The Trans-Atlantic Slave Trade Database

One immediate question is how complete are the data? It is probable that our data set now includes more than 95 percent of all voyages that left British ports—and the British were the second largest of the national slavetrader groups. The data on the eighteenth-century French and Dutch slave trades are also largely complete. The reasons for such comprehensive coverage are fairly obvious. Compared with other slave traders, northwestern European trading nations conducted the great bulk of their business relatively late in the slave trade era when everyone kept better records. Surviving sources in these countries are therefore abundant. Casual inspection ofthe relevant variables in the data set shows that almost all the voyages leaving ports in these countries have more than one source of information, and some have as many as eighteen. Yet the data on the Iberian and Brazilian trades after 1750 are also relatively complete, and information on the earlier period for these regions is vastly greater than it was ten years ago. For a country by country assessment of the completeness of the data, readers are referred to Chapter 1 of Extending the Frontiers and the spreadsheets downloadable from this web site that underpinour estimates of the overall size of the slave trade. 

The 36,000 trans-Atlantic voyages contained in the database allows us to infer the total number of voyages carrying slaves from Africa. The Estimates page suggests that 12 ½ million captives (12,520,000) departed Africa for the Americas. Dividing this total by the average number of people embarked per voyage, 304 individuals, yields 41,190 voyages. Similarly, the Estimates pages suggests that 10.7 million enslaved Africans disembarked,mainly in the Americas. Dividing by the average number disembarked per voyage, 265 people, yields an estimated 40,380 voyages arriving. Not all 36,000 voyages in the database carried slaves from Africa. A total of 633 voyages (1.8%) never reached the African coast because they were lost at sea, captured or suffered some other misfortune. After removing these voyages, the database contains some trace of 85 percent of voyages that embarked captives. The database also contains records of 34,106 voyages that disembarked slaves, or could have done so (in other words, for some of these we do not know the outcome of the voyage). A total of 668 of these disembarked their slaves inthe Old World. The latter group comprised mainly ships captured in the nineteenth century which were taken to Sierra Leone and St. Helena as part of the attempt to suppress the trade. A further 277 sank after leaving Africa with theloss of their slaves. In all, the database contains some record of almost 80 percent of vessels disembarking captives. Of course, there are other estimates of the volume of the trans-Atlantic slave trade. If we take a higher estimate of, say, 15.4 million departures, (2) then the Voyages Database documents two-thirds of all slaving voyages that sailed between 1514 and 1866.

Much of what is new in this data set lies in the sources, which call for some discussion. The published data draw on a wide range of published and archival information. Postma, Mettas, and Richardson used new material in the preparation of their published catalogues of voyages. Our data set does not reproduce all the sources that previous authors used and listed. Thus, voyages drawn from published sources are represented here by that single reference to them alone. Anyone wishing to consult their archival references will have to look them up in the hard copy of these published volumes. We edited individual voyage details in these collections only when we found new documentary evidence to support a change; consequently, we added these new references to the source record. Pulling together the results of work carried out in separate national archives was particularly fruitful because trans-Atlantic slave vessels could clear from one jurisdiction and arrive in the Americas in another. Specifically, the international nature of the slave trade meant that a voyage that might appear primarily in one national body of records had a very good chance of showing up, in addition, in the records of other countries. Thus, voyages organized by London merchants operating independently of the Royal African Company (RAC) in the 1670s and 1680s obtained their slaves in South-east Africa, outside the RAC’s English monopoly and where the English East India Company had little physical presence. Because almost all these ships called at the Cape before beginning their trans-Atlantic journeys, several of them appear in the Dutch Cape of Good Hope archives.(3) Portuguese ships leaving Bahia in Brazil for the "Mina"coast appear in English Cape Coast Castle material at the Public Record Office. Slave ships of every nationality appear in Lloyd’s Lists, and of course the hundreds of slave ships captured in the many European wars are often carefully documented in the archives of the captors, as well as in the records of the nations to which the ships belonged. Indeed, the South Atlantic Portuguese trade has fewer sources per voyage, precisely because this branch of the traffic operated independently of the others. Winds and ocean currents kept the South Atlantic trade out of the non-Portuguese archives,as well as keeping northern Europeans out of South Atlantic ports.

Of the 36,000 voyages in the data set, 12,560 have only one source listed. But in fact, more than half of these apparently single-source records are taken from already published material where, in nearly all cases, additional primary references are to be found. Furthermore, other publications on which the present data set draws, such as Coughtry’s listing of Rhode Island voyages, are based on a range of primary documents which are not listed by voyage in the publication itself. After allowing for these multi-sourced single references, it would appear that perhaps only one in six of the 36,000 voyages are based on a single historical record. Sixty percent of the voyages in the set have three or more separate sources each. The full reference may be viewed by rolling the cursor over the abbreviated source or by clicking on the row of a voyage to view voyage details. Alphabetic listings of references and full descriptions of locations are to be found in Home> Voyages Database> Understanding the database> Sources .

While the sources are relatively rich, diversity brings a new set of problems. We can hardly expect that reports on voyages made several thousand miles—as well as several months—apart, often in different languages and under different bureaucracies, each with a separate set of official procedures to follow, to always generate perfectly consistent information. For example, over 200 voyages in the data set apparently arrived in the Americas with more slaves on board than when they left Africa. Others left port more than once on the same voyage, and some ships reportedly changed tonnage and even rig in the course of the voyage.The same ship occasionally appears under more than one name on the same voyage. Those used to working with a single source per voyage and generating data sets without any conflicting information should be warned that the editors have not attempted to correct all these problems. The data set offered here is by no means "clean" in the sense of being entirely internally consistent.We have pursued and eliminated many of the inconsistencies, but to eliminate all would have imposed an order on the historical record that anyone who has visited the archives (or indeed examined published sources such as Mettas or Richardson) knows does not exist. The editors always entered only one value per variable when faced with alternative information. In making such choices, we followed certain rules that researchers can change after going back to the sources. If users elect to do this, however, they, too, will have a set which is both not "clean" and not necessarily reflective of the historical records from which it is drawn.

New material tends to raise the question of the appropriateness of the variables used. The selection offered here has changed several times in the last decade and will no doubt change again in the future as interests shift. Each entry in our data set is a single voyage, assigned a unique identification number as the first of piece of information (VOYAGEID). The question of what voyages to include is also to some extent arbitrary. The term "trans-Atlantic" is less straightforward than it appears. Omitting ships sailing to the Mascarene Islands was an easy decision, but several French ships in the late eighteenth century began their slaving activities in the Indian Ocean, but then on the same voyage brought slaves to the Americas after selling some Africans in Bourbon and the Cape of Good Hope. What to do with the British ships that carried hundreds of children from the Upper Guinea coast to Lisbon in the mid-eighteenth century? These we included on the basis of length of voyage. Should one include the Portuguese trade to São Tomé in the Bight of Biafra—probably the most enduring branch of the Atlantic slavetrade? (excluded on the same basis). Then there were the more than 1,200 slave ships engaged in trans-Atlantic voyages, nearly half with slaves on board, that the British captured and carried into Sierra Leone, the Cape of Good Hope, St. Helena, Fernando Po, or Luanda, before they were able to reach their intended American destinations. These we included. Or even more confusing, the 1,060 slaves awaiting shipment in barracoons in Ambriz, Angola, in May 1842, but carried off in British cruisers to St. Helena and Sierra Leone and never subjected to court proceedings of any kind, because they had never been on board a slave ship (excluded). (4) Limits had to be established, but the data set provides a basis for those who disagree with those limits to use our work to create their own data sets. A total of 74 variables are made available on the search the database interface, general category, and 293 variables are available in the downloadable version of the database. Users should note, though, that the website set combines all sources into one variable and the day, month, year values are also combined into one variable for each of the nine dates entered. The names of captives, names of Caribbean agents, names of crew other than captain, the details of shipboard insurrection, and much other information are not included in the present data set, but may be added fairly easily or linked with it via the unique voyage identification number.

