1 Introduction

Unified Growth Theory explains the emergence of modern economic growth as the result of a Malthusian interaction between technological innovation and the size and composition of population, which stimulate demand for human capital, resulting in further technological discoveries as well as demographic transition, as households begin to favor quality over quantity of offspring (Galor & Weil, 2000; Ashraf and Galor, 2018). Empirical evidence backs these proposed mechanisms, see e.g. Ashraf and Galor (2011) on the determination of population density by productivity, and Galor and Klemp (2019) on changes in preference regarding quantity and quality of children. Time series analyses have found convincing evidence of a transition from a post-Malthusian era characterized by Malthusian population dynamics but accelerating technological progress, to a modern growth regime, for example Møller and Sharp (2014) for England, and Klemp and Møller (2016) for Scandinavia. More recently, Jensen et al. (2022) provide evidence of a transition from the Malthusian to the post-Malthusian regime for Denmark. Together, this work has demonstrated that the Malthusian regime in Denmark came to an end around 1775, initiating a post-Malthusian regime, which itself was replaced by modern economic growth around a century later. Jensen et al. (2022) suggest historical underpinnings for these developments, which in a wider context simply emphasize the fact that the timing of the transition from stagnation to growth reflects the obvious implications of Unified Growth Theory: that it is determined by initial differences in geographic factors and historical accidents which themselves are reflected in a variety of institutional, social, cultural, and political differences between countries.

Nevertheless, as the large and ever-expanding literature on the British Industrial Revolution demonstrates, understanding these differences can be enlightening, and we consider here a related case, that of the development of Denmark where industrialization was not so much through textiles but rather food processing, and from where exports of foodstuffs played a not insignificant role for sustaining the British Industrial Revolution itself. Workers in industrial cities needed to eat, and by the end of the nineteenth century, Denmark established itself as the dominant foreign supplier of UK staples such as butter and bacon. Lampe and Sharp (2018) have suggested an important role for certain landowning elites from the eighteenth century for this story and argue that they set in motion a process which would lead to the emergence of industrialized dairying by the end of the nineteenth century. We test their hypothesis, but also provide evidence that their impact went beyond the economic sphere and to a particular set of values which make Denmark the society it is today: one characterized by very high standards of living, equality, democracy, freedom, and more.Footnote 1

Outside Denmark, the impact of elites on economic and political institutions is well-established,Footnote 2 and in this literature, agricultural elites are only very occasionally good for development, for example because they can lobby the government to the benefit of citizens (see for example Bates, 1974 and Dell, 2010). Less well-understood however is how elites may foster growth through knowledge diffusion, and in the present work we shed new light on this channel in an agricultural setting. We exploit a specific example centered on the Kingdom of DenmarkFootnote 3 and the emergence of a modern dairy industry based on a new technology, the automatic cream separator (a steam-powered centrifuge), and a new institution, the cooperative creamery, which propelled the country towards prosperity in the last decades of the nineteenth century (for a brief account, see Henriksen, 1993). After the first was founded in 1882, hundreds of cooperative butter factories, or creameries, spread throughout the whole country within a decade. Massive increases in productivity followed, production boomed, Denmark captured a large share of the important UK marketFootnote 4 for butter and other agricultural products and witnessed rapid economic catch up with the leading economies of the day,Footnote 5 as traditional suppliers of agricultural goods such as Ireland and the Netherlands lost market share. An important point motivating the present work is the rapidity with which this industrialization of the countryside spread, with the first wave of butter factories established by 1890; see Fig. 1. The vast majority of these were cooperatively owned by the farmers, conferring various practical benefits, regarding for example the ability to raise capital and for quality control, relative to their privately managed competitors, which they vastly outnumbered and which were often more focused on supplying fresh milk to the cities (see for example Henriksen, 1993).

