) doubly-robust estimator, which specifies never-treated or not-yet-treated units as the comparison group. In all models, robust standard errors remain clustered by settlement.

Identification Strategy

In the absence of time-series data on our district-level development outcomes, a major inferential challenge is the possibility that the location of slave raids is endogenous to omitted determinants of development—or to development itself. It should be noted, however, that neither our urban population analysis nor historical accounts suggest that prosperous locations were a more attractive target; on the contrary, raiders were known to favor rural areas with weak defenses (Gliwa Reference Gliwa and Roşu2021, 197).

To address potential endogeneity in the geographical distribution of raids, we pursue an instrumental variables strategy that leverages natural topographical features affecting districts’ accessibility to raiders. Our approach is motivated by the observation that the trails used to conduct raids closely follow the boundaries between watershed zones—areas of highest elevation around lakes and river segments—enabling raiders to avoid deep ravines, marshland, steep slopes, and river crossings as well as discovery by enemy watchmen. A district’s proximity to watershed boundary lines should thus predict its accessibility to raiders with comparable precision to its distance from raiding trails. Since watershed zones solely reflect characteristics of the steppe terrain, however, they are more plausibly exogenous to long-run development and its correlates. We thus instrument a district’s exposure to raids with its minimum distance to Perekop or Akkerman—the typical starting points for raiding expeditions from the northern Black Sea coast—along a watershed boundary.Footnote ⁴⁸

A key threat to the exclusion restriction is the potential for our instrument to impact development directly, for instance, because watershed boundaries overlap with historical trade routes connecting to the Black Sea coast or because proximity to Crimea correlates with soil fertility and favorable climatic conditions (Moon Reference Moon2013). There is no historical evidence to suggest that watershed boundaries served as conduits for trade; nor does this possibility seem likely, given that goods were usually transported via rivers, proximity to which we control for in our instrumental variables specification.Footnote ⁴⁹ We also account for terrain ruggedness, soil fertility, temperature, and other climatic and topographical characteristics influencing agricultural potential. Our identifying assumption is that, conditional on these natural features, a district’s proximity to Perekop and Akkerman via the watershed boundary network only affects its long-run development through its exposure to raids.

As an alternative approach, we follow Blaydes and Paik (Reference Blaydes and Paik2021) and Matranga and Natkhov (Reference Matranga and Natkhov2022) by instrumenting raid exposure with distance to the geographically most efficient routes—or “least-cost paths”—between the starting points (Akkerman and Perekop) and destinations (Lviv and Moscow) of raiding trails. As detailed in Section E of the Supplementary Material, we construct nine such paths using an algorithm that minimizes cumulative resistance to the flow of water from one endpoint to the other, which generally increases with terrain gradient and decreases with elevation.Footnote ⁵⁰ The advantage of this strategy is that least-cost paths provide a slightly closer approximation to raiding trails—which are less convoluted than watershed boundaries—increasing the instrument’s predictive power.

We implement these approaches using a two-stage least squares (2SLS) estimator. The first stage models a district’s aggregate exposure to raids as a function of the instrument plus a battery of controls:

(3) $$ \log \left({R}_d\right)={\varphi}_0+{\varphi}_1{L}_d+{\varphi}_{\mathbf{X}}{\mathbf{X}}_d^{\prime }+{\varepsilon}_d, $$

where $ {R}_d $ is the cumulative number of raids on district $ d $ over the course of the slave trade; $ {L}_d $ is $ d $ ’s minimum kilometer distance either to Crimea via watershed boundaries (main instrument) or to one of our nine least-cost paths (alternative instrument); and $ {\mathbf{X}}_d^{\prime } $ is a vector of district-level controls described below. In the second stage, we regress a given development outcome on predicted values of $ {R}_d $ and the same set of controls:

(4) $$ {Y}_d=\alpha +\hat{\beta \log \left({R}_d\right)}+{\psi}_{\mathbf{X}}{\mathbf{X}}_d^{\prime }+{\varepsilon}_d. $$

If the exclusion restriction holds, $ \beta $ captures the average change in $ {Y}_d $ resulting from a 1% increase in cumulative raids on a district due to the topographical characteristics captured by the instrument.