The database contains two broad types of variables: data variables and imputed variables. The largest group, 58 of those in the search interface, are data variables, including derived variables - the latter comprising percentages calculated from the data, or geographic regions when we know a particular port. They incorporate information collected from the sources. Imputed variables, indicated by the abbreviation "IMP" or "imputed," are drawn from data variables, but where data are unavailable, resonable inferences are oten possible.  There are seven of these and they are discussed more fully below.  In augmenting the number of voyages on which analysis can be conducted, the imputed variables produce more statistically significant results in using the options in the “Search the database” interface to create tables and custom graphs. They form the basis of the tables and graphs that users can build.

The variables include information that, for convenience, have been grouped into eight categories:  (1) year range; (2) vessel characteristics (name, tonnage, rig, guns, place and  year of construction, owners); (3) the itinerary of the voyage; (4) the numbers, age/sex and mortality of enslaved people; (5) dates at which the vessel left and arrived; (6) the captain and crew of the vessel; (7) the outcome of the voyage; and (8) the sources for the record. The Variable List in the “Understanding the Database” section presents a complete listing of the data variables as well as the imputed variables in the data set. Labels for imputed variables are always followed by the abbreviation “IMP”. No voyage, however, includes information for all data or imputed variables.

Generally, we attempted to preserve the written documentary record in adding to the data variables. Numeric variables, such as vessel tonnage, numbers of crew, and numbers of slaves, demanded a ranking of sources, particularly for the well-documented British trade. (5) Sources often report different numbers of slaves embarked on or "taken on board" the coast of Africa or landed in the Americas. Furthermore, for some years there are inconsistencies in slave age or gender totals per voyage. Regarding slave exports, we were careful to distinguish between the number of slaves purchased and the number who in fact were shipped from the coast. We used slave departure totals, whether reported by slave traders, African merchants, or European captains, agents, or merchants. We included the slave departures reported in sources such as logs kept by the Dutch and English castles on the Gold Coast, even though the totals often were rounded numbers, such as 400 or 500 slaves, and even though the totals occasionally were significantly less than the numbers of slaves who were disembarked in the Americas. Users should keep these biases in mind, not least for any calculations of mortality they may wish to try. For slave arrivals recorded in customs documents or shipping gazettes, we decided to use maximum totals where there was conflicting information, under the assumption that these differences might indicate deaths of slaves before slaves disembarked.

Age categories must also be used with care. (6) The expanded version of the Voyages Database includes four variables distinguishing the number of captives by age and gender: adult males ("men"), adult females ("women"), male children ("boys"), and female children ("girls"); three variables classifying slaves by age, without specifying gender: “adults,” “children,”and “infants” (reported often as "infants at the breast"); and two variables classifying them by gender, but not by age: “males” and “females.” These data variables replicate how slaves were categorized in primary sources.

Age and sex definitions changed over time and among carriers. Arrivals in the early Iberian Americas were assigned a ratio of what a prime male slave would cost—the latter being termed a pieza de Indias. A child would receive a rating of half a pieza, a woman 0.8, and so on. It has not proved possible to infer age and gender breakdowns from aggregated pieças. In the 1660–1730 period, the London-based Royal African Company (RAC) defined children as about ten years of age or younger. For most of the British and French slave trades, a height (about four feet four inches) and/or age (about puberty) criterion distinguished adults from children. In the nineteenth century, captured slave ships of all nations, but mainly Spanish and Brazilian, had their human cargoes recorded by a variety of courts, some British, some international. There is little doubt that the criterion used to separate out adults was sexual maturity as assessed by physical appearance, which for most Africans at this time would probably occur in the mid-teens, but could vary according to the diet prevalent in the areas from which Africans were drawn as well as according to the eye of the purchasers. Yet another categorization emerges from Cuban slave trade data (1790–1820) taken from the Seville archives, which adds "men-boys" and "women-girls" to the previous categories. These we included among men and women, respectively.

All these measurements are of course imprecise, with even a clear age definition of "ten years and younger" hinging on casual inspection by Europeans, because many African cultures did not attach importance to knowledge of exact ages. In nineteenth-century court records, different officials often recorded slightly different distributions of the same group of slaves. However, in documents with information on height of slaves, classification as an adult correlates well with height (specifically the teenage growth spurt) and sexual maturity. Thus the RAC’s definition of children as under the age of ten excluded individuals that other definitions would have included as children. As the RAC records form the bulk of the age and gender information for 1660–1710, the share of children for this period is biased downward.

Dates of slave voyage sailings and arrivals are useful categories for sorting voyages in the data set. We therefore decided to broaden the definition of "departure" dates and we estimated years of arrival in the Americas for select voyages. For the non-British slave voyages, dates of departure generally refer to the date the vessels sailed from port, although inthe records of massive departures from Bahia de Todos os Santos in Brazil, the date of the issue of the license to export tobacco (the only data available for hundreds of voyages) is sometimes more than a month different from the date of permission to clear port (where this source is also available). Similar variance appears in the large British trade. We assumed that the datecrew entered pay (listed on Bristol or Liverpool muster rolls) was the date of departure (or close to it); and, for many voyages,this assumption is confirmed by analysis of other sources. Other "departure" dates included in the new slave trade data set are:the date vessels cleared customs; the dates Mediterranean passes were issued; the dates bonds were given; the dates sailors’ pension monies were lodged; and the dates of vessel registration. These events usually occurred within 1-2 months of departure.  London slave vessels cleared customs at Gravesend and often sailed from the "Downs," the shallows off the Kent coast. During contrary south-westerly winds, departures from the Downs were often delayed for several weeks. We included these dates of sail from the Downs as London "departure" dates occurred within 1-2 months of departure. London slave vessels cleared customs at Gravesend and often sailed from the "Downs," the shallows off the Kent coast. During contrary south-westerly winds, departures from the Downs were often delayed for several weeks. We included these dates of sail from the Downs as London "departure" dates when no other information was available. A separate variable defines these various "departure" dates. Reports of slave vessel arrivals in the Americas generally reached Europe within six to ten weeks. Without other documentary support we assumed the year or (in select cases) month of arrival, when the evidence was clear from the timing of gazette listings. Bristol and Liverpool muster rolls frequently record the dates the crew deserted ship or were discharged from pay in the Americas. For many voyages we assumed that these were arrival dates. When multiple dates were reported in the sources, the editors chose the latest dates for departures and the earliest dates for arrivals.(7)

One further convention is adopted for vessels captured off Africa without slaves. Some voyages in the data set that were captured by pirates or ships of other nations display a date of departure from Africa even though they had no slaves on board. In such cases the date of departure from the coast should be read as the date of capture. The great majority of such instances occurred in the nineteenth century when British naval vessels captured many empty slave vessels as part of their efforts to suppress the slave trade.