Fig. 1
figure 1

Location of cooperative butter factories in 1890, Source: own work, based on Bjørn (1988)

In a light-hearted manner, Francis Fukuyama (2011) has described the issue facing developing countries as the problem of ‘getting to Denmark’, a metaphor for a society characterized by wealth, the rule of law, good governance, and related virtues. But how did Denmark get to Denmark? The rise of the cooperatives is often considered a prominent part of the answer, but, following an argument first laid out by Lampe and Sharp (2018), we demonstrate that elitesFootnote 6 facilitated this. This was despite the fact that these were elites in the traditional sense of the word, being large, often aristocratic landowners, and like elites elsewhere they were jealous of their status and were not always supportive of economic progress for the wider population. Some members of these elites were however also ‘knowledge elites’, transferring technology, in particular through the introduction of new agricultural methods including the idea of centralizing the production of dairy products and reaping the fruits of supplying the growing cities at home and abroad. This latter also means that their presence is easy to measure since the locations of their proto-modern dairies, so-called hollænderier, are well documented. Most importantly for the present work, however, these were enlightened elites, providing inspiration, education, and enlightenment thought in general to local agriculture, and we do indeed find that their presence is a major factor behind the spread of the cooperative butter factories a century later, and is associated with greater local wealth. These elites, the contributions of whom we discuss more below, were of course a subset of the wider class of elite aristocratic landlords (not to mention an even wider class of intellectual, spiritual, royal, etc. elites), but we refer to them simply as ‘elites’ below.

Our econometric analysis relies on a novel database for which we have collected a substantial amount of data from a combination of primary and secondary sources. For the main analysis we have this data on the parish level (for 1,675 parishes), but we also demonstrate that our results are robust to an alternative specification where we divide Denmark into 38,010 1 × 1 km grid cells. Controlling for a large number of other relevant determinants of cooperation, including indicators of land quality and suitability for dairying, we find that each parish’s exposure to cooperative butter factories in 1890 (measured as the sum of the inverse distance to every cooperative from that location) correlates with that parish’s treatment by the elites, measured by the sum of the (tax assessed land quality adjustedFootnote 7) sizes of all estates with hollænderier in 1782 weighted by their distance to the parish centroid. Then, we provide evidence consistent with causality using the fact that these spread throughout the country in a particular way, starting with the first, which was established on an estate called Sofiendal, the distance to which we use as an instrument.Footnote 8 Our results are not just statistically but also highly economically significant, implying that an increase of one standard deviation in the elite influence increases the cooperative exposure by 56 percent of the mean exposure in one of our preferred specifications. They are also robust to a variety of alternative specifications.

This represents a significant revision of our understanding of the Danish development path, with important policy implications, largely due to the role Denmark often plays as a poster child for the idea that countries can develop through cooperation in the countryside, as is perhaps best exemplified by the classic account commissioned by the Food and Agricultural Organization of the United Nations after the Second World War (Skrubbeltrang, 1953). However, the idea that a country with a large number of peasants and cows (for example Ireland at the turn of the twentieth century, or India after the Second World War) could simply cooperate its way out of underdevelopment has been severely criticized by Lampe and Sharp (2018) for exactly the reason presented here, that cooperatives stood on the shoulders of at least a century of accumulated improvements first introduced by elites.

Our findings also contribute in other important ways. Cooperation is often seen as something which is facilitated by or promotes social capital (see for example Valentinov, 2004), which in turn is often given as one of the reasons for other ‘typically’ Danish features such as the development of a large welfare state and inclusive democracy. We do not however consider the cooperative institution as such to have been the defining feature of Denmark’s transition to modern economic growth, and indeed there is evidence that their largest contribution was simply to make it possible for peasant farmers to pool resources and invest in modern technology, rather than any inherent productivity advantage as such (Henriksen et al., 2011). Nevertheless, the cooperatives do in fact play a wider role for Danish national identity today and have been associated with a particular set of values, as highlighted by Fukuyama.

