Wealth was highly concentrated in much of Western Europe by the eve of the Industrial Revolution. This was due to a gradual concentration of wealth that may stem back to the fourteenth century (Alfani Reference Alfani2015; Alfani and Ryckbosch Reference Alfani and Ryckbosch2016; Alfani and Ammannati Reference Alfani and Ammannati2017; Bengtsson et al. Reference Bengtsson, Missiaia, Olsson and Svensson2018; Alfani, Gierok, and Schaff Reference Alfani, Gierok and Schaff2022). Only the catastrophic shocks of the black death and world wars allowed wealth inequality to decline (Piketty, Postel-Vinay, and Rosenthal Reference Piketty, Postel-Vinay and Rosenthal2006; Roine and Waldenström Reference Roine and Waldenström2009; Alfani Reference Alfani2015; Saez and Zucman 2016; Scheidel Reference Scheidel2017; Alvaredo, Atkinson, and Morelli Reference Alvaredo, Atkinson and Morelli2018). Such evidence has given rise to a narrative of inevitable wealth concentration over the very long run in the absence of catastrophic shocks (Scheidel Reference Scheidel2017). However, the evidence is overwhelmingly from Western societies, so we cannot be sure whether these inequality trends are a Western or a universal phenomenon.

This paper uses new data from 584 villages to measure landownership inequality in pre-industrial rural Japan, 1640–1870. I focus on the distribution of lands because it was by far the most important form of wealth in the pre-industrial context. Surprisingly, land distribution was highly equal. The Gini coefficient averaged 0.57 and 85 percent of households owned some land. Japan was a society of landowning peasants. Further, I also find landownership inequality in Japan had no trend, so this was a stable equality.

In comparison, the most comparable village-level data from pre-industrial Italy, Germany, and England show substantially higher Gini coefficients of 0.7–0.9 (Alfani Reference Alfani2015; Alfani and Ammannati Reference Alfani and Ammannati2017; Kumon Reference Kumon2021; Alfani, Gierok, and Schaff Reference Alfani, Gierok and Schaff2022) and an upward trend in inequality over time. Although not as comparable, data from other Western European societies shows much greater inequality in land and other forms of wealth (Alfani and Ryckbosch Reference Alfani and Ryckbosch2016; Bengtsson et al. Reference Bengtsson, Missiaia, Olsson and Svensson2018), while evidence from China suggests similarly low landownership inequality (Buck Reference Buck1937). I further address data comparability concerns but conclude that such concerns are unlikely to explain differences in inequality. Overall, there is strong evidence of a divergence in inequality outcomes between England, Japan, Germany, and Italy. Further, there is suggestive evidence of a regional divergence in inequality. Western Europe converged toward societies of landless laborers since medieval times. In contrast, records from East Asia dating back to ancient times consistently show landowning peasant societies.

How did Japanese villages have such equal land distributions? I hypothesize that the key difference was in the institution of adoption, which was common in East Asia but not in Western Europe. Adoptions were used during this period as an heirship strategy when the household failed to biologically produce a son (Kurosu and Ochiai Reference Kurosu and Ochiai1995). Therefore, in the absence of a biological son, an adopted son could inherit the lands and keep the land in the family. An additional benefit was the reduction of landless heirs in Japan, where impartible inheritance (primogeniture), the inheritance of lands by one heir, was the norm. The surplus sons, who were not in line to inherit lands, were often adopted into other households. Thus, this reduced the number of landless households in the next generation. Together, adoption could lead Japan and other East Asian societies to converge toward an equal landownership equilibrium.

In contrast, Western Europeans had very limited options for securing heirs due to the lack of both polygyny and adoptions. This greatly increased household extinctions. Upon household extinction, these lands were redistributed to other male lines via wills or the marriage of surviving daughters (Habakkuk 1994). Such redistributions were highly unequal. Assortative mating meant wealthy heiresses married wealthy men, leading to greater concentrations of land, of which there are many examples (Clay Reference Clay1968). Thus, household extinctions were a channel that generated greater inequality.

How far can adoption explain differences between Japan and Western Europe? I study this using linked household landownership data across multiple generations, which allows me to look at how lands were transmitted across generations. If adoptions were important, they should have greatly reduced household extinctions among landowning households.

I begin by showing there were no large differences in features of inheritance beyond adoption within Japanese and Western European villages. First, the fertility patterns of the rural elite in Japan were similar to their English counterparts. Thus, without adoptions, the household extinction rates would have been similar. Second, the Japanese mostly practiced impartible inheritance (or primogeniture), which leads to unequal land distributions (Bartels, Jäger, and Obergruber Reference Bartels, Jäger and Obergruber2020; Wegge Reference Wegge2021). This institutional setting was similar to much of England but not to other European regions where partible inheritance also occurred. If anything, this feature would suggest more unequal land distribution in Japan.

I next show that adoption was motivated by the securing of heirship, which has lacked strong empirical grounding. Using the linked village data, I show that adoptions were much more likely when a household lacked a male biological heir. I use the sex of the first child as an instrumental variable to show that this was not driven by endogeneity. Therefore, adoptions were being used to reduce household extinctions, as hypothesized. This contrasts with the adoptions motivated by child welfare in Western societies today.

Finally, I show the effects of landownership on household extinction. I find that households with more than the average amount of land rarely went extinct despite having a 20 percent chance of having no male biological heir. This matches the patterns of adoption among rich households. In contrast, poorer households also had greatly reduced rates of household extinction due to adoptions but they did occasionally go extinct. This was presumably because they failed to attract an adoptee before death due to their limited amounts of land wealth.

Overall, adoption in Japan was functioning such that only 10 percent of lands needed to be redistributed due to household extinctions per century. This contrasts with data from England, where male lines went extinct at least 25 percent per generation, 1200–1800 (Russel 1948; Gobbi and Gõni Reference Gobbi and Gõni2021). This implies that 25 percent of lands were passed on and concentrated in other male lines per generation.

Adoptions made a clear difference in landownership inequality but why did some societies have adoptions while others did not? The historical evidence shows differences did not exist preceding the fourth century, when adoption was practiced across Eurasia. Studies from both ancient East Asia and ancient Greece/Rome show adoptions were used as a means of ensuring against the significant risk of not having a son. However, the church began preaching against adoptions in the fourth century (Goody Reference Goody1983; Gager Reference Gager2014). The institutional change was gradual but effective, and the use of adoption beyond the early Middle Ages became rare in most of Western Europe. This led household extinctions to play a major role in land distributions in Western Europe. This finding is consistent with the higher inequality in Western Europe relative to East Asia. This novel institutional mechanism is therefore a plausible partial explanation of the observed divergence in landownership inequality between East Asia and Western Europe.