The data set includes variables that report the place and year of construction for most vessels in the British slave trade. Merchants often purchased ships from other trades and converted them to slavers. The British slaving fleet also included prizes captured from the French, Spanish, Dutch, or Americans during the many wars of the eighteenth century. Shipping sources, such as Lloyd’s Registers of Shipping, often do not distinguish whether a vessel was "French-built" or a "French prize." Similarly, the year of construction in the documents may refer to the year the vessel was captured, made free or rebuilt. As it is unlikely that many British merchants purchased vessels built in France in an open market, the user should assume that a "French-built" vessel was likely a war prize. Vessels reported as "French prizes," on the other hand, may not have been built in France. These ships could in fact have been built in Britain and subsequently captured and renamed by French merchants.

We included three variables for captains in the data set, though in Voyages user interface these are combined into a single variable. The ordering of these names indicates the order these men appeared,chronologically, to be associated with the voyage. For some British and French voyages, sources list different captains during the ship’s outfitting. A slave vessel may have cleared customs under the command of one captain but sailed to Africa under a subsequent captain. Evidence from the British trade suggests that for some voyages the first captain, rather than leaving the vessel, worked as a supercargo for the voyage. Therefore, we decided to keep the names and their ordering in the data set. Theuser will not be able to determine which captains were in charge of the vessels on the Middle Passage for all voyages. Some of the captains died before slaving on the coast; other captains’ listings include the man who commanded the vessels on the homeward passages from the Americas. We kept all abbreviations in captains’ names, consistent with the documentary evidence. From the Mettas-Daget catalog of French slave voyages, we attempted to maintain a consistent spelling of captains’ names as indicated in the index to the two-volume French set. Double surnames and indicators of rank (Sieur, Chevalier, de, de la) pose problems singular to the organization of the French subset. In short, the spelling of names is not fixed in the French language.We followed the spellings preferred in the index, though we transcribed first-name abbreviations as indicated in the documentary evidence. To facilitate sorting the Voyages Database’ file by captains’ names, we maintained the ordering of surnames as indicated in the published index.

Similarly, we followed, as closely as possible, the spelling and ordering of ship owners’ names given in the documentary evidence.The user will note some voyages "owned" by the RAC, Compagnie du Sénégal, or other monopoly trading groups. For these voyages, companies hired the vessels from ship owners and a group of partners or shareholders invested in the trading cargoes. The names of these individuals are not known. For most of the slave voyages in the data set, however, merchants owned fractional shares of the vessel and trading cargo.The listing of merchants in the set probably reflects the size of each shareholder, though this fact can be confirmed only for a few voyages. For some voyages we only know the principal owner "and Company." This is true particularly for many Bristol (England) voyages.To indicate the fact that the voyage was owned and/or organized by additional owners, we placed an asterisk, *, at the end of the last recorded merchant’s name, as in "Jones, Thomas*" (read: "Thomas Jones and Company"). For some other British voyages, father–son partnerships are listed, as in "Richard Farr, Sons and Company." For such voyages, we included the second owner with surname "Farr" as "Farr (Son)" and indicated that subsequent partners may be present by adding an asterisk after the third owner, "Farr (Son)*." Similarly, for the Dutch firm Jan Swart & Zoon (son), we entered the second owner as "Swart (Zoon)."

Ownership information contained in the French slave trade documents presents additional problems for the researcher. Unlike the British trade, in which many records of extended partnerships survive, French documents usually list single armateurs who organized slave voyages. According to Stein, an armateur was "the merchant who organized and usually financed a large part of the slaving expedition." (8) Other merchant-investors, therefore, are not recorded in the documents. In cases in which additional owners are suggested by the words "company" (Compagnie or Cie.) or "associates" (consorts), we inserted an asterisk. Many French slavevoyages were organized by family members. French documents include these familial relationships: brother(s) (frères), father (père),wife (épouse), widow (veuve or vve), eldest son (fils aîné), and son(s) (fils). These relationships are integral to the archival record and have been maintained in the Voyages Database. Because the French words frères and fils can imply multiple brothers and sons, we inserted an asterisk in the second ownership column, as in "Portier (Frères)*." In some cases, the document may record owners as "Brunaud Frères et Compagnie." For these few cases, we inserted a double asterisk as in "Brunaud (Frères)**." Some documents report the names of the propriéteurs who hired out their vessels to the armateurs, the affreteurs who freighted the slave ships, or the local agents who transacted business for absentee armateurs. We excluded these names from Voyages Database. French owners’ names often include complex double surnames and aristocratic titles. As in the case of French captains’ names, we attempted to preserve the spelling in the original documents while following the Mettas-Daget index to standardize the basic spelling and name ordering. We did this to allow the user to analyze ownership patterns easily through an A–Z owner–variable sort. The user should refer to the index of volume 2 of Mettas-Daget’s Répertoire for a complete listing of the variant spellings of French merchants’ names.

The common multiple Iberian names-- whether vessels, captains or owners—causes particular problems for researchers. Spanish and Portuguese names often incorporate the surnames of both the father and the mother. In the case of ship’s names, length stemmed from the habit of introducing multiple saints’ names and objects of religious veneration into the name of the vessel -at least before 1800. Length is a problem in this context because the official record of the vessel (and person), as well as the common usage of names by sailors and owners often recorded and employed only fragments of the full name, but unfortunately not always the same fragments. Users are consequently warned that it is often more difficult to track and eliminate double-counting of Iberian vessels and people than it is of their non-Iberian counterparts. Upward bias from double-counting is thus more likely in the records of the Iberian than of the non-Iberian slave trades.

There are seven imputed variables, indicated by the abbreviation "IMP" or "Imputed."  They draw first on data variables, but where hard data are unavailable, reasonable assumptions are often possible.  The seven are Flag IMP (National carrier), Slaximp, Slamimp, Year IMP, Modified tonnate, and place where a voyage began, and embarked or disembarked captives.  Their derivation is discussed more fully below.  To make the imputed variables as transparent as possible and to facilitate refinements, alternative assessments, and corrections that users might think necessary, we have included a download page which makes the database available to users in various formats. Indeed, it provides users with the data needed to create their own imputed variables.

The most straightforward of the imputed variables are geographic. The nearly one thousand locations in the data set where slave ships were built, registered, cleared for a slaving voyage, embarked or disembarked slaves are grouped into seven broad regions and 90 specific regions. A full listing of these places is contained in the SPSS code book, but space constraints prevent all these from being represented in the maps. We have adopted a convention that a geographical placename, such as London or Luanda or Charleston, needs to be mentioned five times in the historical sources in order to appear on the online maps. Those wishing to use alternative groupings of ports and places into regions and broad regions may use the geographic appendix to the SPSS code book. For Europe and the Americas, the groupings are self- explanatory. For Africa eight regional definitions are employed, following conventions in the literature. Senegambia is anywhere north of the Rio Nunez. Sierra Leone comprises the Rio Nunez to just west of Cape Mount inclusive.The Windward Coast is defined as Cape Mount up to and including the Assini River. The Gold Coast runs east of here up to and including the Volta River. The Bight of Benin covers the Rio Volta to Rio Nun, and the Bight of Biafra, east of the Nun to Cape Lopez inclusive.West-central Africa is defined as the rest of the western coast of the continent south of this point, and South-east Africa anywhere east of the Cape of Good Hope. A few locations mentioned in the documents, for example Casnasonis, and Touau-Toro cannot be identified,and one other major designation, a definition of the Windward Coast associated with no less than seventy voyages, straddles the definition of Windward Coast and Sierra Leone adopted here and is excluded from the African regional groupings.

