January 2019 newsletter - Was Domar Right? Serfdom and Factor Endowments in Bohemia
22 Jan 2019
Alex Klein1 reports on the use of a unique database to shed light on the origin and persistence of ‘coerced-labor’ systems.
Are institutions shaped by factor endowments? Coerced labor institutions such as serfdom and slavery prevailed in many societies for centuries and had wide economic repercussions. But what caused such institutions to arise and survive? For decades, researchers have struggled to understand why slavery and serfdom dominated certain historical societies, and why they lasted so long, through the nineteenth century. Research thus far has largely been theoretical, relying on qualitative and descriptive approaches. Recent advancements in digitisation of historical data sources, however, now make it possible to address these questions from a new perspective: a quantitative one. In recent research with Sheilagh Ogilvie2, (Klein and Ogilvie, 2017)we quantitatively examine serfdom in bohemia (part of what is now the Czech republic) by taking advantage of enormously rich eighteenth century censuses, unique not only for a society in Eastern Europe but for any society at that time. Our research allows us for the first time to empirically investigate a leading and controversial hypothesis that suggests that the land-labor ratio of the time affected labor coercion.
We rely on a comprehensive tax register that provides data on all the 11,349 serf villages in bohemia in 1757. This register, known as the Theresian cadaster (Tereziánský katastr), recorded serfs’ coerced labor obligations at the level of each village in bohemia. The cadastre provides information on the number of serf households that were required to provide coerced labor and the number of days they had to do it. Bohemia experienced classical medieval serfdom, in which peasants were obliged to deliver coerced labor along with other payments to their landlords in return for being allowed to occupy land. To enforce the delivery of coerced labor, as well as other rents and taxes, bohemian landlords imposed restrictions on geographical mobility, marriage, household formation, settlement, inheritance, and land transfers. In most of Western Europe these obligations petered out by early modern times, but in bohemia and most of Eastern Europe, they survived and intensified in a development known as the ‘second serfdom’ lasting all the way to the early nineteenth century. During this time, many landlords began increasing the coerced labor they extracted from serfs, demanding it from previously exempt groups, and using it not just for farm work but also for many other non-agricultural activities such textile manufacturing, ironworking, glassmaking, brewing, fish-farming, or transportation, just to name a few. What were the causes of it? To answer this question, we tested one of the most well-known, and most vigorously criticised explanations. Domar (1970) speculated that coerced labor systems were caused by high land-labor ratios. Under his theory, landowners devised serfdom and slavery in economies where labor was scarce relative to land; thus, these institutions ensured they could get labor to work their land at a lower cost. This hypothesis, however, stirred a long-lasting debate. It was largely dismissed because it did not accurately reflect historical reality; for example, increases in land-labor ratios after the Black Death, in the middle of the fourteenth century, were followed by a decline of serfdom in some societies and an intensification in others.
Acemoglu and Wolitzky (2011) breathed new life into the Domar theory by putting forward an explanation of why land-labor ratios might affect labor coercion differently in different contexts. In line with theories advanced by Postan (1966) and North and Thomas (1971), they argued that an ‘outside option’ played a role. That is, labor scarcity increased the wage that serfs could earn in outside activities, such those in the urban sector. The presence of alternative work options would discourage coercion. So, a rise in the land-labor ratio could increase the use of coerced labor in some contexts, but it could also decrease its use via its effects on serfs' other work opportunities. The relative size of these two effects will vary, so that the same rise in land-labor ratios can result in different outcomes. Despite the theoretical work, quantitative empirical evidence on this issue has been practically non- existent. Thus, our research with data from the Theresian cadaster provides what we believe to be the first quantitative analysis of labor coercion under serfdom.
To investigate the effect of factor proportions on coerced labor under serfdom, we used our data on 11,349 Bohemian villages in 1757 to estimate a reduced-form relationship between labor coercion and the land-labor ratio, controlling for urban potential and other village characteristics. Generally, our regression specification can be written as follows:
Coercioni,j = f(Land-Labori,j, Urban Potentiali,j, Xi,j, εi,j)
where i denotes a village and j an estate and f is the function relating coerced labor to the regressors. Coercioni,j denotes the number of days of coerced labor extorted from serfs per week in village i on estate j. Land-Labori,j denotes the land-labor ratio in village i on estate j. The vector Urban Potentiali,j is a vector of five variables denoting the potential for towns to offer serfs outside options in village i on estate j. The vector Xi,j includes village, estate, and region controls: the number of households in village i on estate j, village-level latitude and longitude, dummies for each type of estate lordship (noble, royal, ecclesiastical, etc.), and controls for the region (kraj) in which the village was located. We also allow for estate-level fixed effects although, for the reasons explained below, we do not estimate them directly. The error term in the equation is denoted by εi,j. We use two alternative measure of labor coercion. One focuses on human time only and comprises the total number of days of human labor the village was obliged to provide to its landlord each week. The other is the total work energy extracted from serf households: animal energy was combined with human labor to yield the total number of ‘serf-equivalent’ days of work the village had to provide each week.
Our general regression specification allows for the possibility that the relationship between the land-labor ratio and labor coercion was a non-linear one by including the square of the land-labor ratio as a regressor. As labor scarcity rose, landlords might have approached a technical frontier of coercion, at which they were no longer able to extort additional labor regardless of its value to them. When the land-labor ratio rose above a certain level, labor might become so scarce that most of it was required to keep serfs themselves alive, reducing the increment the landlord could extract despite his intensified demand for it.