A contribution of this paper is to show a long-run regional divergence in wealth inequality. The past literature had mostly focused on sporadically available estimates of inequality. In the case of Asia, this mostly came from the eighteenth century onward. Milanovic (Reference Milanovic2018) found cross-country evidence of pre-industrial income inequality being lower in Asian societies after the eighteenth century, and this is consistent with my findings.Footnote ¹ Scheidel (Reference Scheidel2017) attempted to look at a longer time scale with more fragmentary data but he concluded that all societies were converging toward high inequality in the absence of catastrophic shocks such as the world wars and the black death. In contrast, I show evidence for an alternative path of stable equality in Japan and perhaps China, which long predates the eighteenth century. This newly documented dimension of divergence in landownership inequality also adds to the well-known case of divergence in living standards across East Asia and Western Europe (Kumon Reference Kumon2022).

A second contribution is to show a novel mechanism that can explain differing wealth inequality outcomes across countries. I add to literature that has attempted to explain inequality through three approaches. At the macro level, studies have shown the importance of capital and its inheritance as a mechanism that generates inequality (Piketty Reference Piketty2011, Reference Piketty2014; Saez and Zucman 2016). At the micro level, studies using detailed micro-data have identified potential explanations such as differential returns on wealth (Bach, Clavet, and Sodini 2020), taxation (Jakobsen et al. Reference Jakobsen, Jakobsen, Kleven and Zucman2020), tax evasion (Alstadsæter, Johannesen, and Zucman Reference Alstadsæter, Johannesen and Zucman2019), genetics (Barth, Papageorge, and Thom Reference Barth, Papageorge and Thom2020), or inheritance (Boserup, Kopczuk, and Kreiner Reference Boserup, Kopczuk and Kreiner2016; Elinder, Erixson, and Waldenström Reference Elinder, Erixson and Waldenström2018; Adermon, Lindahl, and Waldenström Reference Adermon, Lindahl and Waldenström2018). Interestingly, Japanese peasants mostly practiced impartible inheritance, which increased wealth inequality in Germany (Bartels, Jäger, and Obergruber Reference Bartels, Jäger and Obergruber2020; Wegge Reference Wegge2021). Despite this, I show that inequality was low because adoption can significantly mitigate this mechanism.

An implication of this paper is that religious institutions may have impacted economic development through their effect on inequality. This is a novel channel in a longstanding literature linking religion to economic development since the seminal book by Weber (Reference Weber1930). The past literature had argued religion played a major role in economic development through its effect on human behavior (Becker and Woessmann Reference Becker and Woessmann2009; Schulz Reference Schulz2022; Henrich Reference Henrich2020), social organization (Greif and Tabellini Reference Greif and Tabellini2017), or its effect on resource allocation (Cantoni, Dittmar, and Yuchtman Reference Cantoni, Dittmar and Yuchtman2018). This paper suggests that church reforms also affected economic development through landownership inequality.

DATA

My main data source are the Japanese village population censuses (Shumon Ninbetsu Aratame Cho) from 584 villages with sporadic observations between 1634–1872 (Kumon 2024). The population censuses were annually compiled by all villages in Japan by order of the lords. The original motive was to use censuses to help enforce a ban on Christians by the Tokugawa shogunate. The censuses included the names, ages, household compositions, and a declaration of religion to identify Christians. Despite Christianity having almost disappeared in Japan by the eighteenth century, the surveys continued until 1870 by taking on new administrative roles. Many of these censuses began listing information on household landholdings, which was the main source of wealth at these times. Importantly, I also observe landless households, which are often not registered in tax registers as they are not taxable.

From the perspective of economists, landholdings can be interpreted as landownership because peasants had well-established rights to sell, rent, use, and inherit their lands. The lords owned land by law but in effect only taxed lands. Further, an institution separating the samurai from farmers (heinōbunri) meant the samurai class lived in urban areas and did not own lands. Instead, they earned a salary from the lords. This contrasts with the landed aristocracy of contemporary Europe. The samurai class, who earned wages, are therefore not included in the analysis. An analysis including the samurai requires studying income inequality, and a past study has shown the samurai were surprisingly poor; they earned only 20 percent more income than the average peasant (Saito Reference Saito2015), which is comparable to high-wage professions such as urban carpenters and tatami makers (Yamamura Reference Yamamura1974). The remainder of this paper will use the term landownership to refer to what has traditionally been labeled landholdings in the case of Japan and other European societies such as England.

I have collected population censuses from three sources. The first are data published in local histories, which were digitized.Footnote ² The second is the “Population and Family History Project” database at Reitaku University. The third is the online database of Hiroshi Kawaguchi entitled DANJURO. To focus on land inequality in an agricultural setting, I drop all observations from cities, post stations, and coastal villages where non-agricultural activities are common. This leaves 2,476 village-year observations from 584 villages, which I refer to as the Japanese inequality data.Footnote ³ There are unsurprisingly fewer observations for earlier years, due to survival bias with a dip in the 1870s when the censuses ended (see Figure 1b). I also observe 84 villages over the long run, defined as multiple observations spanning more than two decades. I use this long-run data to investigate time trends. Unfortunately, the data is highly sporadic, so villages can reappear in my sample after being missing for decades. For econometric purposes, this precludes the use of many time series techniques that require complete time series.

Figure 1 OBSERVATIONS ACROSS SPACE AND TIME

Source: Japanese inequality data.

The geographic breadth of the data is rich and representative of the main island of Honshu, with approximately 80 percent of the population (see Figure 1b). The topographic map (with white shade indicating higher elevations) shows how mountains dominate much of the landscape, amounting to approximately two-thirds of land area. Unsurprisingly, there are few observations from mountainous terrain, which only had small pockets of habitable areas. On the other hand, there are many observations in the plains where population was concentrated. The sampling for the islands of Kyushu and Shikoku in the southwest are poor and results from these areas must be interpreted with caution.

The village censuses contained information on landownership that were expressed in the value of the yield in units of koku (volume of rice grain equivalent) or mon (copper coins) and, in some rare cases, the land area. These values often came from cadastral surveys in the late sixteenth to early seventeenth centuries. These “official yields” were therefore outdated, so they failed to account for increased plot size or increased productivity. They also did not include landownership outside the village. Thus, there are sources of measurement error but how far can they affect a measure of inequality at the village level? Ideally, I want landownership to be in the value of land rent net of tax. The land rent net of tax in each year is a function of official yields, as in Equation (1).