On maps accompanying the “Search the database” interface, the default is “Broad Regions” at the lowest zoom level.  “Specific regions,” labelled in large font, are best viewed at mid-range zoom levels. “Port” locations are shown by black dots and labels at the highest zoom level.  Estimates maps have a similar structure except that no data are provided for ports, and broad and specific regions have definitions that differ slightly from those in the main database interface.

The variable “Place where voyage began” has an important imputed value added. We have assumed, if a ship brought slaves into a Brazilian port south of Amazonia, but left no record of where it began its voyage, that the voyage originated in the same place that it ended. The justification for this is the very strong bilateral nature of the Brazilian slave trade.  Users who do not wish to use this imputed variable can find the equivalent data variable in the downloadable version of the database.

There are also imputed variables for both voyage dates and on numbers of captives. Because most slaving voyages lasted for many months or even years, and no voyages have complete information for all ten date variables, we have created three definitions of "year" in the full downloadable database from which users can choose for purposes of analysis: the year in which the voyage originated, “YEARDEP,” the year of embarkation of slaves “YEARAF,” or the year of arrival at point of disembarkation, “YEARAM.” In the Voyages Database only “YEARAM” or “year of arrival” is provided. We created imputed year values when the sources did not record the years when voyages departed their homeport, or departed the African coast, or the year when vessels arrived in the Americas. If, for example, a London-based vessel arrived in Jamaica in September 1770, and this is the only documented date for the voyage, the year 1770 becomes the imputed African departure year, as well as the imputed departure year from the homeport. Further, years of arrival in the Americas are grouped into periods of five, twenty-five, fifty, and one hundred years. For the numbers of slaves carried and the numbers who perished during the voyage, as well as the age and gender categories, information is also frequently incomplete and additional imputed values are added, the creation of which is discussed more fully below. Researchers can of course make their own estimates and these, like the inferences on which alternative estimates are based, may well be different from what we regard as optimal. We would like to emphasize that in many cases the optimal solution is not obvious, and one researcher’s estimates (and inferences) may be different from, but as good as, another’s, despite the fact that all are working with the same data set. Anyone using the data, including ourselves, therefore needs to specify clearly the assumptions he or she is using.

One important problem slave-trade researchers need to address is whether vessels bound for “Africa” in the sources are slaving vessels. As late as the end of the seventeenth century, slaves formed less than half of trade by value between the Atlantic world and Africa. Many captains sailed to Africa to purchase gold, ivory, dyewoods, or spices., Numerous naval vessels, troop transports or store ships sailed from Lisbon to the Portuguese forts at Benguela or Luanda, and other European powers needed to supply their trading forts as well. For most of the French, Portuguese and Dutch voyages to Africa, researchers other than ourselves have made the decision on whether or not a ship was a slaver, though we have uncovered a few additional voyages from these nations where the object of the voyage remains unclear. It might be noted that records of ship departures have typically survived better in the historical record than records of ship arrivals.

For many British and Portuguese voyages, however, we have had to make some hard decisions in determining whether vessels were slavers or non-slavers. Many British and North American voyages returned to the port of origin after an interval of time during which a slave voyage could have taken place, but no information survives of the places of trade in Africa or the Americas. For most of these ships, clearance was for "Africa and the Americas" and many of the remainder in this group are ships leaving British American ports for Africa. Before the nineteenth century, ships rarely went from the Americas to Africa for anything but slaves. In all these cases the ship is assumed to have been a slaver. For several hundred more voyages from Brazil, even less is known. In Bahia, the main source of information on these is licenses of ship departures which specify "Elmina" (in West Africa) as the permitted destination. Large rolls of local tobacco were the trade good for African-bound Bahian ships, and in the eighteenth century slave traders of all nations depended on this tobacco. And as with North American Africa-bound ships, there is no evidence of a significant produce trade between Brazil and Africa. Gold was important in the first half of the eighteenth century and alcohol became more important later in the century, but return cargoes were always human. After 1680 many Portuguese voyages from Brazil show up in British records from Cape Coast Castle as well as the Dutch records for Elmina, and in the nineteenth century there is very good overlap between these licences and the observations of British observers on the movements of slave ships. We have made the decision to include these Bahia–Africa voyages in the data set. Nevertheless, we do have a file of 1,400 Atlantic voyages that might have carried slaves, but for which we are awaiting additional evidence. The majority of these certainly sailed from Europe to the Caribbean and then back again without sailing anywhere near Africa, but we cannot be absolutely certain and we retain information on them for future use as necessary. These "doubtfuls" are troublesome, but their numbers, compared to the voyages about which we are quite certain, are not great.

There remains the question of produce ships—defined as ships that sailed to Africa to trade for animal products, agricultural commodities or minerals. We have identified 1,450 voyages that departed Africa without obtaining slaves. In some cases they carried supplies for the European castles on the coast, but in the majority of instances they traded for African produce before returning directly to Europe. In addition, there were always a few "tenders" each year that went to the coast to supply slaves for a larger ship, but did not themselves carry slaves across the Atlantic. The great majority of these non-slaving ships were Dutch and British, the two nations that carried on the largest trades in African produce. We have identified produce (as opposed to slave) vessels sometimes on the basis of their voyage histories, sometimes on the known activities of their captains and sometimes on the basis of small crew-to-tonnage ratios, suggesting they were not vessels that required additional crew to control slaves. (9) Both the produce traders and the doubtful traders are held in a separate file and in the former case will be used as the basis for separate work on the African produce traffic.

Finally, not all voyages that crossed the Atlantic from Africa carried slaves. Generally we have assumed that all such voyages were slaving voyages, and have included them in the data set, though there is a slight possibility that a few of these vessels traded at produce markets on the coast. In summary, about 5 percent of the voyages included in the data set lack information about their activities after the voyages began. We nevertheless feel fairly confident that these were slaving voyages, and, as noted, those about which we feel less confident we retain in a separate file.

Obviously, not all slave ships made it to the Americas, or even to Africa. Fortunately the data set is quite rich in information on the outcome of voyages. The data set allows for 184 different voyage outcomes. As with the geographical variables, some regrouping is required to make these more manageable. The first regrouping, “Outcome of voyage for slaves” (FATE2), takes the standpoint of the Africans on board, and asks where the ship disembarked its slaves. The majority were disembarked in the Americas, but about 12 percent in the present sample died during the voyage. In addition, some who left African ports actually disembarked in another part of Africa or on the island of St. Helena (about two percent of all slaves recorded). Most of this latter group were captured by British naval cruisers in the nineteenth century, though a very few, in the previous century, ended up in Europe. A second regrouping, “Outcome of voyage if ship captured” (FATE3), is concerned with the fate of the ship and who might have interfered with its voyage. Slaves rebelled, shore-based Africans or pirates attacked ships, and one European power would often try to seize ships flying the flag of other powers, especially in wartime. Finally, a third regrouping, “Outcome of voyage for owner” (FATE4), takes the standpoint of the owners, and groups voyages on the basis of whether the ships reached the Americas, and if not, whether it was human agency or natural hazard that was responsible. As indicated, each of these three regroupings is represented by a different variable.