As Galor (2022, p. 168) has emphasized, ‘The great diversity of cultures across the world is an outcome of each society's adaptation to its unique ecological niche and historical circumstances.’ In the Danish case, the industrialization of agriculture emerged during the dominance of a particular brand of Lutheran revivalist thinking, embodied by the clergyman, historian, and poet Nikolaj Frederik Severin Grundtvig, whose impact ‘has been so comprehensive in Denmark that he is a fundamental and inescapable reference point whenever themes such as people, nation, democracy, freedom, church, and education are to be discussed’ (Hall et al., 2014, 7). His philosophy (or interpretations of it) came to be taught at a network of so-called ‘folk high schools’, a type of adult education college. This philosophy came to be intertwined with the peasant cooperative movement and a major influence on the Danish liberal movement, which supported them in parliament. As more modern dairying expanded, farmers demanded more education, in line with the aforementioned demand for education in the production process highlighted by Unified Growth Theory, which was satisfied through the folk high school movement. Thus, peasants took in Grundtvigian philosophy alongside lectures on best practice within agriculture, which there was an increasing demand for as industrialized dairying expanded, and the human capital accumulation, together with the physical capital embodied in the cows, contributed to the rapid adoption of industrialized dairying, but only once the automatic cream separator was invented as we discuss below.

We find evidence that folk high schools were more common in areas that had specialized in dairying, following the inspiration provided by the elites, and that these areas also witnessed the rise of industrial dairying (the cooperative butter factories) and greater (accumulated) local incomes, which we measure using wealth tax data from the early twentieth century. Of course, the structure of the Danish economy changed massively following that, with economic activity moving into the cities, and local (parish-level) wealth, even if we had data on it today, would surely not reflect the location of cooperative butter factories more than a century previously. But beyond the obvious point that Danish industrialization through dairying marked the onset of a modern growth regime, we also consider evidence that the cultural impact of Grundtvigianism might cast longer shadows, and in ways which have significant implications for modern Denmark: first, through a belief in democracy (which we measure by electoral participation); and second, through a belief in individual freedom (which might in part also be captured by electoral participation, but which we in particular measure through participation in the Danish government’s mass COVID-19 testing and vaccination regime).

Figure 2 aims to place our findings both in time, but also in relation to Unified Growth Theory. As noted above, Jensen et al. (2022) explain, and test for using time series data on fertility, mortality, and real wages, how Denmark escaped from Malthusian crisis in the eighteenth century through a combination of agrarian reform, and the sort of improvements in the countryside we consider here. Boserupian technological progress led to the emergence of a post-Malthusian regime and population growth, with the fertility transition and modern growth occurring alongside industrialization.

Fig. 2
figure 2

The main mechanisms and outcomes described in the text

Here, we demonstrate how enlightened rural elites both inspired investment in dairying, but also laid the basis for an important educational channel which would lead to the rapid adoption of industrialized agriculture following the invention of the centrifuge. This in turn had an initial impact on living standards, but a more permanent impact on culture. Thus, this study contributes in the broadest sense to our understanding of what made Denmark Denmark. Moreover, Denmark’s current status as an ‘agricultural superpower’,Footnote 9 dominated by massive firms such as Arla (a dairy cooperative) and Danish Crown (a food, especially meat, processing firm previously also a cooperative until 2010), is usually traced back to the aforementioned developments in the 1880s. The existing literature (basically Henriksen, 1999, inspired by Ó Gráda 1977) has attributed the rapid diffusion of industrialized dairying in Denmark mostly to pre-existing cow densities, while the cooperative form of production has been associated with (low land) inequality, (protestant) religion and social capital (Fernández, 2014), while O’Rourke (2006, 2007) has argued that it was the absence of conflicts and the egalitarianism of the Danish population which distinguished it from Ireland, where cooperation emerged later and less successfully.Footnote 10

The following section provides a brief review of the relevant literature on the role of elites for development, and provides the historical background for Denmark. Section 3 provides an empirical analysis of the impact of this for the emergence of the cooperatives more than a century later. Section 4 tests the robustness of the main results, and Sect. 5 presents evidence for the relationship between the cooperatives and local economic development. Section 6 concludes.