(1) ${\rm{\;}}land{\rm{\;}}rent{\rm{\;}}net{\rm{\;}}of{\rm{\;}}ta{x_{i,t}} = yiel{d_{i,0}}\left( {\Delta \,pro{d_{i,t}} \times land{\rm{\;}}rental{\rm{\;}}rat{e_{i,t}} - tax{\rm{\;}}rat{e_{v,t}}} \right){\alpha _{i,t}}$

The left-hand side refers to the land rent net of tax, which is the economic value of owning the land. Yield is the value of the yield in period 0 (or the official yield) when yields were measured; Δprod _i,t is the change in productivity since the measurement of yield for field i in period t; and α _i,t captures other factors that cannot be controlled but affect land prices, such as yield risk. This would include any investments or depreciation on the plot that affects the value. The land rental rate is the implicit or explicit share of yield being awarded to the landowner in return for his rights. Finally, tax rate is what was paid by the landowner to the lord in proportion to the official yield.Footnote ⁴ As I am computing inequality measures that rely on landownership relative to total land owned, such as Gini coefficients, there is no problem if relative value is a function of the official yields multiplied by a constant or

$\frac{{land{\rm{\;}}rent{\rm{\;}}net{\rm{\;}}of{\rm{\;}}ta{x_{i,t}}}}{{total{\rm{\;}}land{\rm{\;}}rent{\rm{\;}}net{\rm{\;}}of{\rm{\;}}ta{x_v}}} = \frac{{yiel{d_{i,0}}(\left( {\Delta \,pro{d_{i,t}} \times land{\rm{\;}}rental{\rm{\;}}rat{e_{i,t}} - tax{\rm{\;}}rat{e_{v,t}}} \right)}}{{\mathop \sum \nolimits_{i = 1}^N yiel{d_{i,0}}(\left( {\Delta \,pro{d_{i,t}} \times land{\rm{\;}}rental{\rm{\;}}rat{e_{i,t}} - tax{\rm{\;}}rat{e_{v,t}}} \right)}}$

$$ = {\gamma _{v,t}} \times yiel{d_{i,0}}$$

where γ _v,t is constant within a village-year. This would hold if changes in productivity, land rental rates, and tax rates were uniform within the village. It is not possible to make the stronger assumption that γ _v,t is constant across villages in a given year (or γ _v,t = γ _t ) due to widely differing tax rates, which precludes inequality measures beyond the village level.

The weaker assumption that γ _v,t is constant within the village raises two concerns. First, did the land rent per official yield (Δprod _i,t × land rental rate _i,t ) vary across plots? Second, did tax rates per official yield vary across plots?

I can test these assumptions using land records from large landowners who recorded the land rent, land tax, and official yield of their plots. These records were made for the purposes of land rentals and land sales for which this information was required. I use records that were transcribed in Takeyasu (Reference Takeyasu1966) and Shoji (Reference Shoji1986), which come from the regions of the Kinai (centered around Osaka) and the island of Shikoku in the nineteenth century. Although the data is from a limited number of villages, the basic institutions were largely similar across Japan, so these can be considered as representative case studies. There are 64 records that have all variables, while another 63 have all data except the land tax.

Figure 2 shows both the land rent and land tax for plots of land owned by two landowners. It is immediately clear that the land tax was almost perfectly correlated with the official yield.Footnote ⁵ The few outliers are almost certainly due to these plots being located in another village. This shows the land tax was a fixed rate based on the official yield. Therefore, it is safe to assume the tax rate was constant within any village-year.

Figure 2 LAND RENTS AND LAND TAX RELATIVE TO OFFICIAL YIELDS

Sources: Takeyasu (Reference Takeyasu1966) and Shoji (Reference Shoji1986).

The land rental rates (gross of tax) relative to the official yield show more variation. When I include data from the other villages, the coefficient of variation of land rents relative to the official yield is 0.3 (see Online Appendix B.1).Footnote ⁶ Therefore, the official yield is a decent proxy of true land incomes (especially when we consider this issue is shared by modern wealth data). Perhaps more importantly, land rental rates do not vary strongly with plot size in these two villages nor the other villages (see Online Appendix B.1).

A related concern is that productivity differences existed across landowners. If large landowners had faster technological growth, this cannot be detected by the exercise noted previously that focuses on large landowners. However, when true land values have been compared to the outdated official yield across landowners, such correlations are not observed (Takeyasu Reference Takeyasu1966). There was little reason for productivity growth to be widely different within villages when available technologies were similar.

A final issue is that censuses only recorded lands within the village. This will tend to bias my inequality estimates downward because it was large landowners who were most likely to have holdings in other villages. However, landownership outside the village would have been small due to a system of law that gave less legal protection to land rights outside of the village of residence (Nakabayashi Reference Nakabayashi2013). I can also estimate the extent of land owned in other villages by looking at the proportion of land owned by non-residents in 47 villages where such data is available. The average is 15 percent, a small proportion of land. I later show this can cause a modest downward bias in my inequality estimates, which does not affect my conclusions.

In the following sections, I will compute the Gini coefficient, share landless, and share of land held by the bottom 20 percent, bottom 40 percent, top 20 percent, and top 10 percent, which are the standard inequality measures within the literature. They are all computed using standard formulae at the household level, the unit at which land was owned, inclusive of the landless. I focus on Gini coefficients when looking at time trends but this is to avoid repetition due to the high correlation of these measures (see Online Appendix B.3). The observations will be weighted by total households-village-decade. The village-decade weight gives equal weight to each village-decade so that I am able to capture long-run fluctuations. The total household weight gives higher weight to larger villages, although I show the results do not change when I weight each village equally.

INEQUALITY ESTIMATES

Time Trends

I first estimate time-trends in inequality using long-run data from 85 Japanese villages, 1647–1872. I estimate a linear time trend using village fixed effects.Footnote ⁷ As there is potential regional heterogeneity, I estimate it in aggregate and by region (as defined in Online Appendix A).

Figure 3 plots the data with the regression prediction by region. It shows there is much heterogeneity in inequality trends by village due to local phenomena. However, there is no clear trend in aggregate. This is confirmed in the regression as shown in Table 1, where I find a negative but insignificant coefficient. When splitting the sample by region, the central and northeast regions have marginal significance but with opposite trends.

