Economic inequality in Latin America and Africa, 1650 to 1950: Can a comparison of historical trajectories help to understand underdevelopment?

ABSTRACT The present article provides a comparative review of historical economic inequality in the two most unequal regions of the world, namely Latin America and Africa. This contribution examines novel studies that provide quantitative estimates of income and/or wealth inequality in the two continents in terms of sources, methods, results and interpretations, focusing on the period 1650 to 1950. The article shows that although scholars in the two regions have often employed similar methodologies, their results are far from conforming to a uniform pattern. The present review highlights how scholars of Latin America and Africa tend to remain geographically isolated, failing to capture the learning opportunities stemming from the work of their continental counterparts in terms of both sources and methods.


Introduction
Studies of inequality have seen a resurgence in recent decades, not least following a growing interest in the consequences of inequality on economic development. In this context, history has been deemed vital to the understanding of the present, a belief that has fuelled a new wave of studies on historical inequality. Previous studies on the subject were few and far between (see e.g. Gudmundson 1983;Bigsten 1986;Berry 1990), yet historical inequality is currently experiencing a much-needed revival (Federico 2021). Notwithstanding the prevalence of historical studies of inequality focusing on Europe and North America, the new wave of research has increasingly brought other regions into the spotlightparticularly in the last two decadesamong them Latin America and Africa.
It is not by chance that Latin America and Africa have received ever-increasing attention in terms of research on inequality. It is well-established that these two regions are the most unequal in today's world both from an economic and a social perspective (The World Bank 2016).

The present study
The widespread use of new archival material from both regions, in combination with the use of new and adapted estimation methods, has brought to light new empirical evidence on long-term inequality trends. These are presented in a large number of studies, primarily published within the past couple of decades. Despite the increased interest in historical inequality, this literature is yet to be systematically reviewed, leaving the question of what can be learnt from this recent strand of literature still unanswered. In the present contribution we review the current state of the research on long-term historical economic inequality in Latin America and Africa, from the 1650s to 1950, with the intention of answering the following research question: what can we learn about the long-term historical development of socio-economic inequality by employing a comparative perspective on Latin American and African societies? Our study shows that, despite the use of similar methods and types of sources, scholars have seldom taken advantage of opportunities for cross-fertilization stemming from geography. We find that the heterogeneity of the estimates makes it difficult to extrapolate patterns of long-term inequality between and within continents. We conclude that the reliance on individual case studies focusing on either of the two regions makes it difficult to solve the issue of comparability between estimates. More intentional comparative research in the future may thus contribute significantly to shedding light on common determinants of inequality in the two continents.

Delimitations
In the present article we focus on historical studies concerned with economic inequalities in wealth and/or income in Latin America (including the Caribbean) and Africa. We focus on studies that have contributed to the scholarly debate through quantitative, empirical estimates of aggregate wealth and income inequality. The latter is calculated in various ways, such as income in the formal sector or income that includes estimates for both formal and informal labour. Studies that focus on the distribution of a few specific wealth assets, i.e. land or slaves, will only be discussed in passing. The reason is that studies that focus solely on particular assets do not provide a comprehensive picture of economic inequality in a society. Furthermore, our contribution will be delimited to examining research that has provided estimates on inequality before 1950. The mid-twentieth century marks a turning point in the availability of official statistics, which has resulted in a plethora of studies on economic inequality in the two regions that stretch from this period until the present day. Such studies have been discussed in some previous literature reviews of the field, which motivates our temporal delimitation (see e.g. Cornia and Kiiski 2001;Anand and Segal 2008;Cornia 2014).
Due to length constraints, we will focus on studies that have examined inequality between individuals/households or between classes in a society. Studies of wage or income gaps (skilled/unskilled, female/male, urban/rural or between regions) or other aspects of inequality (e.g. human capital, anthropometrics) fall outside the scope of the present review. Although such studies help to shed light on specific aspects of socioeconomic inequality, they are based on very different methodologies than those discussed in the present contribution and would merit reviews of their own.
With the aim of reviewing the most relevant publications, we began by including several studies known to the authors. We later expanded our sample by undertaking a search of Google Scholar, using keywords such as wealth inequality, income inequality, long-term wealth/income inequality, historical inequality, inequality in Latin America/Africa, longterm wealth/income distribution, historical wealth/income distribution, and wealth/ income distribution in Latin America/Africa (in both English and Spanish). Further publications were then identified by using a snowballing technique from the studies found in the first round of identification. In order to avoid a bias towards a certain type of study, e.g. highly cited published work, we did not rule out studies that were unpublished working papers at the time of the writing of this review. In total, we identified in excess of 100 publications that at first glance seemed to fit the criteria of the present review. All these studies were separately examined by at least two of the three authors of this contribution, so as to minimize bias in the collection and reporting of information. At this stage some of the studies were deemed not to meet the criteria for inclusion in this review. In this way we arrived at a sample of 65 studies that met our criteria of providing novel quantitative empirical evidence of historical economic inequality in Latin America or Africa.