Our preferred estimation approach is one in which in which we allow for the possibility that different mechanisms generate the zero and the positive values of labor coercion. In this two-part model, the first part is a logit regression which models the probability that a village has positive coerced labor, while the second part uses OLS to estimate a linear model of coerced labor conditional on such labor being positive. The same set of regressors was used in both parts. We estimated this two-part model using the Stata command twopm of Belotti et al. (2015). Although the two-part specification is the one on which we place the most emphasis, we also report the results of using the OLS and Tobit specifications.
The regression we estimate is a reduced-form one, so the coefficients on the land-labor ratio obtained from estimating this equation do not measure the Domar effect. They measure the net outcome of the two possible effects pointed out by Acemoglu and Wolitzky (2011), the positive Domar effect and the negative outside options effect. If the net effect of the land-labor ratio is positive, then one can say that the Domar effect dominates, even though the precise sizes of it and the outside options effect are unknown.
Tables 1 and 2 show the results of estimating the regression equation for human-only and human-animal coerced labor respectively. The tables report the marginal effects implied by the two-part and RE Tobit regressions for easier comparison with the OLS coefficients. All three estimation methods yield virtually the same marginal effects and statistical significance for all variables except the land-labor ratio, where the two-part and RE Tobit marginal effects are both approximately twice the size of the OLS coefficient. The characteristics of our data strongly indicate the use of the two-part model, so we focus mainly on the two-part results in the discussion that follows.
What light do our regression results shed on the Acemoglu-Wolitzky theory about coerced labor under serfdom? As Tables 1 and 2 show, for both definitions of coerced labor, the marginal effect of the land-labor ratio is significantly different from zero, as is its squared term, implying a curvilinear relationship. Figure 1 graphs the elasticity of labor coercion with respect to the land-labor ratio according to the regression models in Tables 1 and 2, setting all other regressors at their sample mean values. All three estimation approaches imply that the elasticity of coercion with respect to the land-labor ratio is positive, indicating that the Domar effect outweighs the outside options effect, over virtually the whole range of values. What explains the second feature of our results, the decline in the elasticity of coerced labor with respect to the land-labor ratio as the latter rose? In villages with very high land-labor ratios, labor was so scarce that even the impressive coercive capacities of landlords reached a technical frontier at which it became impossible to extract more coerced labor. There was an irreducible minimum of labor which serf households themselves required in order to ensure survival and availability of any coerced labor. In villages with very high land-labor ratios, labor was so scarce that most of it was needed just to keep serfs themselves alive, so lords encountered technical constraints in extracting more of it. This accounts for the declining, and ultimately zero or negative, elasticity of labor coercion with respect to the land-labor ratio when the latter reached very high values. In other words, when labor reached a state of extreme scarcity, market pressures broke through and even highly effective coercive techniques could not counteract them. Opportunities in the urban sector also had the potential to affect labor coercion under serfdom. However, as Tables 1 and 2 show, most categories of town exercise no statistically significant effect on either measure of labor coercion, and their economic significance of almost all measures of urban potential is very minor.
Overall, we find that where the land-labor ratio was higher, labor coercion was also higher, implying that the Domar effect outweighed any countervailing outside options effect. The effect has two additional features, both arising from the technology of coercion under serfdom: the effect of the land-labor ratio on labor coercion was much larger for human-animal than for human-only labor, and it declined as the land-labor ratio rose. Indeed, our findings show an inverted U-shape effect on land-labor ratio of labor coercion, meaning that in the villages with only a few serfs, coerced labor obligations were not very severe.
We also present evidence that supports Acemoglu and Wolitzky’s conjecture that serfdom was strong in Eastern Europe partly because the urban sector was too weak to generate outside labor options for serfs. Indeed, our econometric results show that urban potential exercised little statistically or economically significant effects on labor coercion. Towns’ lack of impact on labor coercion reflects evidence that in Bohemia and other parts of eastern- central Europe towns were too few and too weak to have any serious impact on serfdom.
It has long been believed that serfdom arose from class struggle, royal strength, urban power, or other society-specific variables. Our findings, by contrast, show that even when such issues are considered, economic fundamentals prove paramount. A very important implication of our results is that factor proportions — the comparative value of labor and land — do indeed affect institutions. Even though political economy and a number of other variables can impact the labor coercion, our results show that economic fundamentals help shape the institution of serfdom.
It has long been believed that serfdom arose from class struggle, royal strength, urban power, or other society-specific variables. Our findings, by contrast, show that even when such issues are considered, economic fundamentals prove paramount. A very important implication of our results is that factor proportions — the comparative value of labor and land — do indeed affect institut-
ions. Even though political economy and a number of other variables can impact the labor coercion, our results show that economic fundamentals help shape the institution of serfdom.
Notes and references
1. Director of the Macroeconomics, Growth and History Centre, University of Kent.
2. Professor of Economic History, University of Cambridge.
Acemoglu D, and A Wolitzky (2011). ‘The Economics of Labor Coercion.’ Econometrica 79(2): 555-600.
Belotti F, P Deb, W G Manning and E C Norton (2015). ‘twopm: Two-part models.’ Stata Journal 15(1): 3-20.
Domar E D (1970). ‘The Causes of Slavery or Serfdom: a Hypothesis.’ Journal of Economic History 30(1): 18-32.
Klein A and S Ogilvie (2017): Was Domar Right? Serfdom and Factor Endowments in Bohemia, CAGE WP 344/2017
North D C and R P Thomas (1971). ‘The Rise and Fall of the Manorial System: a Theoretical Model.’ Journal of Economic History 31(4): 777-803.
Postan M M (1966). ‘Medieval Agrarian Society in its Prime: England’, In M M Postan, ed., The Cambridge Economic History of Europe, Volume 1: The Agrarian Life of the Middle Ages. Cambridge, Cambridge University Press: 548-632.
From issue no. 184, January 2019, pp. 5-9