Establishing the outcome of the voyage is an important prerequisite to inferring information about both places of trade and numbers of people purchased. We have a good basis for imputing locations of slave trading as well as estimating the numbers of slaves embarked and disembarked. We turn first to the geography of the traffic. For some voyages we know the intended ports of trade on both the African coast and in the Americas. Private correspondence, newspaper reports, and official records of clearances from ports in Europe and the Americas frequently provide such information. Of the 36,000 voyages in the data set, at least 1,580 did not embark slaves, usually on account of capture or natural hazard. Of those that did obtain slaves, several hundred failed to complete the Middle Passage.

The data set provides some information on African place of trade for 21,700 voyages or about half of those that are likely to have obtained slaves. While this information surpasses current knowledge of the geography of the slave trade, it is possible to glean yet more. For 4,142 voyages that left Africa with slaves, or could have done so in the sense that the ship was not wrecked or captured prior to trade beginning, we may not know the African place of embarkation, but we do know where the captain intended to buy slaves. If we assume that he did in fact do what was intended, then after eliminating those locations that are not easy to group into regions (for example the French designation Côte d’Or, which ranged from the Windward Coast to the Bight of Biafra), we are left with 25,842 voyages that contain useful information on place of African trade—or about 60 percent of those vessels in our sample that actually did or could have left Africa with slaves. The imputed variable “MAJBYPTIMP” combines actual, with intended places of embarkation, and “MAJBYIMP” and “MAJBYIMP1” group these values into specific regions and broad regions of embarkation, respectively.

Switching to the other side of the Atlantic, the data set yields some information on ports of arrival for 25,589 voyages. Once more we have additional information on where 3,946 voyages intended to trade their slaves even though we cannot be certain that they actually did so. If we assume that captains completed the voyage according to plan, then the sample for places of disembarkation increases from 25,589 to 30,535 voyages, or close to 75 percent of all those ventures disembarking captives. The imputed variable “MJSLPTIMP” combines actual with intended places of disembarkation, and “MAJBYIMP” and MAJBYIMP1 group these values into specific regions and broad regions of arrival, respectively.

How valid are these assumptions on imputing places of trade from information on intended place of trade? Most slave ships traded in the regions where owners declared they would trade. After eliminating captured ships that rarely completed their voyages as intended, as well as those ships with very broadly defined destinations ("Americas" or "British North America"), a Pearson product moment correlation run on ports of arrival in the Americas generated a coefficient of 0.83 (n=9,541). A similar procedure for region of trade in Africa and intended region of trade produced a Pearson product moment correlation of 0.714 (n=13,951). Ii should also be kept in mind that merchandise always had to be loaded in Europe and the Americas for a specific African region and was often impossible to sell in another region. It was unusual to find a specific manufactured good selling in more than one region. (10) Taken together, this evidence appears sufficiently strong to allow some modest inferences for those voyages that we know purchased slaves in Africa, or subsequently disembarked slaves in other parts of the Atlantic world, and for which the intended, but not the actual, region of trade is known.

In addition to these inferential issues, there are also known biases in the geographic data. The British signed three treaties with the Portuguese between 1810 and 1817 that contained clauses limiting Portuguese slave traders to regions of Africa south of the equator, and the last two of the treaties allowed British cruisers to capture Portuguese ships that did not adhere to these provisions. Brazil assumed these treaties when that country became independent in 1822. From 1815, slave ships arriving in Bahia, which had strong trading relations with the Bight of Benin or Slave Coast (north of the equator), usually reported their African port of departure as Cabinda or Malembo, ports just north of the Congo. British officials in Bahia, as well as naval officers patrolling the African coast, were convinced that all Bahian ships nevertheless continued to trade on the Slave Coast. (11)

A second set of inferences is suggested by the data on numbers of slaves leaving Africa and arriving in the New (and in some cases, the Old) World. The sources provide the actual number on board at arrival for only 16,394 voyages. On the African side, the data are much weaker, with only 9.105 yielding information on the number of slaves leaving Africa, out of 30,622 voyages that left with slaves, and a further 3,711 that could have done so. Because most of those studying the slave trade are interested in the captives rather than the ships, some inference would seem appropriate for those ships that traded without leaving anything in the historical record about the slaves they carried. The first step in making reasonable inferences is to draw on the number of captives who might have been reported for the same itinerary at an earlier or later stage of its itinerary.

Only 6,683 voyages in the revised database contain numbers of both captives embarked and captives disembarked, but for a further 2,223 we have the figure for departures alone and for 12,667, numbers arrived alone. Imputed totals for the missing information may be made from the large subset of voyages that provide information on deaths during the passage. The Voyages Database contains 6,430 voyages for which a ratio of deaths to slaves embarked may be calculated. Deaths as a proportion of those embarked differed markedly by African region of embarkation. Table 2 shows breakdowns of shipboard mortality as a percentage of those slaves taken on board.

The breakdown of mortality ratios by African region is used here as the basis for imputing numbers arrived in the Americas where totals leaving Africa exist, and for numbers leaving Africa where the numbers on board at arrival in the Americas are known.

There remain 12,789 voyages with no information about the numbers of slaves on board ship. Indeed, we do not even know if some of these ships carried slaves. The current data set allows us to focus on the type of vessel as well as the route the voyage followed in forming estimates of captives transported where such information is missing. The number of captives on board was very much a function of the type of ship as well as place of trade in Africa, and to a lesser extent in the Americas. Moreover, the size of the type of vessel as reflected in its rig changed over time. The Appendix table attempts to take into account these factors by showing average number of captives both embarked and disembarked for 155 separate combinations of first, rig of vessel and time period; and second, where these were not available, place of trade in Africa; and third, a separate grouping of 18 types of vessels – smaller than those from the rest of the Atlantic World before 1800 - built in North America. A small group of vessels have no information on either rig or place of trade and estimates of captives for these are classed as “No rig” in the Appendix table. The means for these 155 categories were added to voyage records, as appropriate, whenever data on slaves could not be extracted from the sources. Users should note that in all cases these averages might be termed “running” in the sense that that as we added data to the database we recalculated the statistics reported in the Appendix table. And as augmentation of the dataset is set to continue then the imputed values will vary as data are added in the future. Users should thus not expect to find that the imputed values assigned to any given ship type are always the same or, indeed, will remain the same.