Measuring historical economic inequality
In Tables 1 and 2 we present key aspects of the studies examined in the present review, sorted by type of inequality (wealth or income), continent, and time period.
Overall, the tables show that Latin America has received substantially more attention than Africa, with 51 and 14 studies respectively. Historically, Africa suffers from a lack of Estimates both wealth and income inequality, b estimates the distribution of land wealth assets c estimates the distribution of land and livestock wealth assets d estimates the distribution of slaves wealth assets.  esearch prior to the twentieth century, a result of the scarcity of written sources for the period preceding the European colonization in the nineteenth century. The type of economic inequality studied in the literature is nearly equally divided between income and wealth. There are, however, relatively few studies that provide estimates on aggregate wealth inequality, as many of the studies concerned with wealth inequality only produce estimates for specific assets: slaves, land, or livestock. These provide a picture of inequality that is inherently limited and will therefore only be discussed briefly in this review.

Wealth inequality
Virtually all studies concerned with wealth inequality are based on various types of household-level data. Although the sources that enable such studies may vary, the most common sources we identify are tax records, census data and probate inventories. Census data have often been employed in research, from both Africa and Latin America (e.g. Gudmundson 1983;Gelman and Santilli 2010a;Bragoni 2011b;Gelman and Santilli 2011a;Koby 2014;Galli and Rönnbäck 2020).
While many might assume that a census only contains demographic information, some historical censuses also report valuable information on various assets owned by the individuals or households surveyed. One key strength of census-based estimates is that they tend to cover the entire population. This is, however, not always the case, as some censuses show a geographical bias (Galli and Rönnbäck 2020). Another type of source is tax records reporting data at the household level (Gelman and Santilli 2003;Fourie and von Fintel 2010;Parolo and Fandos 2011). A common challenge for both censuses and tax records is that they tend to include only certain types of assets, with taxable wealth depending on the type of taxation imposed.
Many of the assets included in studies that rely on either census or tax data seem to have a rural bias, as assets more commonly held by urban dwellerse.g. financial assets or capital invested in factories, workshops, or merchant inventoriesare seldom included in the primary sources. While some studies have therefore limited their scope to solely studying the distribution of wealth in rural areas (Gelman and Santilli 2003;Bragoni 2011b;Koby 2014;Galli and Rönnbäck 2020), others have attempted to adjust their estimates to compensate for this shortcoming (Fourie and von Fintel 2010). It is often also necessary to supplement data from censuses or tax records with data on the value of the assets they record. Where there is a lack of data on the value of these assets, wealth inequality can be estimated based on a Principal Component Analysis 1 (Fourie and von Fintel 2010).
An important methodological challenge for wealth inequality research based on census or tax data is how to deal with households that appear to have no wealth. This is a problem for many of the studies reviewed here (among others Gudmundson 1983;Parolo and Fandos 2009;Gelman and Santilli 2010a;Bragoni 2011b). Although it is very possible that a number of households in a society may not have any wealthand especially no wealth of the type recorded in the sourcesin some cases it can be difficult to tell from the source whether a household really had zero wealth, or if the records for some reason are incomplete (see e.g. Gudmundson 1983). Omitting the propertyless can, however, substantially bias the levels of inequality estimated for the whole society downwards. Due to their nature, certain sources, most importantly tax records, also tend to completely omit households that do not own any of the taxed assets, or that own wealth below a certain threshold. In order to draw any inference about the distribution of wealth in society at large, not just among property holders, information may have to be cross-checked against the size of the entire population, to establish how large a share of the population presumably held no assets (e.g. Parolo and Fandos 2011). Once the number of propertyless households is known, scholars have dealt in various ways with this share of the population, from assuming that the propertyless simply had zero wealth (Johnson and Frank 2006;Gelman and Santilli 2010a), to assuming that their wealth was below the threshold for tax liability (Gudmundson 1983).
A third common source for the study of historical wealth inequality is probate inventories. This research design has long been employed in studies concerned with wealth distribution in Europe or North America (see e.g. Main 1977;Jones 1980). While probate records have also been employed in the study of Latin America, so far only one ongoing project makes use of this type of source for Africa (Fourie and Green 2018). The main strength of probate inventories is that all assets held by the household under probate are included in the inventory. Nevertheless, previous research has shown that this is not always the case. On one hand, financial credits and debits are not always accounted for in the records, and on the other hand, some assets may have been removed prior to the probate being compiled. The main challenge of studies based on probate inventories, however, is how accurate a representation they provide. In most societies, wealthier households were more likely to be subject to probate than poor or propertyless households. Furthermore, in most slave societies, some reviewed here, slaves were legally banned from owning property and would therefore not be subject to probate. As probate inventories are established once a head of household has deceased, there is also an inherent age bias, as wealth tends to be accumulated over a lifetime. All the studies we reviewed that rely on probate inventories as a primary source employ comparatively small samples of households, at best a few hundred per point of observation (Burnard 2001;Frank 2005;Johnson and Frank 2006;Álvarez 2011;Moraes et al. 2021). This renders their degree of representativeness of inequality among the overall population uncertain at best.
A number of studies concerned with inequality focus on the distribution of specific types of assets, such as land or slaves. While some such studies are based on either of the types of source discussed above, and therefore face similar challenges to those already mentioned (Bergad 1999;Gelman and Santilli 2010a;Djenderedjian and Martirén 2012;Vicario 2017;Gelman and Santilli 2018;Galli and Rönnbäck 2021), several are based on other, asset-specific sourcescadastral surveys or particular farm/plantation censuses in the case of land inequality (Bergad 1999;Acemoglu et al. 2007;Djenderedjian and Schmit 2008;Summerhill 2010;Fandos 2013), or slave registers of various types in the case of slave ownership (Marcondes 2005). Although records such as these might be valuable per se, they only tend to include those members of a population who own such assets, therefore making it very difficult to draw inferences about the levels of inequality that include the propertyless.
A final issue that is worth noting is how to deal with slaves, as slavery remained legal and constituted a major form of capital investment well into the nineteenth century in several countries on both continents. That ownership of slaves was one important aspect of the wealth distribution in slave societies is obvious, and slaves are consequently also studied in several of the studies reviewed here for their contribution to wealth (see e.g. Bergad 1999;Marcondes 2005;Johnson and Frank 2006;Fourie and von Fintel 2010;Koby 2014;Moraes et al. 2021). Some of the studies have, however, also attempted to include slaves as members of the population, albeit propertyless, when estimating wealth inequality levels for the society at large (Johnson and Frank 2006;Fourie and von Fintel 2010).