2 The role of elites for development in the Danish context

An elite class might emerge through the evolutionary and Malthusian mechanisms associated with Unified Growth Theory, or they might be in place due to historical accident. In the Danish case, as we will discuss below, agricultural innovations were pioneered by a largely immigrant group within traditional landowning elites, taking advantage of the privatization of crown estates to introduce new agricultural methods to the land. Their role links to a long running theme in growth and development as to how elites influence the long run economic trajectories of societies (Amsden et al., 2012). It is undeniable that elites exert disproportionate influence on how institutions are designed, and factor endowments are used, but it is less well known how their actions feed into the development process.

For the case of agricultural elites, the existing literature has provided conflicting views on the role of large landowners for economic development. On the one hand, the concentration of agricultural resources in the hands of large landowners and accompanying high levels of land inequality are often seen as an impediment to development. Engerman and Sokoloff (2002) stress the interaction between factor endowments and the resulting impact of land inequality and how higher land inequality leads to agricultural elites who favor slavery and extractive institutions, which produces poor economic outcomes, see also Acemoglu et al. (2001) and Bannerje and Iyer (2005). Galor et al. (2009), Baten and Juif (2014) and Cinnirella and Hornung (2016) demonstrate that high land inequality causes elites to block investment in human capital, although Ashraf et al. (2020) demonstrate that elites in nineteenth century Prussia moved to emancipate laborers as they increasingly required skilled labor and wanted to incentivize human capital acquisition. Others have stressed that high land inequality limits the scope for agricultural cooperation both through lower social capital (Fernández, 2014) and through direct crowding-out (Henriksen, 1999). On the other hand, in contexts in which property rights are poorly defined, large landowners can ‘shelter’ dependent peasants from extractive state institutions (Dell, 2010) and effectively lobby for better provision of collective goods and infrastructure than politically weak peasant communities (Dell, 2010, 2012).

By contrast, we stress that agricultural elites may spread knowledge, which then subsequently aids development in the agricultural sector, the industrialization of which, as explained above, was the key factor in bringing Denmark decisively into the modern growth regime. In other words, our work suggests that agricultural elites may also be knowledge elites, who facilitate later development. Squicciarini and Voigtländer (2015, 2016) have demonstrated that knowledge elites played a significant role in the industrialization of France by e.g. running businesses themselves or exchanging knowledge with entrepreneurs. Our work emphasizes the importance of knowledge spillovers and agricultural enlightenment by and through ‘improving landlords’ (Mokyr, 2009, ch. 9), and shares some similarities with Hornung’s (2014) work on high-skilled immigration of Huguenots into Prussia and the diffusion of technology in that country.

As noted above, we rest to a certain extent on the work of Lampe and Sharp (2018), and we provide a more detailed historical background in Online Appendix A. They explain how the origins of modern dairying in Denmark begin in the Duchies of Schleswig and Holstein, ruled by the King of Denmark in personal union until 1864 when they were lost to Prussia. There, an intensified crop rotation system with an important and professional dairy component, known as Koppelwirtschaft in German, or kobbelbrug in Danish, was developed on the large manorial estates (Porskrog Rasmussen, 2010a). It became the dominant manorial field system in the Duchies in the 1700s, and included unprecedentedly large herds of milch cows and the invention of an innovative centralized system of butter production, the hollænderi,Footnote 11 which was essentially a proto-modern dairy with a centralized production facility for separation of cream from milk and production of butter much like, but of course technologically very different from, the cooperative butter factories a century later. This system was introduced to the demesne farming of large manorial estates, and not in peasant agriculture, because these were the most commercially oriented agricultural units, the most likely to be able to sustain the considerable capital investments and labor efforts (via corvée or hired labor) required for reorganizing the fields, and also the ones with the largest freedom to act under the institutional framework of the time.