The above procedures generate imputed figures, where necessary, for the data used to produce series of slave departures and arrivals for eight different regions in Africa and sixty-four different regions of disembarkation. The variables “Total slaves embarked” (SLAXIMP) and “Total slaves disembarked” (SLAMIMP) incorporate these imputed data. “Total slaves embarked” is derived first from the data variable, “TSLAVESD." When values for this are not available, then a SLAXIMP value is derived from the sum of values available for “Slaves carried from first port of purchase” (NCAR13), “Slaves carried from second port of purchase” (NCAR15), and “Slaves carried from third port of purchase” (NCAR17)--in other words, the slaves embarked at up to three ports in Africa where these are available. If these data are incomplete or missing then SLAXIMP draws from the data variable “Number of slaves arriving at first place of landing” (SLAARRIV) adjusted for voyage mortality (imputed, if necessary), and finally, if this is not available then “Total slaves purchased” (TSLAVESP) – a variable available only in the downloadable database is used. Only if none of these data variables can provide adequate information, does SLAXIMP incorporate one of the imputed values in the Appendix table. Similarly SLAMIMP is derived first from SLAARRIV, then, if necessary, from the sum of “Number of slaves disembarked at first place of landing “ (SLAS32), “Number of slaves disembarked at second place of landing “ (SLAS36), and “Number of slaves disembarked at third place of landing “ (SLAS39) - in other words the slaves disembarked at up to three ports at the completion of the Middle Passage. All four of these are data variables. If these are blank, then SLAMIMP draws on TSLAVESD when available, adjusted for shipboard mortality (again, imputed, as necessary). As with SLAXIMP, SLAMIMP resorts to the imputed values in the Appendix table only when sources do not document the number of Africans who were disembarked.

Most users will probably prefer to use “SLAXIMP” and “SLAMIMP” in their analyses, but they should be aware that captive numbers are to be found in four different variables each at embarkation  - PLAC1TRA, PLAC2TRA, PLAC3TRA, MJBYPTIMP and disembarkation SLA1PORT, ADPSALE1, ADPSALE2, MJSLPTIMP. Eighty-seven percent of the slaving voyages documented at specific African ports embarked slaves at only one location. This bulk loading pattern was particularly pronounced in regions east of the Bight of Benin. The captains who traded at more than one African location usually did so along the coastline from Senegal to Ouidah (Whydah). Single-port slave-trading occurred even more often in the Americas. Ninety-five percent of the slaving voyages that anchored at specific New World markets discharged their human cargoes at one port. Nevertheless, users should note that information on partial cargoes and full cargoes are located in four different data variables, one containing the total arrived or departed and the other three reflecting purchases (or sales) in possibly three different places. The imputed variables “SLAXIMP” and “SLAMIMP” are based only on the principal place of trade in Africa or principal place of disembarkation. A more refined assessment of numbers of captives boarded or discharged requires a search of all eight variables (four for the embarked, and four for disembarked).

As noted above in description of data variables for “Age categories,” they do provide enough information to describe in gross terms the demographic structure of the trade—the age and gender composition of the Africans carried off as slaves. For 93 voyages, age and sex data is available for up to three places of embarkation and two places of landing of slaves, and for 437 voyages information exists on the age and gender of slaves who died on the crossing. When available, the quality of the data varies considerably. The most precise data is for 3,404 voyages with records that report the number of “men,” “women,” boys,” and ”girls,” as well as in some cases “children” and “infants.” For another 811 voyages, the number of “adults” and “children” are distinguished without indication of their gender; while 484 voyages reported the number of “males”and “females” they carried without specifying how many were adults and children.

Several age and gender variables were derived from the data variables. First, information from voyages with more than one place of embarkation and/or disembarkation of slaves was aggregated to produce one set of four variables – adults, children, males, and females – for places of departure and another for places of arrival. An additional set of the same variables was calculated for deaths on the Middle Passage. From these variables, we calculated two ratios: the proportion of children for all voyages with information on adults and children, and the proportion of males for all voyages with information on males and females. The proportion of adults and females are the reciprocals of the variables that were computed. The ratios are to be understood as the proportion of all slaves for whom a characteristic can be determined. The child ratio is the percentage of slaves identified as either children or adults, and the male ratio is the percentage of slaves identified as males or females. For the smaller number of voyages with information on both age and gender, we also calculated the ratios of men, women, boys, and girls. Ratios were derived only for voyages with at least 20 slaves whose age and/or gender is documented.

Each ratio was calculated for places of departure, arrival, and deaths; but to further simplify the information for inclusion in the set of variables available on the Voyages website, one set of ratios is provided for each voyage with age and sex information. It is the same as the proportion at arrival when that is documented; otherwise it is the proportion at departure. The age and sex of captives was recorded almost two times more often at places of landing (3,731 voyages) than it was at places of embarkation (1,970 voyages). Information on the demographic composition of slave cargoes at both embarkation and disembarkation exists for only 609 voyages.

Caution is required in inferring age and gender patterns from the ratios. Ships left the African coast with varying numbers of men, women and children on board. It makes little sense to combine, say, the Merced, taken into Sierra Leone with only one man slave on board, and the Alerta, which landed 69 men among 606 slaves disembarked in Havana in September 1818. The ratio of men in the first voyage was 100 percent, the ratio in the second case was 11 percent. Averaging without any further adjustment produces a ratio of men of 56 percent, which, given the different numbers of people on board, misrepresents historical reality. With large enough numbers of cases, this problem diminishes to the point of becoming negligible; but if users select a small number of cases, they should employ a simple weighting technique to correct for the differences in the number of people being counted. Thus, in the above example, the weighted average of men on the two ships is very much closer to the 11 percent on the Alerta than the 100 percent onthe Merced. Alternatively, users might disregard our voyage-based age and gender ratios and simply divide the total of males (or females) by the total number of slaves in the sample they select.

As the above discussion suggests, the ratios for age and sex made available in the Voyages Database are calculated without weighting. For example, “Percentage male” (MALRAT7) is computed by averaging the ratios computed for each voyage. Thus a mean of, say, 70 percent male for a group of years or a region is the unweighted average of male ratios for individual voyages in the selected group. If users wish to group all males in the selection (or all children) and divide by all slaves, they may obtain somewhat different results from those provided in the search interface, but they will have to first download the database to make that calculation.

Users should also remember that age and sex information was recorded on some vessels at the beginning of the voyage and on others at the end of the voyage. We have created composite male and child ratio variables, “Percentage male” (MALRAT7) and “Percentage children” (CHILRAT7) that lump together information from both ends depending on availability of data, and where information has survived on both we gave precedence to the ratios at the point of disembarkation. This procedure is justified by the finding that shipboard mortality was only modestly age and sex specific. Those users who wish to eliminate these modest effects should download that database first.(12)

We have also made some assumptions in order to project a fuller picture of national carriers in the trade. The set identifies the national affiliations of 26,391 or 75 percent of the voyages in the data set. The variable that carries this information is “Flag” (NATIONAL). For a further 7,364 ventures, the context of the voyage and the name of the ship owner or captain make inferences about place of registration possible, and thus we created an imputed variable of national affiliation that contains affiliations for 33,794 voyages. This is “Flag*” (NATINIMP). For some ships, no method enables one to impute the national registration. From 1839, the British allowed their cruisers to take slave ships flying the Portuguese flag into British Vice-Admiralty Courts for confiscation under British law. In response to this (and to similar legislation in 1845 that extended the provision to the Brazilian flag), many slave ships abandoned ship registration papers altogether. In addition, there are undoubtedly some voyages that registered in one country but that belonged to nationals of another, and others that sailed under false papers. Some British ships sailing under the French flag in the late eighteenth century are examples of the first; both British and United States owners sailing with Portuguese and Spanish papers after 1807—sometimes fraudulent, sometimes not—are examples of the second. Overall, these cases probably account for less than 1 percent of the ships included in the data set. It is also difficult to separate voyages made by ships owned in Britain from those owned in the British Americas and later in the United States. Some vessels identified as "British" were likely registered in the British Americas, and a similar problem of geographic specificity arises with the Portuguese and Brazilian ships in the nineteenth century. A frequency count of this imputed flag variable is nevertheless of interest and indicates that 12,014 voyages were British and that a further 2,261 were registered in the British Americas. Of the other principal nations, 11,367 are identified as Portuguese or Brazilian; 4,201 were French; 1,711 were Dutch; 1,840 were Spanish or Uruguayan; and 409 flew the flags of various Baltic states (mainly Danish), including Brandenburg-Prussia. Many of the voyages of unknown nationality were likely Portuguese, but even if they all were, Portuguese voyages are still somewhat underrepresented in the Voyages Database.