Income inequality
Scholars studying income inequality have employed methodologies based on social tables, tabulated tax data or the Williamson ratio, with social tables being the most widely used methodology ( Table 2).
The idea of estimating inequality based on social tables is elegant in its simplicity, and has a long history (see Milanovic, Lindert, and Williamson 2011 for an overview of many such studies). Such studies most commonly start out from census data reporting information on the occupational structure of the society in question. The research design is thus highly dependent upon the existence and reliability of this type of data. Based on the data, the population is divided into classes of income earners, and to each member of a class is assigned a particular income. Although this procedure may seem straightforward, major challenges arise when estimating the income of individuals involved in agriculture, or that of slaves, both of which constituted a large share of the population in the period under examination in our areas of interest. A further methodological element worth noting concerns the number of classes a population is divided into, which affects the level of detail of a study. As evident in Table 2, some studies are very crude, based on a handful of classes, while some are substantially more detailed (e.g. Rodríguez Weber 2014; Burnard, Panza, and Williamson 2019; Tadei and Alfani 2019). A simple count of the number of classes may nevertheless hide the fact that a single class, i.e. small-scale farmers as in several of the African studies, dominates the sample numerically (e.g. Aboagye and Bolt 2021). Furthermore, most of the studies reviewed here rely on a basic social table design, which captures inequality between classes but not within, although some authors (e.g. Bigsten 1986) have attempted to make some assumptions concerning within-class inequality. It is noteworthy that within-class inequality can be high, as suggested by Fourie and von Fintel in their study of the Cape Colony (Fourie and von Fintel 2011). Studies that do not adjust for within-class inequality can therefore substantially underestimate the real levels of inequality in a society.
Other authors, among them Anthony Atkinson, Facundo Alvaredo, and their coauthors, estimate income inequality based on tabulated tax data (Alvaredo 2010;Alvaredo and Atkinson 2010;Atkinson 2014Atkinson , 2015aAtkinson , 2015bAtkinson , 2015cSouza 2016;Alvaredo, Cogneau, and Piketty 2021). This methodology is not devoid of issues, as reliance on tax registers obscures the issue of tax evasion. A perhaps more important problem is the minimum level of income required to pay taxes. In some of the countries studied by Atkinson and co-authors, this threshold level is indeed very high, so that the vast majority of the population goes unrecorded. In order to arrive at a meaningful estimate of income inequality, these types of studies require additional data on national income, i.e. from historical national accounts. Furthermore, this research design only permits study of the top income shares, rather than the whole distribution, as there is only reliable information for this share of the population.
A couple of recent Latin American studies have, finally, also attempted to estimate what has come to be known as the 'Williamson ratio', the ratio of median wages to average GDP per worker. This method, devised as an alternative to the wage/rental ratio that has figured in some previous research, is intended to overcome the paucity of data concerning land rents and therefore to examine income inequality levels in a larger pool of societies. Nevertheless, this methodology does not come without challenges. The main challenge lies in the assumptions that authors are required to make concerning which wage rate is most appropriate to use, leading to issues of comparability and reliability. Furthermore, the degree of representativeness of studies that rely on this method remains unclear, as wages tend to be available for the formal sector of the economy, which may account for only a small fraction of the labour force in agrarian societies, which applies to most of the studies reviewed here.
The availability of sources and the time-consuming nature of data compilation has limited the chronological scope of studies on inequality. Some studies merely report data for a single year, making it impossible to uncover chronological trends (e.g. Gudmundson 1983;Berry 1990;Burnard 2001;Gelman andSantilli 2003, 2010a;Bragoni 2011b;Koby 2014;Burnard, Panza, and Williamson 2019;Galli and Rönnbäck 2020;Moraes et al. 2021), while the vast majority of the studies in our review rely on estimates calculated over a few benchmark years. However, there are a few studies that have been able to create annual data series based on primary sources (e.g.

Empirical estimates of inequality
The studies reviewed in this article rely on a variety of inequality measures, from Gini coefficients to top income shares, to the Williamson ratio and more. The inequality measure of choice is an important element in studies of inequality as it influences the level of comparability across time and space.