Figure 3 RURAL WEALTH INEQUALITY DYNAMICS IN PRE-INDUSTRIAL JAPAN

Source: Japanese Inequality Data.

Table 1 LONG RUN TRENDS BY REGION

Notes: Standard errors are in parenthesis and clustered by village. *p < 0.1, **p < 0.05, ***p < 0.01.

Source: Japanese inequality data.

There are a number of concerns with this analysis. First, there may be non-linear trends in the data that are hidden when only looking for linear trends. In particular, the major famines of the 1780s and 1830s may have functioned like a “catastrophic shock” that reduced inequality and counteracted an overall increase. Although Figure 3 shows no sign of this, I formally test this in Online Appendix D.1. I show the famine years in the 1780s coincided with decreased inequality while the famine in the 1830s had no effect. However, this did not mask a general positive trend in other decades.

Second, villagers could be increasing landownership outside the village that is not included in the village census. I test this in Online Appendix D.2, where I proxy external holdings using the total land owned in each village. The total land owned is the sum of all within-village landownership by residents. The remainder of the lands were either owned by non-residents or abandoned due to depopulation or environmental reasons. Therefore, decreasing total landownership could reflect increasing external holdings in the region. However, I find no evidence for this.

For the purposes of comparison, I can conduct similar analysis in the case of 18 rural Italian villages, 1307–1809, which were collected from tax records and made available (Alfani Reference Alfani2015; Alfani and Ammannati Reference Alfani and Ammannati2017). I refer to the original articles for a detailed description of the data. For the purposes of measuring village-level wealth inequality, many aspects of the data are comparable. The Italian data is based on real estate at the household level. The inequality measures are also at the village level. However, the Italian data is inclusive of housing. The distribution of housing is unknown for Italy but a detailed study from eighteenth-century Spain by Nicolini and Ramos Palencia (Reference Nicolini and Ramos Palencia2016) showed non-land properties were more equally distributed than agricultural lands. Further, non-land assets were only 12 percent of the value of real estate. If these societies were similar, the real estate inequality measures for Italy would be a lower bound for land inequality, and most of the inequality would be generated by land. Another issue is that the propertyless are not included. This will decrease the measured inequality, although the available evidence suggests the bias is small (Alfani Reference Alfani2015; Alfani and Ammannati Reference Alfani and Ammannati2017; Alfani and Di Tullio Reference Alfani and Di Tullio2019).

The differences in wealth inequality trends between Japan and Italy are large (see Online Appendix G). All villages in Italy were trending upward after the black death. The rate of increase was also very high, with a 0.07 increase in Gini points per century. Italian villages clearly had very different dynamics to Japan.

Inequality Levels

I next estimate inequality levels for Japan as a whole. Although my observations are limited to a sample of village, there is much geographic breadth. In Online Appendix C, I use the data to show there is strong spatial auto-correlation in inequality up to 100 miles. Therefore, each observation contains information on its unobserved neighbors. These facts can be used to justify regional-level estimates.

One issue with this approach is that some regions lack any observations. In order to partially account for unobserved areas, I assume persistence in land ownership inequality between the Tokugawa period, 1647–1872, and the modern era, 1883–1895. This assumption is reasonable due to the slow-moving nature of wealth inequality combined with there being less than two decades between the two points of observation. To proxy landownership inequality in modern Japan, I use prefectural-level data on the share of farmlands under tenancy as a proxy for landownership inequality. This data was originally recorded in prefectural statistics and later compiled by Arimoto et al. (Reference Arimoto, Nakayama, Kimura and Nishimura1984). The share of land under tenancy records the area of plots farmed by tenants divided by the total area.

Given this assumption of persistence, I can “backwardly project” inequality in the Tokugawa period for regions lacking observation. The estimation is conducted in two steps. First, I estimate an OLS regression of inequality measures in the two periods. Second, I predict inequality in regions with few observations. I summarize the important parts of this process here with further details in Online Appendix B.

In the first step, I coded all village locations into modern prefectures and estimated the correlation in inequality between the modern and pre-industrial periods.Footnote ⁸ I use the share of land under tenancy in the 1880s as my measure of modern inequality. Unsurprisingly, I find a very strong correlation except for the share landless, for which I get the correct sign but no significance. In the second step, I estimate landownership inequality in Japan as a whole by combining my sample averages with the OLS predictions of inequality in prefectures with less than three observations.

The resulting estimates reveal a society of landowning peasants (see Table 2). The share of land owned by each class of peasant sheds light on the structure of landownership. The bottom 15 percent were landless, while the next 25 percent owned 7 percent of the land. These were small landowners who would need to additionally rent large amounts of land to make a living. The middle-class peasants, from the 5th to the 8th decile of landownership, owned 32 percent of the land and would have been less reliant on tenancy. Finally, the top 20 percent were the large landowning class, who owned 61 percent of the land. They could rent out their surplus lands to the lower classes within the village. Despite the existence of inequalities, it is surprising that such a large share of the population owned significant amounts of land.

Table 2 VILLAGE-LEVEL LANDOWNERSHIP INEQUALITY ESTIMATES FOR TOKUGAWA JAPAN

Notes: All prefectures are weighted by rural population in the 1870s in order to calculate overall inequality.

Source: Japanese inequality data.

The regional estimates show some regional variation. Gini coefficients varied from 0.5 to 0.69, with the most equal areas being the Kanto region surrounding Edo (current-day Tokyo) and the coastal prefectures to the west of it. In contrast, the most unequal areas were the Kinai region surrounding Osaka and Kyoto and the Hokuriku region. Importantly, the next section shows that even the most unequal regions in Japan were more equal than the typical Western European village by 1800.

There are a number of concerns with the methodology that are addressed in detail in Online Appendix B. First, the weighting could be changed to give equal weight to each village. However, re-estimating inequality using this method only changes the Gini coefficient to 0.59 and is not driving the results. Second, I could alternatively use only predicted inequality from the backward projection. Such an approach leads to a lower inequality estimate of 0.54. Third, I can also include coastal villages, where fishing was an important industry and agricultural lands were less important. However, I still find a Gini coefficient of 0.58. Fourth, the lack of observations of landownership outside the village may be causing a severe downward bias. I can estimate an extreme upper bound for inequality by assuming the richest person owned all of the land outside the village. Assuming such land amounted to 18 percent of within-village landownership values, as implied by the available data, the implied extreme upper bound Gini coefficient will only modestly increase to 0.64. Thus, the narrative of a landowning peasantry is robust to various concerns.