Perhaps the most difficult of the imputed variables developed for the Voyages Database is “Standardized tonnage”(TONMOD) derived from the data variable “Tonnage.” This may be the least reliable of the imputed variables provided in the database. It is offered here as a guide to ship size and to provide a crude basis for calculating indices of crowding on ships for both slaves and crew, as well as a basis for examining the efficiency with which the trade was conducted over time and between major ports and carriers. The reasons for the relative unreliability of this variable begin with the differences between deadweight tonnage, tons burden (for loose-packed cargo), and freight tons (for merchandise). But even within these types, jurisdictions often had their own methods of computation. (13) Sometimes more than one method was used simultaneously, and in most countries the various methods changed over time.In England, for example, the Royal African Company, the Naval Office shipping returns, the Royal Navy, and Lloyd’s Registers of Shipping all appear to have computed tons differently until 1786, when the measured ton became standard by Parliamentary statute. The formula was changed in 1836. It is possible to ignore some of these (the Royal Navy, which did occasionally carry slaves to the Americas, appears to have used the same formula as the RAC) and develop an equivalency for some others. But some jurisdictions introduced political bias because tonnage might be tied to subsidies or figures might be altered to circumvent the efforts of another branch of officialdom to control the numbers of slaves carried per ton.

The standard adopted here is the one established by the British--the largest of national traders when the slave trade was at its height. Beginning in 1773, British ships were required to use measured tons as well as registeredtons in their official documentation; from 1786, measured tons alone became the standard. (14) After 1807, slave ships were not usually of British origin, but reports of their activities originated from or weretransmitted through British channels. Much of the data were converted into British tons in the process. Reports from the British Vice Admiralty Courts almost invariably list British tonnage, and in 1840 the Foreign Office instructed its overseas "observers" to give tonnages as provided in the ships’ papers where possible, as well as in British tons. (15) Many tonnage data, however, are from non-British jurisdictions. Several independent contemporary observers suggested that the Portuguese (and Brazilian) ton was perhaps 5 percent smaller than its 1773–1835 British counterpart, and the Spanish ton 50 percent larger. (16) The differences between Portuguese and British tonnage for the nineteenth century, at least, seem small enough to disregard. A regression equation is estimated for converting Spanish into British tons that suggests that the former was perhaps two-thirds larger, with the difference varying somewhat by size of ship. United States tonnages are taken to be the so-called "Custom House Measure" of 1789, which was modeled on the British formula. Although some differences existed in the application of this rule among American ports, no adjustment is made here. (17)

For the period before 1786, a further regression equation is estimated for converting Royal African Company tonnages into the pre-1786 registered ton. (18) Also for this period,the Dutch ton, or last, is taken to be double the size of the British registered ton, and the French tonne aude mer is treated as equivalent to the British registered ton. (19) In addition to these adjustments, it is, of course, necessary to convert all pre-1786 tonnages to the standard British measured ton adopted for the set. Once more the British registered-to-measured conversion formulae are called into service.There remain several tonnages for British ships between 1714 and 1786, the provenance of which we are not certain.We have used registered tons wherever we could, but much of the data were collected by others and it is not always clear which tonnage measurement is used. We have made the assumption that such tonnages were the same as registered tons down to 1786. As noted above, ships could use either the registered or the new measured ton in their papers between 1773 and 1786, but slavers sailing before and after 1773 appear not to have changed their tonnage. Finally,it should be noted that there are almost no Spanish and Portuguese tonnage observations in the set before 1773 and very few Dutch tonnage records after 1786. No conversion has been attempted for Scandinavian, Hanseatic League (or Brandenburg-Prussian), Sardinian, or Mexican tonnages, values for which thus do not appear in the “Standardized tonnage” variable.

One last tonnage adjustment (not made here) is required for known bias. Tonnages of French slave ships between 1784 and 1792 were inflated (that is to say the size of the ton was deflated) substantially, as the French government based their subsidy of the slave trade on tonnages. (20) The size of the bias is unknown and tonnages of French ships in this period are simply ignored in the conversion procedure. A second bias (also not made here) is apparent in Portuguese tonnages between 1815 and 1830. A Portuguese law of 1684, and clauses in the 1815 and 1817 Anglo-Portuguese slave trade treaties, limited Portuguese and later Brazilian ships to a ratio of between 2.5 and 3.5 slaves per ton, depending on the construction of the ship. (21) Ratios were normally lower than this inevery branch of the trade for which data survive, and the regulation must have had little practical impact. As pressure to suppress the trade mounted in the nineteenth century and conditions on board deteriorated, it is possible that these strictures began to have some application. In any event, British officials in Brazil between 1815 and 1830 (after which the complete Brazilian trade was illegal and such regulations became moot) became convinced that the Portuguese tonnage measurements were being inflated by 60 percent on average so that more slaves could be confined on board. (22) The issue cannot be resolved on the available evidence and no adjustment is made here, but users have been warned.

Finally, we consider two variables that will attract more attention than most. First, we use the data variable “African resistance” (RESISTANCE) to compute a “Rate of Resistance” variable which is available only on the time line and in Custom Graphs. This is simply the number of vessels experiencing some recorded act of resistance divided by the total number of vessels in a given year and expressed as a percentage. The second is the “Sterling cash price (of slaves) in Jamaica” variable (JAMCASPR), which may be used to track the price paid for slaves in the Americas as they were sold from the vessel. The Voyages Database contains prices for those on board 957 voyages. The full derivation of these data is described elsewhere, but a summary description is appropriate here.(23) Prices for human beings in the Americas were subject to as many influences as were prices in any other market. Key factors included the characteristics of the person being sold, the distance between slave markets in the Americas and Africa and the price of the captive in Africa. This variable attempts to adjust for several of these factors so that the underlying price trends become apparent to the user of Voyages. In most cases the data are taken from the slave traders’ accounts and correspondence. Our first goal was to ensure that we recorded a single category of captive – what was frequently referred to at the time as “a prime male”.(24) Second, we adjusted that price for the price differential between the market in which the slave was actually sold and the price in Jamaica. Thus, if the captive was sold in one of the eastern Caribbean islands we would make a small adjustment upwards to reflect the ten extra days sailing time it would take to reach Jamaica. Third, we converted all prices into pounds sterling. What we did not do was to express the price in constant pounds (adjusted for inflation) – in other words, in real terms. This variable is thus based on archival data.

The above discussion is not exhaustive in the sense that we not have touched on and explained every single variable in either the Voyages interface or the two databases offered for downloading. Many of the variables need no more explanation than is available on the Variable List.

¹ David Eltis and David Richardson, eds., Extending the Frontiers: Essays on the New Transatlantic Slave Trade Database (New Haven: Yale University Press, 2008).