Analysis of Imperial Russian Districts

We first apply our instrumental variables strategy to districts (uezdy) of Imperial Russia, which are mapped in Figure 5.Footnote ⁵¹ Drawing on data collected from imperial statistical volumes and administrative records by Dower et al. (Reference Dower, Finkel, Gehlbach and Nafziger2018), we construct three sets of outcome variables: (1) population (1863), logged urban and per square kilometers; (2) the number of markets (1867), logged and per square kilometers; and (3) the number of factories (1867), logged and per square kilometers. In both estimation stages, we control for several variables critical to the exclusion restriction’s plausibility: mean terrain ruggedness, computed with raster data from Shaver, Carter, and Shawa (Reference Shaver, Carter and Shawa2019); average annual precipitation and temperature seasonality, from the WorldClim 2 dataset (Fick and Hijmans Reference Fick and Hijmans2017); straight-line distance to the nearest river and coastline; soil fertility, as operationalized in the urban population analysis; the logarithm of urban population in 1400 (a proxy for pre-slave trade development), calculated using the European Urban Population data; the number of military conflicts during the slave trade, measured with the Historical Conflict Event Dataset; the logarithm of land area (in square kilometers); distance to Moscow; and an indicator for whether the district lay within Muscovy’s 1490 borders.Footnote ⁵²

Figure 5. Distribution of Slave Raids in Imperial Russia and Austrian Galicia and Silesia

Note: Cumulative slave raids on locations within mid-nineteenth-century districts of (1) the Russian Empire, excluding Tatar khanates and territories under Ottoman control during the Black Sea slave trade, and (2) Austrian Galicia and Silesia.

Panel A in Table 3 reports second-stage 2SLS estimates and first-stage F-statistics (bottom row). As indicated by the latter, a district’s minimum distance to Perekop and Akkerman along a watershed boundary is a strong predictor of its exposure to raids. In the second stage, the coefficient on cumulative raids is positive for all six development outcomes, statistically significant for four, and close to significant for the remaining two. A 1% rise in aggregate raids is associated with 0.49% more urban inhabitants, 0.8% more markets, and 0.32% more factories. Per 100 square kilometes, this translates into an additional 7 inhabitants, 0.003 markets, and 0.007 factories.

Table 3. Slave Raids and Development Outcomes in Imperial Russia and Austria: Instrumental Variables Estimates

Note: 2SLS estimates of the impact of slave raids, instrumented by minimum distance to Perekop or Akkerman along a watershed boundary line, on district-level development outcomes in mid-nineteenth century Russia (panel A) and Austrian Galicia and Silesia (panel B). All models control for urban population in 1400, land area, distance to the nearest river and coastline, soil fertility, terrain ruggedness, and cumulative military conflicts during the raiding period. In panel A, temperature seasonality, precipitation, and distance to Moscow are also included. Robust standard errors in parentheses; Conley standard errors (cutoff = 500 kilometers in panel A, 250 kilometers in panel B) in brackets. For full first- and second-stage results, see Tables A15–A17, Section E of the Supplementary Material. * $ p<0.1 $ ; ^** $ p<0.05 $ ; ^*** $ p<0.01 $ .

In Section E of the Supplementary Material, we confirm that estimating Equation 4 using observed rather than predicted values of cumulative raids—that is, substituting OLS for 2SLS—does not materially alter the results (panel A, Table A16 in the Supplementary Material). In addition, we demonstrate robustness to our alternative instrument, distance to nine least-cost paths from Akkerman and Perekop to Moscow and Lviv (panel A, Table A19 in the Supplementary Material).