INTERNATIONAL COMPARISONS

Table 3 compares a cross-section of inequality across societies as they approached industrialization. Some caution is required in interpreting the results due to the differing data and methodologies across these studies.

Table 3 WEALTH INEQUALITY IN PRE-INDUSTRIAL SOCIETIES

Notes: + indicates propertyless are excluded. * indicates village-level estimates. Chinese estimates from the 1930s use figures for North China and South China to get a range of Gini coefficient. The proportion landless is from two different estimates for all of China in Reference CermanBuck (1937). English estimates are based on land areas rather than values. French estimates are based on tabulated data from Heywood (1981) as described in Online Appendix F. Northern Spain’s estimates are from Palencia, Northwest Italy’s estimates are from Piedmont, Northeast Italy’s estimates are from the Republic of Venice, and Central Italy’s estimates are from Tuscany.

Sources: Buck (1937), Soltow (1979, Reference Soltow1981), Heywood (Reference Heywood1981), Brandt and Sands (Reference Brandt and Sands1990), Kung, Wu, and Wu (2012), Alfani (2015), Nicolini and Ramos Palencia (Reference Nicolini and Ramos Palencia2016), Alfani and Ammannati (2017), Bengtsson et al. (Reference Bengtsson, Missiaia, Olsson and Svensson2018), Alfani and Di Tullio (2019), and Kumon (2021).

The most comparable data come from Germany, Italy, Japan, and England, which are measured at the village-level and wealth is in the form of real estate or lands. Like the case of Japan, these records do not include real estate outside of the village. They also do not include the state, church, or feudal properties, much like the exclusion of the samurai class in Japan. However, there are a number of major differences. First, the German and Italian estimates include non-land properties. However, as stated earlier, Nicolini and Ramos Palencia (Reference Nicolini and Ramos Palencia2016) showed that lands were 88 percent of the value of real estate in eighteenth-century Spain. Further, lands were more unequally distributed than non-land properties. If these societies were similar to Spain, the bias due to this difference is downward. Second, the measures for Germany and Italy do not include propertyless people (indicated by a “+” on the table), leading to a downward bias, although indicators from these studies suggest it is small. Third, the measures for England are from commons that got enclosed via parliamentary enclosure (Kumon Reference Kumon2021). As the commons were relatively equal, this is a significant underestimation of inequality. Fourth, the unit of measurement in England is adult males instead of households. The direction of this bias is unclear, although most adult men would have had their own household and biases are likely to be minor. Finally, the large landowners whose holdings spanned many villages appear as relatively small landowners in village-level inequality estimates. This was a small issue in Japan, where cross-village landownership was limited as explained earlier but may lead to greater downward biases in Europe, where there were fewer limitations on cross-village landownership. The overall impression is that the European village-level estimates are downwardly biased.

The measures from all other societies were made at the national level. As the national-level inequality also accounts for between-village inequality, it will upwardly bias inequality relative to my measure, although the magnitude remains unclear. However, the percentage of landless is an alternative measure of inequality that does not account for differences in levels of inequality. Therefore, it is robust to this concern. Therefore, the share landless (where available) is the key comparison measure for these societies.

With these limitations in mind, the evidence suggests East Asian societies were more equal than those in Western Europe.Footnote ⁹ Village-level Gini coefficients in Western Europe are generally close to 0.8, while that in Japan is closer to 0.6. Only Germany has comparable inequality, but this is explained by the catastrophic shocks that decreased its inequality, as I will explain later. As shown earlier, an upper bound estimate of Japanese village-level inequality would suggest a Gini coefficient of 0.64, which is still far lower than the Gini coefficient of most village-level inequality estimates in Western Europe. Moreover, the other potential biases mostly work against these conclusions. The alternative measure of the share landless, which can be compared with national-level estimates, show East Asia had very low shares of landless (mostly below 20 percent), unlike in Europe, where it is generally above 50 percent. These high shares of landless in itself indicate that Gini coefficients must also have been much higher in these regions compared to Japan.Footnote ¹⁰

Another concern is that these results are driven by the timing of observations. Western Europe was about to start an Industrial Revolution, and the underlying factors that created modern growth may have also increased inequality. However, the trends in inequality in Italy and Germany were upward long before industrialization (see Figure 4). The only exception is the catastrophic shocks of the black death and, in the case of Germany, the Thirty Years War, 1618–1648, which temporarily reduced wealth inequality. Although there is no time series for England, Campbell (Reference Campbell2008) suggests 47 percent of the rural population were landless laborers in 1290, and other data from the hundred rolls in 1280 suggests land ownership inequality among peasants had a Gini coefficient of roughly 0.75.Footnote ¹¹ Therefore, increasing wealth inequality had little to do with industrialization.

Figure 4 INTERNATIONAL TRENDS IN RURAL WEALTH INEQUALITY

Sources: Japanese inequality data, Alfani (Reference Alfani2015), Bengtsson et al. (Reference Bengtsson, Missiaia, Olsson and Svensson2018), Alfani and Di Tullio (Reference Alfani and Di Tullio2019), Alfani, Gierok, and Schaff (Reference Alfani, Gierok and Schaff2022), and Kumon (Reference Kumon2021).

In contrast, the more patchy data from Japan and China suggest my observations are not from a time when society was abnormally equal. Rather, these societies were equal throughout the records stretching millennia. China introduced the equal fields system in 485 that was continued up to 780. Each man aged 15–59 was theoretically allocated 100 mu of land, although the reality was less equality due to land scarcity (Mitani Reference Mitani, Otsu, Sakurai, Fujii, Yoshida and Ri2015). This was also inherited by Japan via the Handen system from the seventh–tenth centuries (Iyanaga Reference Iyanaga1980). Although we know little of the context of these policies, it is likely that these policies accepted realities of relatively equal landownership rather than being a radical redistribution imposed by the state (for more detail, see Online Appendix H).

My findings also match wider findings from the literature. Lundh and Kurosu (2014) compare landownership distributions within villages in seven rural areas across pre-industrial Eurasia and find similar patterns. Another approach by Milanovic, Lindert, and Williamson (Reference Milanovic, Lindert and Williamson2010) uses social tables to estimate income inequality across classes. While this captures other aspects of inequality, it should be highly correlated with wealth inequality because wage inequality was relatively low. Their study finds similar regional patterns in income inequality. Importantly, Japan was relatively more equal in income inequality despite including the samurai class (Saito Reference Saito2015). Thus, these results are not an artifact of the differing social structures.