Gini coefficient
The Gini coefficient is the measure more often employed by scholars due to its long and widespread use in research. 2 Figure 1 reports the available wealth inequality estimates provided in the form of Gini coefficients for countries in Latin America and Africa by decade. Only estimates pertaining to total wealth and to the whole population in each society studied appear in the figure.
The studies in our review span the period from the 1660s to the 1900s, with a concentration of estimates in the first half of the nineteenth century. The figure shows a predominance of studies on Latin America, although on closer scrutiny many of these estimates refer to regions in Brazil and Argentina, while only two studies refer to Africa, examining the Cape Colony and the Sierra Leone Colony. The range of the estimates spans from an exceptionally low Gini coefficient of 0.51 for the Sierra Leone Colony to an exceptionally high value of 0.95 for the province of Córdoba in Argentina. More notable, however, is the large difference in wealth inequality estimates among studies that focus on the same regions. Due to the use of studies that employ different methodologies or sources, those focusing on Rio de Janeiro and Buenos Aires in the mid-nineteenth century have concluded significantly different levels of inequality: a difference in the range of 0.15-0.2 Gini points (see Figure 1).
In Figure 2 we turn our attention to Gini estimates of income inequality. All estimates extracted from studies included in this review span from the 1700s to our cut-off point in 1950. 3 Estimates of income and wealth inequality therefore overlap for much of the period under study, with the exception of the years prior to 1700.
The range of income inequality estimates shows a wider range than for wealth inequality, with a low of 0.19 suggested for Mexico in the 1940s and 0.2 for Chile in the 1910s, and a high of 0.79, estimated for the Cape Colony in the 1700s (Fourie and von Fintel 2011;. The only studies dating from the eighteenth century, from the Cape Colony, colonial Jamaica and colonial Nueva España, suggest that income inequality in these three colonies was extremely high (Fourie and von Fintel 2011;Milanovic, Lindert, and Williamson 2011, 201;Burnard, Panza, and Williamson 2019). Unsurprisingly, inequality was the highest (in the range 0.74-0.79) in the two slave colonies, the Cape Colony and Jamaica (Fourie and von Fintel 2011;Burnard, Panza, and Williamson 2019). The estimated Gini coefficient of income inequality from Nueva España is somewhat lower (0.64), but the authors argue that it nonetheless exceeded the inequality possibility frontier of that particular colony (Milanovic, Lindert, and Williamson 2011).
The majority of the estimates of Gini coefficients in our sample date from the late nineteenth century and the pattern they reveal is rather heterogenous. Figure 2 highlights how current estimates of historical income inequality are 'fragmentary and not fully comparable', to borrow Branko Milanovic's words (Milanovic 2016, 4). Despite the large range of Gini estimates no clear trend can be discerned for Latin America and Africa as a whole. Differences in research design and varying geographical units of analysis make direct comparisons between studies problematic. Figure 3 shows, however, that even among studies concerned with a single territory and employing similar research methods, estimates may vary enormously in both levels and trends over time. Especially notable is the case of Brazil, where FitzGerald (2008), Bértola et al. (2010) and Astorga Junquera (2017) find a common trend in income inequality but a striking difference in the levels of Gini estimates (a difference of at least a factor of three) during the same   Table 2. ECONOMIC HISTORY OF DEVELOPING REGIONS years. The estimates by Astorga Junquera suggest that Brazil at this time would have been among the most equal societies in human history, whereas the estimates by Fitzgerland and Bértola et al. would place Brazil among the most unequal societies. The great disparity between estimates can also be seen for the other countries reported in Figure 3.