² Joseph Inikori has given a preferred global figure of 15.4 million for the European slave trade. Adjusting for those carried to the offshore islands and Europe, this implies 14.9 million headed for the Americas. See Cahiers d'Etudes africaines, 32 (1993): 686.

³ H.C.V. Leibbrandt, Precis of the Archives of the Cape of Good Hope: Journal 1662-1670 (Cape Town: Richards, Government Printers, 1901), 127-128.

⁴ British slave ships trading from Africa to Lisbon include the Kent (1731), the Mary (1737), and the Betsey and Hennie (1755). For sources see the data set. For the removal of slaves from Ambriz to St. Helena and Sierra Leone, see Kelly Muspratt to Aberdeen, 31 July 1843, British National Archives (henceforth BNA), FO84/501.

⁵ A separate discussion of tonnage is to be found below.

⁶ One frequently cited shipping list reports that there were no children on board several British slave voyages in the 1790s (House of Lords Record Office, House of Lords, Main Papers, 28 July 1800). This document, however, omitted to report the children embarked (cf. BNA, T70/1574; House of Lords Record Office, House of Lords, Main Papers, 14, 25 June 1799). In Luanda and Benguela Portuguese customs reports of departures for Brazil report very low numbers of children embarked. But in this instance, 'children' refers to infants only, and was above a tax category that indicated exemption from customs duties.

⁷ Mediterranean passes were issued by most European nations as a result of treaties with the Barbary powers. In theory, these documents allowed the vessels of the signing nation to pass freely through the Mediterranean waters frequented by Barbary corsairs. The passes record vessels and captain's names, tonnage, the date the pass was issued, and intended trading location, such as Africa or Africa and the Americas or Barbary or Madeira. See David Richardson, The Mediterranean Passes in the Public Records Office (East Ardsley, UK: EP Microform Ltd., 1981).

⁸ Robert Louis Stein, The French Slave Trade in the Eighteenth Century: An Old Regime Business (Madison: University of Wisconsin Press, 1979), xv. Some armateurs also may have owned the vessel. French dictionaries define armateurs firstly as those merchants who fit out the ship or expedition and secondly as (ship)owners. See Emile Littré, Dictionnaire de la Langue Française, vol. I (Paris: Hachette, 1881), 194.

⁹ One major aid in identifying produce vessels is the Seaman's Sixpence ledgers (BNA, ADM 68 series). One of our team, Jelmer Vos, went through this large series with great care.

¹⁰ One of the most widely used contemporary surveys of African regional preferences was Lt. Edward Bold, The Merchants and Mariners, African Guide (J. W. Norie and Co.: London, 1822). For a very detailed private record, see the manuscript in the Sidney Jones Library, University of Liverpool, Memorandum of African Trade, 1830-1840, for W.A. Maxwell and Co.

¹¹ See Pierre Verger, Trade Relations Between Bahia de Todos os Santos and the Bight of Benin, 17th to the 19th Century (Ibadan: Ibadan University Press, 1976), 358-361; and David Eltis, "The Export of Slaves from Africa, 1820-43," Journal of Economic History 37 (1977): 417-420, for a fuller discussion.

¹² In the downloadable version users should use the data variables Percentage male embarked (MALRAT1), Percentage male disembarked (MALRAT3), Percentage children embarked (CHILRAT1) and Percentage children disembarked (CHILRAT3) variables instead of the variables, Percentage male (MALRAT7) and Percentage children CHILRAT7.

¹³ For discussion of the general problem see Frederick C. Lane, "Tonnages, Medieval and Modern," Economic History Review, 17 (1964-5): 213-33.

¹⁴ The 1773 legislation is 13 Geo III, c. 74. See W. Salisbury, "Early Tonnage Measurements in England: I," H.M. Customs and Statutory Rules, Mariner's Mirror, 52 (1966): 329-40. To convert registered tons into measured tons, we used the formulae in Christopher J. French, "Eighteenth Century Shipping Tonnage Measurements," Journal of Economic History 33 (1973): 434-43. The 1786 act is 26 Geo III, c. 60, and its 1835 counterpart is 5 and 6 Will IV, c. 56, which introduced different rules for empty ships (s. 2) and those with cargo (s. 6). As the latter appears to have been used on slave ships, it is the one adopted here, and a further regression equation allows us to convert post-1836 tonnages into the measured ton of 1773-1835. It is:
Y = 52.86 + (1.22 x X) N = 63, R² = 0.77
where Y = measured tons, 1773-1835, and X = measured tons after 1835. For a critique of this approach see Peter M. Solar and Nicholas Duquette, "Ship Crowding and Slave Mortality: Missing Observations or Incorrect Measurements," Journal of Economic History, 77 (2017): 1177-1202.

¹⁵ Palmerston to Kennedy, May 4, 1840 (circular dispatch), BNA, FO84/312.

¹⁶ H. Chamberlain to Canning, 18 Sept. 1824 (enc.), FO84/31; W. Cole and H. W. Macaulay to Palmerston, 1 Jan. 1835 (enc.), BNA, FO84/169; W. W. Lewis and R. Docherty to Palmerston, 9 Sept. 1837 (enc.), BNA, FO84/214; J. Barrow to Aberdeen, 16 May 1842 (enc.), BNA, FO84/439; G. Jackson and F. Grigg to Aberdeen, 2 Jan. 1841 (enc.), BNA, FO84/350.

¹⁷ For Spanish into British tonnage, data are limited. The equation is:
Y = 71 + (0.86 x X) N = 32, R² = 0.66.
Where Y = British measured tons, 1773-1835, and X = Spanish tons.
For US and British, see An Act for Registering and Clearing Vessels, Regulating the Coasting Trade, and for other purposes, Statutes at Large of the United States of America, 1789-1873, 1 (1789): 55. For a discussion, see W. Salisbury, "Early Tonnage Measurements in England: IV," Rules Used by Shipwrights and Merchants, Mariner's Mirror 53 (1967): 260-64.

¹⁸ See David Eltis and David Richardson, "Productivity in the Transatlantic Slave Trade," Explorations in Economic History 32 (1995): 481, for the formula and a discussion.

¹⁹ See Lane, "Tonnages, Medieval and Modern," 217-233 for a discussion.

²⁰ Stein, French Slave Trade, 40-1; Patrick Villiers, "The Slave and Colonial Trade in France just before the Revolution," in Slavery and the Rise of the Atlantic System, ed. Barbara L. Solow (Cambridge, UK: Cambridge University Press, 1991), 228.

²¹ See Herbert S. Klein, The Middle Passage: Comparative Studies in the Atlantic Slave Trade (Princeton: Princeton University Press, 1978), 29-31.

²² See, for example, H. Chamberlain to Canning, 7 July 1824 (enc.), BNA, FO84/31.

²³ David Eltis and David Richardson, "Slave Prices of Newly Arrived Africans in the Americas, 1673-1807: A Quinquennial Series," in Historical Statistics of the United States, Susan Carter et al., eds. (Cambridge, UK: Cambridge University Press, 2006), 5: 690-691; idem, "Markets for Newly Arrived Slaves in the Americas, 1673-1864," in Slavery in the Development of the Americas, David Eltis, Frank Lewis, and Kenneth Sokoloff, eds. (Cambridge, UK: Cambridge University Press, 2004), 183-221.

²⁴ More specifically, the price presented here is the average of the first ten males sold off the vessel.