Overall, the large magnitude of difference across societies lends confidence to East Asia being more equal than the West despite the differences in measures. However, we cannot be certain of the exact magnitude of difference. What we do know is that the share of landless households, where available, show East Asian households commonly owned lands, unlike their Western European counterparts. These measures also match the wider historiography of the rise of the landless class in pre-industrial England (Shaw-Taylor Reference Shaw-Taylor2001), and Holland (Van Bavel 2006). In contrast, the East Asian literature has often focused on the landowning peasant (Huang 1990). The new puzzle that emerges is why landownership patterns differed so markedly across these two regions.

EXPLAINING INEQUALITY

How can we explain the low landownership inequality in Japan relative to Western Europe? The literature on pre-industrial inequality has surprisingly few explanations. A cross-country descriptive study by Milanovic (Reference Milanovic2018) shows areas with higher population densities had lower income inequality. One prominent hypothesis, which lacks empirical evidence, is the effect of struggles between the state and peasants over land rights in Europe. The state wanted to maximize taxes while peasants resisted such efforts. Although the reasons remain unclear, serfdom emerged in the East and free peasants in the West (Brenner Reference Brenner1976). Episodes of struggles include the parliamentary enclosures (Marx Reference Marx1867; Humphries Reference Humphries1990) and the Danish agrarian reforms (Boberg-Fazlić et al. 2022), among others.Footnote ¹² These institutions changed existing land rights where many individuals held rights over a single plots of land (often known as commons) to private ownership by individuals. This also allowed for plots of land to be sold, which made land accumulation easier.Footnote ¹³ In contrast, the argument is that Eastern European peasants were repressed and lacked land rights, although there is little empirical evidence.

Where does Japan fit in terms of the struggle between the state and peasants? Despite a feudal structure, peasants had secure de-facto landownership rights that were more similar to Western Europe. Unlike in Western Europe, this was due to the state indirectly pushing for clearly defined land rights since the Taikō Kenchi in the sixteenth century. This allowed for land accumulation among peasants. However, Nakabayashi (Reference Nakabayashi2013) shows evidence that land markets were less functional in Japan due to limited land rights for holdings outside one’s village of residence. This reduced the amount of land accumulation that extended beyond the home village. Therefore, part of the story of lower inequality could be attributed to the differing arrangements of landownership rights in Feudal Japan.

Other important hypotheses include the effects of regressive taxation (Alfani and Di Tullio Reference Alfani and Di Tullio2019), the development of mortgage markets (Allen 2006), agricultural productivity (Clark Reference Clark1998), and inheritance institutions. Regarding the last mechanism, Wegge (Reference Wegge2021) shows areas with partible inheritance in Germany had lower inequality, although this explanation has limitations due to the endogeneity of such institutions (Hrdy and Judge Reference Hrdy and Judge1993; Huning and Wahl Reference Huning and Wahl2021), as can be seen in East Asia where such institutions depended on land abundance (Hirai Reference Hirai2003; Zhu, Son, and Seo Reference Zhu, Son and Seo2015). Further, this fails to explain the case of Japan, where impartible inheritance was common (Hayami Reference Hayami1983).

Adoption and Inequality

This paper proposes an additional mechanism based on the use of adoption as an heirship strategy. Adoption is where a person becomes the legal parent of an adoptee and the adoptee gains the rights associated with being a biological child. Adoption affects land distributions because it gives the adoptee the rights over wealth inheritance. Unlike in the modern era, where adoption is often about the welfare of the adoptee, most adoptions in the pre-industrial era were about the continuation of male lines. Importantly, adoption was widespread in East Asia but not in Western Europe, so this institution may have contributed to differences across these regions.

Adoption was useful in the pre-industrial world, where child mortality was high. Assuming similar fertility across societies, some households succeeded in producing one or many male heirs while others did not. In Figure 5, I illustrate this with a case of five households, two with two heirs, two with no heirs, and one with one heir. In Japan, a household with more than one male heir could give them up to a household in need of male heirs via adoption (see Figure 5c). This benefits the adoptee, who will own more land than if he remained in his parent’s household. The overall effect is that lands were kept in the family, resulting in stability across generations.

Figure 5 LAND DISTRIBUTION ACROSS TWO GENERATIONS

Notes: Households a and b refer to new households formed by two sons from the same household, excluding the case of adoptions.

Source: Author’s illustration.

In contrast, Western European households very rarely adopted children. When they failed to produce male heirs, the household went extinct. Thereafter, marriage and inheritance norms redistributed the lands of the extinct. If a daughter survived, she would inherit the lands and usually marry into another household, leading to greater concentration of lands. If no children survived, the lands will be redistributed via will. Most notably, rich heiresses tended to marry rich men due to positive assortative mating at this time (Kurosu and Ochiai Reference Kurosu and Ochiai1995; De La Croix, Schneider, and Weisdorf Reference De La Croix, Schneider and Weisdorf2019; Clark and Cummins Reference Clark and Cummins2022), leading to land concentration.

There were further problems due to the lack of an option to make surplus heirs an adoptee. If the society had impartible inheritance (primogeniture), the surplus son would become landless, which further increased inequality (see Figure 5b). If the society had partible inheritance, there is some inequality generated by the randomness of fertility and mortality across households (see Figure 5d). These factors can lead to inequality beyond the effects of marriages and wills.

One important extension beyond the simple cases is to allow for differing inequality across households. In this case, adoption need not be practiced among all households. In fact, the extinction of landless households will reduce landownership inequality. Moreover, it is more important that the rich practice adoption because their extinction leads to greater land concentration than the extinction of a poorer household.

Adoption was not the only heirship strategy available at the time. One could marry early and have high fertility within marriage. Another strategy was polygyny, which could further increase births while also hedging against the risk that the wife was infertile (Kumon and Saleh Reference Kumon and Saleh2023). However, both strategies were unreliable due to the high variance in fertility and mortality. Moreover, adoption was special due to two factors. First, it could fully ensure against the risks of varying fertility and high child mortality. Second, other strategies had the opposite risk of creating too many children, which was expensive. Thus, institutions like polygyny were restricted to the richest in pre-industrial Japan but did not occur in peasant households. In contrast, adoption was used often among landowning peasants.