Despite the highly contradictory results in previous research, we will try to distinguish some commonalities. Starting with the studies from Latin America, it seems possible to distinguish a few different patterns.
(1) One pattern is an increase in inequality throughout the period for which data is available. This seems to be the case in Brazil (FitzGerald 2008;Bértola et al. 2010; and in Mexico (FitzGerald 2008; Astorga Junquera 2017; Castañeda Garza and Bengtsson 2020). The increase is very dramatic in the case of Mexico, in the range of 0.15-0.2 Gini points, whereas the estimates for Brazil increase more modestly, arguably due to a higher level of inequality from the outset. (2) A second pattern tends to hint at an increase in inequality during the nineteenth century, followed by a decrease in the twentieth century. This would seem to have been the case in Uruguay (Bértola 2005;Bértola et al. 2010) andin Argentina (FitzGerald 2008;Bértola et al. 2010).  Table 2.
(3) A potential third pattern would be the mirror image of the previous one: decreasing inequality during the nineteenth century and increasing inequality during the twentieth century. This may have been the case for Chile; however, estimates for Chile vary substantially in terms of both levels and trends depending on the study one focuses on (FitzGerald 2008;Bértola et al. 2010;Rodríguez Weber 2014;Llorca-Jaña et al. 2018a). The estimates of FitzGerald (2008) and Astorga Junquera (2017) suggest that income inequality in Chile from the 1900s to 1950 was rather low, hovering around 0.4 and 0.3 respectively, and stagnant, with only some minor variations. Rodríguez Weber (2014) estimates cover a longer time span and suggest a higher level of income inequality, albeit decreasing in the nineteenth century and increasing afterwards. Estimates from Llorca-Jaña et al. (2018a) and Bértola et al. (2010) are even higher and trending upwards, with inequality in early nineteenth century Chile being among the highest recorded in our sample. Rodríguez Weber's work would seem the most reliable as it builds on an extensive construction of a social table with over 50 classes. Competing estimates have also been calculated for Colombia (see Figure 3), where we find that the existing estimates hint at three different trends in inequality: a stationary, a growing, and a declining trend (FitzGerald 2008; Astorga Junquera 2017; Rodríguez Weber 2017a). Fitzgerald's and Astorga Junquera's data cover a slightly longer period, and exhibit no trend until the 1940s, when Fitzgerald's estimates converge towards Astorga Junquera's lower Gini estimates. Rodríguez Weber's data, although covering a shorter period, suggests increasing income inequality between the 1930s and 1950. The work by Rodríguez Weber is here largely based on previous estimates by Londoño (1991), 4 which seem to be more reliable than Fitzgerald's proxy estimates, upon which Astorga Junquera (2017) also relies heavily.
Turning then to the six African countries for which Gini coefficients are available, they exhibit one of three patterns (see Figure 4).
(1) The Botswana and Kenya estimates (Bigsten 1986;Bolt and Hillbom 2016) suggest increasing income inequality during the twentieth century. In Botswana, the estimated Gini coefficient increased from the 1920s, and in Kenya from the 1910s.
(2) The estimates from the Ivory Coast and Senegal (Tadei and Alfani 2019), in contrast, both suggest that income inequality in these countries decreased, at least from the 1930s onwards, the period for which they provide estimates.   Table 2.