Analysis of Austrian Galicia and Silesia

Another part of Eastern Europe for which rich local development data from the mid-nineteenth century have recently become available is the southern strip of the Polish-Lithuanian Commonwealth annexed by the Austrian Empire in the eighteenth century. This relatively small and homogeneous area comprised the Polish regions of Galicia and Silesia, whose 99 districts (Kreise) varied markedly in their exposure to slave raids (see Figure 5).

Using geocoded data gathered from censuses, historical maps, and satellite images by Kaim et al. (Reference Kaim, Szwagrzyk, Dobosz, Troll and Ostafin2021), we construct six outcome variables measuring the number and density of a district’s inhabitants (1857–69), houses (1840–63), and farm structures (1840–63). Following our earlier identification strategy, we regress these measures on the logarithm of cumulative slave raids on a district instrumented by its minimum distance to Akkerman or Perekop via a watershed boundary line. In both stages of the 2SLS specification, we include a similar set of controls to the Imperial Russia analysis.Footnote ⁵³

The results are presented in Panel B of Table 3. The first-stage F-statistics (bottom row) again confirm that the instrument strongly predicts raid exposure. The second-stage estimates reveal a consistently positive treatment effect that reaches statistical significance for five of the six development outcomes. A 1% increment in cumulative raids raises a district’s population by 0.36%, its housing stock by 0.25%, and its number of farm structures by 0.04%, which amounts to 672 more inhabitants, 20 more houses, and 15 more farm structures per 100 square kilometers.

Once more, the results are robust both to using observed rather than predicted values of the treatment variable (panel B, Table A16 in the Supplementary Material) and to instrumenting cumulative raids with minimum distance to a least-cost path (panel B, Table A19 in the Supplementary Material).

The Defensive State-Building Mechanism

What explains the positive association between exposure to slave raids and long-run development in Eastern Europe? This section presents a selection of additional, more systematic evidence for our argument that raids stimulated economic growth by incentivizing states to strengthen defensive capacity in targeted areas—both by deploying military forces and by constructing permanent fortifications—and, to this end, consolidating fiscal and administrative capacity. Statistical results not reported in this section can be found in Section F of the Supplementary Material.

Poland–Lithuania

We begin by examining how military deployments and fiscal revenues responded to raiding activity in Poland–Lithuania. Drawing on data from the first half of the sixteenth century assembled by Łopatecki and Bołdyrew (Reference Łopatecki and Bołdyrew2024), Table A20 in the Supplementary Material shows that the number of military servicemen stationed in Red Ruthenia rose by 53%, on average, in years after the region was subjected to raids. A similar pattern characterizes the value of tax collections by the Polish and Lithuanian treasuries between 1588 and 1647—as recorded by Filipczak-Kocur (Reference Filipczak-Kocur2006)—following raids on the state.Footnote ⁵⁴

Second, digitizing and geocoding data gathered by Adamczyk (Reference Adamczyk2004), we analyze the impact of raid exposure on the construction of five types of permanent fortifications—major castles, minor castles, fortified towns, fortified villages, and fortified churches—in southern provinces of Poland–Lithuania (mapped in Figure A11 in the Supplementary Material). As available construction dates are imprecise, we compute the density of each type of fortification per grid cell $ g $ at 50-year intervals between 1100 and 1800, focusing on the approximately rectangular (2,000 km $ \times $ 1,500 km) area studied by Adamczyk.Footnote ⁵⁵ We regress these measures on $ {R}_{gt} $ plus the logarithm of cumulative battles in $ g $ in $ t $ , grid cell fixed effects ( $ {\omega}_g $ ), and period fixed effects ( $ {\delta}_t $ ). As reported in panel A of Table 4, raid exposure is positively and significantly related to the density of all five fortification types. On average, raided grid cells saw the construction of 1.25 more major castles, 0.44 more small castles, 0.79 more fortified towns, 0.04 more fortified villages, and 0.31 more fortified churches over the sample period.Footnote ⁵⁶