One concern with the mechanism is that adoption may have developed endogenously with inequality. Surprisingly, adoption was widely practiced across Eurasia during ancient times. In East Asia, the practice began by the Han period in China, 206 BCE–220 CE, the Nara period in Japan, 710–794, and the early Chosun dynasty in Korea, 1392–1910, as seen by genealogies or law codes (Hayashi Reference Hayashi, Otake, Takeda and Hasegawa1988; Brown and De Crespigny Reference Brown and De Crespigny2009; Peterson Reference Peterson1996). The institution of adoption continued to be widely practiced into the eighteenth century. In particular, adoption is well studied for the elite class, and adoption rates were as high as 8 percent in China, 1750–1849, 21 percent in Korea, 1750–1849, and 37 percent in Japan, 1700–1799 (Moore Reference Moore1970; Feng and Lee Reference Feng and Lee1998; Kim and Park Reference Kim and Park2010).Footnote ¹⁴ Importantly, adoption was clearly motivated by the desire to secure an heir in these societies (Feng and Lee Reference Feng and Lee1998; Kim and Park Reference Kim and Park2010).Footnote ¹⁵

Beyond East Asia, adoption was also common in the ancient Near East and Mediterranean, such as in Babylonia, Middle Assyria, Greek and Roman Egypt, Greece, and Rome (Huebner Reference Huebner2013a).Footnote ¹⁶ In fact, the term “adoption” originated in ancient Rome (Goody Reference Goody1969; Corbier Reference Corbier and Rawson1991).Footnote ¹⁷ The most convincing evidence for its widespread use in ancient Europe comes from a Roman census list of 1450 individuals. For those over age 50, almost all households had a male heir, which could only have been possible with adoption in an ancient mortality regime (Huebner Reference Huebner2013b). This also confirms the people of ancient Rome were using adoption as a means of securing heirs. Such practices survived into the Middle Ages, where Germanic and Frankish people were known to have had various rituals for adoptions (Lynch Reference Lynch2019).

Unlike East Asia, Western Europe began its general abandonment of adoption in the fourth century when the church made concerted efforts to discourage the institution (Goody Reference Goody1983; Gager Reference Gager2014). The institutional change was gradual but effective, and the use of adoption beyond the early Middle Ages became rare in most regions.Footnote ¹⁸ It was only in the nineteenth century that laws began to accept adoptions in Western Europe (Mignot Reference Mignot2019). Thus, East Asia and Western Europe had very similar adoption institutions until a policy shock led to differences emerging by the medieval period.

Why did Western Europe abandon adoptions that were a common feature of many agricultural societies? One explanation is theological. Contemporaries argued that adoption can overshadow the “divine adoption” of becoming God’s child via baptism (Gager Reference Gager2014). The church also discouraged emphasis on earthly concerns, such as by using adoption to determine wealth transmission after death. Despite the Bible including a few cases of adoption, such as that of Moses, the idea was that adoption motivated by wealth inheritance was wrong.

Alternatively, Goody (Reference Goody1983) argues that the change was motivated by the financial benefit of the church. The shift in policy happened after laws changed allowing the church to own property from the fourth century. This encouraged the church to increase its properties by accepting “God’s share” of bequests from childless families who willed it to them.Footnote ¹⁹ Interestingly, both of the major explanations show there was little awareness of the potential consequences for inequality. Further, the financial incentives for the church may have led to the persistence of this policy. The consequence of this institutional change in Europe was the widespread use of marriage among heiresses and wills to redistribute the land of the extinct, which gradually generated greater inequality.

EMPIRICAL EVIDENCE: ADOPTION AND INEQUALITY

I use evidence from three Japanese villages where the annual censuses have continuously survived, allowing for the construction of annual panel data. The data includes information on landownership, household composition, the relationship of each member to the household head, and the names and ages of all individuals over many generations. The use of annual observations is important for identifying adoption. Household members are recorded as adopted when they enter the household but lose this distinction if they become the household head, as they often did. The annual data also allows me to observe households that go extinct due to the lack of heirs.

Due to the need for detailed data, this is a limited sample. Two villages, Ishibushi village, with observation from 1752–1812, and Tonosu village, with observation from 1790–1859, are from the current region of Fukushima in northeast Japan, while Hanakuma village, with observation from 1789–1869, is from the current region of Hyōgo. The size of the villages was close to average with the exception of Ishibushi, which was slightly smaller. They were also relatively equal, with Gini coefficients ranging from 0.39–0.47, but these factors should not affect adoption behavior. While this is just three villages within Japan, adoption occurred throughout Japan at this time (Ōnuma Reference Ōnuma2018). If anything, the adoption rates were lower in these three villages than other villages, so I may underestimate the effects (Ōto Reference Ōto1996; Okada Reference Okada2006) (for further details, see Online Appendix I).

I first show how Japanese households performed at biologically producing male heirs. To do this, I take each household generation and look at the number of surviving heirs at the end of their reproductive cycle. This under counts the number of surviving heirs because I only observe male sons who are residents of the village. However, these were stem households, so it would have been rare for all sons to out-migrate and leave the household heir-less.

I plot the number of surviving heirs against landownership in bins in Figure 6a.Footnote ²⁰ The data shows that the land poor class, with less than 1 koku of holdings, had a 34 percent chance of having no male heir, while the land rich had a slightly lower chance of 24 percent. Therefore, landownership could increase fertility and lessen the probability of having no heir but this also had its limits。This was equally true for other societies, including the elite of England in normal years and times of calamity (see Figure 6b).Footnote ²¹ I find that 28 percent of English elite households had no male heir during normal years, and an even larger 42 percent failed to have a male heir during the century after the black death when mortality rates rose. This relationship was highly stable over many centuries, and we know the later elites did no better at securing heirs (Gobbi and Gõni Reference Gobbi and Gõni2021). Therefore, the securing of male heirs was a common issue for these societies due to (1) half of children being female and (2) the high mortality rate during this era meant approximately one-third of children died before adulthood (Wrigley et al. 1997).

Figure 6 SHARE OF HOUSEHOLDS BY BIOLOGICAL HEIRSHIP

Notes: I use the number of biological heirs listed in the sources at the end and reproduction (Japan) or at the point of death (England).

Sources: Japanese inequality data and Russell (1948).