Top incomes
The discussion above has been concerned with studies estimating inequality in the form of Gini coefficients. Do studies that employ other metrics tell a similar story? Studies on Africa, like the works by Facundo Alvaredo and others (see Table 2) have made important contributions with long series of estimates of top income shares based on tabulated tax data for a large number of African countries in the twentieth century. These studies show a generalized and pronounced decrease in top income shares during the late colonial period and after independence across most African countries studied (Alvaredo and Atkinson 2010;Atkinson 2014;Tadei and Alfani 2019;Alvaredo, Cogneau, and Piketty 2021). In the case of Nigeria and Ghana, however, top income shares rose slightly over the same time, leading to a convergence of top income shares between the African countries.
The Latin American studies vary in their time span, with some providing estimates for a single benchmark year (Koby 2014;Burnard, Panza, and Williamson 2019), others comparing a benchmark year from the period under review here with modern data (Berry 1990), 5 and yet others providing a cumulative picture over inequality in the continent (Cornia 2021). A few works, however, contribute to the debate with long-time series for three of the largest economies in South America in the twentieth century: Chile, Brazil, and Argentina (Alvaredo 2010;Souza 2016;Rodríguez Weber 2017b). In the cases of both Brazil and Argentina, the authors find that inequality as measured by top income shares grew rather dramatically following the Great Depression only to decline subsequently. In both countries, the share of income held by the top 1% reached a peak in 1942/1943, hovering at 31% in Brazil and 25% in Argentina, followed by a decline after which the top income shares stabilized at around 25 and 19% respectively (Alvaredo 2010;Souza 2016). Unlike the case of Argentina and Brazil, inequality fluctuated widely in the case of Chile, following a trendless pattern ranging from approximately 25 to 35% of total income between 1913 and 1940 (Rodríguez Weber 2017b). Like Brazil and Argentina, however, Chile also experienced a decline in inequality after the 1940s and the top income share fluctuated at approximately 15 to 20% from the 1950s onwards.