Table 4. Slave Raids, Fortification Construction, and Castle Ownership in Poland–Lithuania: Difference-in-Differences Estimates

Note: Difference-in-differences estimates of the impact of exposure to slave raids on the construction of fortifications (panel A) and various measures of castle ownership (panel B) in Poland–Lithuania at the grid cell (0.5° x 0.5°) level. The sample comprises southern provinces from 1100 to 1800 in panel A and central and western provinces from 1300 to 1800 in panel B. Robust standard errors, clustered by grid cell, in parentheses; Conley standard errors (distance cutoff = 500 kilometers) in brackets. * $ p<0.1 $ ; ^** $ p<0.05 $ ; ^*** $ p<0.01 $ .

Third, we adopt a similar strategy to investigate whether states were more successful in establishing a monopoly on violence in raided areas, drawing on records of castle ownership in central and western Poland–Lithuania from the CASTLES dataset (Cappelen and Hariri Reference Cappelen and Hariri2022; Reference Cappelen and Hariri2024). For each grid cell in each half-century between 1300 and 1800, we measure the density and share of castles controlled by the crown and its representatives, which we regress on the same right-hand-side variables as in the previous analysis.Footnote ⁵⁷ Panel B of Table 4 indicates that these outcomes exhibit a statistically significant rise in grid cells exposed to raids. Consistent with this finding, raid exposure has a positive but nonsignificant association with the density of castles controlled by non-crown actors, such as the nobility, the church, and external monarchs (column 8).Footnote ⁵⁸

Russia

Next, we shift our focus to the Tsardom of Russia to study how slave-raiding impacted military and commercial activity in urban communities with the right to conduct commerce or industry (posady) (see Figure A13 in the Supplementary Material for a map). Digitizing and geocoding census statistics recorded at four intervals between 1646 and 1722 by Vodarskii (Reference Vodarskii and Shunkov1966), we measure five outcomes: the logarithm of military servicemen (sluzhilye) at (1) the household (dvor) level in 1650 and (2) the individual level in 1670–1680;Footnote ⁵⁹ and the logarithm of tax-paying traders and artisans (posadskie) at the household level in (3) 1646 and (4) 1670–1680 and (5) 1722.Footnote ⁶⁰ Extending the instrumental variables strategy employed in the previous section, we find that the cumulative number of raids on an urban community is positively associated with its population of military servicemen but negatively associated with its population of traders and artisans (see Table 5). Interestingly, however, the latter relationship is reversed by the early eighteenth century, with the coefficient on the instrumented raids treatment becoming positive but not significant.Footnote ⁶¹

Table 5. Slave Raids, Defensive State Capacity, and Commercial Activity in Russian Urban Communities: Instrumental Variables Estimates

Note: 2SLS estimates of the impact of slave raids, instrumented by minimum distance to Perekop and Akkerman along a watershed boundary line, on the population of military servicemen and traders and artisans in Russian urban communities (posady) between 1646 and 1722. All models control for minimum distance to a river and to a coastline, date of founding or first mention, soil fertility, distance to Moscow, and inclusion in Muscovy’s 1490 borders. Robust standard errors in parentheses; Conley standard errors (distance cutoff = 500 kilometers) in brackets. For full results, see Table A24, Section F of the Supplementary Material. * $ p<0.1 $ ; ^** $ p<0.05 $ ; ^*** $ p<0.01 $ .

The urban communities data also allow us to probe our argument’s implication that strengthening defensive capacity promoted development by fostering a more secure environment for economic activity. Table A26 in the Supplementary Material reveals that, holding constant the number of traders and artisans in 1646 as well as the controls from Table 5, communities with larger contingents of military servicemen in the seventeenth century boasted significantly larger commercial populations in 1722. In other words, exposure to raids appears to have bolstered military presence before fueling an expansion in commercial activity at the local level.