Adoption was an institution that could resolve the issue of heirship. The Japanese peasants often adopted adult men to marry into their households. The adoptee would usually be the surplus sons of other peasants, who were not in line to inherit lands. They did not have to be relatives, as in other societies, although this was preferred. In the extreme, the next generation could be composed of total strangers (Kurosu and Ochiai Reference Kurosu and Ochiai1995; Kurosu Reference Kurosu2013). Adopting adults had the advantage of reducing the risk associated with mortality at younger ages.

Unlike modern Western-style adoption, where child welfare is a key motive, the Japanese adopted in order to preserve the male line. Therefore, adoption was not coincidentally solving the issue of heirship in some households. Rather, it was expressly used to resolve the issue of heirship. I can causally show this by testing whether parents had adopted a male heir by the end of their reproduction, conditional on their success at biologically producing male heirs as shown in Equation (2).

(2) $Adopted\,Male\,Hei{r_i} = {\alpha _v} + {\beta _1}Biological\,Male\,Hei{r_i} + {\beta _2}\,landownershi{p_i} + {\varepsilon _i}$

The dependent variable is a dummy for whether the parents adopted a son by the end of their reproductive cycle. This will capture a subset of adoptions that eventually occurred. The key independent variable is either a dummy of whether a male heir exists or the number of male heirs. I instrument this with the sex of the child for the first observed birth, which is exogenous.Footnote ²² As controls, I use village fixed effects and the quantity of landownership.

Table 4 shows the instrument of the sex of the first child is highly correlated with male heir-ship. Using the instrument, I show that the lack of heir increased the chance of adoption by 0.44, which is much higher than the OLS-based estimate. I get a similar result if I change the independent variable to the number of heirs. I can also account for the ages of the children and their chances of survival to adulthood, or whether any sons have left the village, but the results remain the same (see Online Appendix J). The magnitude of these results shows the lack of male heirship was the key driver of adoption in Japan at this time.

Table 4 ADOPTION AND MALE HEIRSHIP

Notes: Huber-White robust standard errors in parentheses. *p < 0.1, ** p < 0.05, ***p < 0.01. The dependent variable is whether there is an adopted heir within the household at the end of reproduction. Heir refers specifically to male heirs. Koku is a local unit measuring value in volume of rice.

Source: DANJURO database.

How did this affect household extinctions? To answer this, I reorganize the data and take one observation at the point of household succession (when the household head changes) or extinction. I estimate how landownership affected household disappearances and extinctions.

(3) $$Extinctio{n_i} = {\alpha _v} + {\beta _1}landownershi{p_i} + {\varepsilon _i}$$

The dependent variable is household extinctions. As household extinctions are not explicitly mentioned in the census, they can be liberally defined as cases in which households disappear from the village census. However, this definition may also capture migration (although household migration was rare), so I also make a conservative definition based on households that both disappear and have no potential heirs. The main explanatory variable is landownership measured by the value of the land in koku, as explained earlier. I use village fixed effects to control for village heterogeneity.

One concern is that households deciding to go extinct may have slowly sold off their lands, leading to reverse causality. I therefore instrument landownership with a 20-year lagged landownership, which should precede the decision to sell off lands and address reverse causality.

I find that regardless of the definition of household extinction, landownership had a strong negative effect (see Table 5). The magnitude appears small but this is because only 14 percent of households disappeared and only 4 percent went extinct per generation. Therefore, landowners ensured household continuity by adopting. As explained earlier, the negative correlation between household extinction and landownership should theoretically have decreased landownership inequality.

Table 5 LANDOWNERSHIP AND EXTINCTION

Notes: Huber-White robust standard errors in parentheses. *p < 0.1, **p < 0.05, ***p < 0.01. The dependent variable is whether a household went extinct at the point of succession. Heir refers specifically to male heirs. Koku is a local unit measuring value in volume of rice.

Source: DANJURO database.

Another way of addressing the effect of adoption is to look at which households adopted or went extinct. To do this, I take each household-year as an observation and estimate the effect of landownership on extinction, which I show in Figure 7a.Footnote ²³ The striking finding is that households with the average (3.5–4 koku) or above in landownership were not going extinct. This was clearly driven by adoption, defined as the number of cases where an adoptee succeeded the household as plotted in Figure 7b. Adoption rates were much higher among the rich, which is consistent with findings from Kurosu (Reference Kurosu2013) for other villages in contemporary Japan. Adoption also functioned close to its theoretical ideal of preventing all household extinctions. This is not surprising, as both the adopter and adoptee had much to gain from the relationship. Adoption also functioned, but to a lesser degree, among poorer households. Most notably, it was land poor households with lands worth less than 1 koku or those that were landless that had the highest rate of extinction. While some of these households succeeded in finding adoptees before death, others failed due to the small amounts of landownership. However, the extinction of the landless must have decreased inequality because poor households tended to disappear, decreasing the share of the poor within villages.

Figure 7 RATES OF EXTINCTION AND ADOPTION PER DECADE BY LANDOWNERSHIP

Source: Japanese inequality data.

How effectively did adoption insulate lands from being redistributed via social mechanisms? One measure of this is the share of lands owned by households that went extinct. I find extinctions led to only 10 percent of lands being redistributed per century. Such lands were taken by relatives or passed to village organizations, who at times found families to take over the land (Okada Reference Okada2006). This contrasts with the English data in Figure 5b, where 20–30 percent of the richest male lines were going extinct per generation. These lands were then transmitted to other male lines. Although there is no comparable data to track landownership in England, there is no shortage of documented cases of households becoming rich due to strategic marriages to heiresses (Clay Reference Clay1968; Habakkuk 1994; Payling Reference Payling2001; Broad 2004).Footnote ²⁴ The assortative matching involved in marriage meant the lands of the rich generally stayed among the rich (De La Croix, Schneider, and Weisdorf Reference De La Croix, Schneider and Weisdorf2019; Clark and Cummins Reference Clark and Cummins2022).

Finally, was there greater equality in Western Europe preceding the fourth century? While this would be a desirable piece of evidence, there is little evidence for land inequality from Ancient Europe. What exists comes from Roman Egypt (Bagnall Reference Bagnall1992). The first is from 216 CE in the village of Philadelphia in the Fayum, where the Gini coefficient was 0.53 for private lands excluding the landless. The second is from 308/309 CE in the village of Karanis, where the Gini coefficient for private lands was 0.43, excluding the landless. These are similar numbers to post-black death rural Italy, when inequality was unusually low due to the catastrophic shock. It is also comparable to Japan, 1640-1870. While these are two small case studies, these findings are consistent with a relatively equal society.