Williamson ratio
Studies that have built Williamson ratio estimates have so far exclusively focused on Latin America, likely due to the scarcity of reliable GDP estimates for Africa. Despite this geographical concentration, their results on inequality are not uniform. The research by Leticia Arroyo Abad (2013) studies Argentina, Uruguay, Venezuela, and Mexico in the nineteenth century, and finds distinct patterns of development in inequality for each country. For Argentina and Uruguay, the results suggest that inequality increased throughout the nineteenth century. In the case of Mexico, Abad's study suggests that inequality has been decreasing since the 1870s. The more recent study of Mexico by Bleynat, Challú, and Segal (2021), on the other hand, also employs the Williamson ratio, but suggests that income inequality actually increased throughout the period. These different results for nineteenth century Mexico are mainly due to methodological differences in the estimation of the Williamson ratios. Abad's study only accounts for land capital rents while Bleynat et al. account for returns on all factors of production. The study by Prados de la Escosura (2007), finally, has also contributed Williamson ratio estimates for Argentina, Chile, Colombia, Uruguay, Brazil, Mexico and Cuba since the 1850s. His results suggest that inequality increased until around the First World War in all these countries, except for Brazil. Following the First World War, Prados de la Escosura's estimates show decreasing inequality in several of the countries (Argentina, Chile, Cuba and Uruguay), whereas it increased drastically in the remaining countries (Brazil, Colombia and Mexico).

Wealth and income inequality compared
There are only two regionsthe Cape Colony and rural Buenos Airesfor which there are estimates of both wealth and income inequality from the same period. In the case of rural Buenos Aires, the estimates of income inequality (Bértola 2005) are considerably lower than the estimates of wealth inequality in this region (Gelman andSantilli 2011b, 2018). The respective estimates from the Cape Colony (Fourie and von Fintel 2010; do not, however, necessarily show that levels of wealth inequality were higher than levels of income inequality. This might potentially be due to biases in the samples of data used in the two respective studies. Taking all studies together, the estimated levels of income inequality in general seem to fall within a lower range than the levels of wealth inequality; the vast majority of estimates of income inequality show a Gini coefficient below the level of 0.6, in contrast to the studies of wealth inequality where the vast majority are to be found in the range above 0.7.

Concluding remarks
Socio-economic inequality in Latin America and Africa today is among the highest in the world. To what extent do these regions share common patterns in terms of historical levels of inequality? In this paper we offer a novel survey of the recent research on inequality in a historical perspective in the two regions.
Our review of the research in the field of historical economic inequality shows that scholars of Africa and Latin America so far have mainly operated in isolation from each other, as there are few references between the respective bodies of literature. We believe that much could be learnt from cross-fertilization. There are a number of studies from both continents that employ similar research designs, most importantly social tables, but also estimates based on specific types of primary data. In such cases, the details on how these methods have been employed and how scholars have attempted to deal with common challenges can undoubtedly lay the ground for improved studies in the future. Some types of research design are, however, solely employed by scholars studying one of the continents, but not as yet by scholars of the other. Estimating the Williamson ratio in a country over time has, for example, been employed by some Latin American scholars, but not yet in African economic history. This is an area where Africanists could potentially take inspiration from their Latin American colleagues. The same goes for the types of sources employed for the studies. Only one study of wealth inequality in Africa has, so far, attempted to use probate inventories as a source.
As for the results arrived at in the research reviewed in this paper, we find very little support for the argument that Latin American and African countries shared a uniform experience in terms of inequality levels and trends, neither between nor within each continent. Furthermore, these estimates span a range from very low to very high inequality levels, suggesting that the picture is far more complex than often argued in the historiography. The results of this survey thus caution against monocausal explanatory models of inequality levels in the two regions.
If the aim is to develop a more coherent, unitary, and comprehensive theory of inequality, a less fragmentary history is a prerequisite. Our study emphasizes the need for more thorough comparative work between scholars of different territories and periods. Only by creating a collaborative effort could we possibly come close to understanding how inequality grows and develops in developing countries, something that has so far escaped isolated efforts.