The introduction of this so-called Holstein System (Koppelwirtschaft combined with the hollænderi) was facilitated by increasing turnover and concurrent change in the ownership structure of Danish estates. In the 1600 s Denmark consisted of a large number of crown estates, under the direct administration of the monarch, and smaller estates owned by the nobility. From the 1600 s and into the 1700s, the bad finances of the crown, largely as a result of continuous wars against Sweden until 1721, led to repeated rounds of crown estate sales, until by the 1740s almost all had been privatized. The defining moment for our purposes came with the ascent of Adam Gottlob Moltke to Lord Chamberlain for King Frederik V in 1746, and more precisely when Moltke sold his estate of Niendorf near Lübeck in Holstein, on which the Holstein System was firmly established, in 1761 and the same year took the former leaseholder, Johann Matthias Völckers, to his estates on ZealandFootnote 12 to become his administrator and agricultural reorganizer there (Porskog Rasmussen, 2010b). Völckers started on the newly established farm of Stenkelstrup (later named Sofiendal after Moltke’s second wife) to implement an exact copy of the Holstein System, finishing this in 1766, and he then proceeded to reform other estates owned by Moltke. Moltke was imitated by his neighbors, and Lampe and Sharp (2018, 75–80) describe how this established a particular pattern around Denmark, with greater concentrations of estates using the Holstein System closer to Sofiendal, which we exploit for our instrument.

The list of estates having a hollænderi in 1782 reads like something of a who’s who of nobility: old, new, high, low, Danish and foreign, and includes some of the leading reformers of the time. Unfortunately, we do not have data on the location of hollænderier for later years. Given however that only 10 years later, Europe embarked on a period of severe social and economic turbulence marked by the French Revolutionary and Napoleonic Wars (1792–1815), which was to have severe consequences for Denmark, we feel that this date represents a reasonable picture of the spread prior to the nineteenth century. Lampe and Sharp (2018) then demonstrate the spread to the peasantry based on a large number of primary and secondary sources. In short, these reveal two things: first, that estates in the western part of Denmark (i.e. further away from Sofiendal) developed slower than in the east, and second, the peasantry began the period as relatively backward, but had an early advantage in the east, where by the late 1830s or 1840s good dairy practices were spreading to the peasantry in the south of the island of Zealand close to where Moltke first introduced the Holstein system in 1766. This supports our narrative that both the adoption of the Holstein system by estates, and the early spread of good practices to the peasantry, was more frequent closer to Sofiendal. Below, we measure the increasing focus of the peasantry on dairying through data on cow densities, which in the nineteenth century was greater closer to the demesnes of the elites.

These elites continued to innovate beyond the pure transfer of technology from the Duchies. Lampe and Sharp (2018) argue that their main contribution was the early introduction of an ‘enlightened’ approach to dairying and agriculture more generally, involving accurate measuring and recordkeeping, combined with sophisticated bookkeeping and accounting. This allowed first of all for a scientific and experimental approach to agriculture, answering questions such as how best to feed, breed, and milk cows. Second, accounting allowed for a better idea of profitability, allowing for profit-maximizing behavior and for example the discovery that specialization in dairying was the best strategy. The knowledge built up in this process laid the foundation for the rapid spread of smallholder dairying later on, in particular because – apart from the example of specialization—it was increasingly taught through specialized agricultural schooling and apprenticeship programs, providing an important human capital channel.

During the latter part of this process, the teachings of Grundtvig gained followers around Denmark. Grundtvig was born in 1783, the son of a pietist Lutheran preacher, and by 1803 he had graduated with a first-class degree in theology from the University of Copenhagen. Already in 1824, he described himself in a poem as ‘Luther the Little’ but following a libel case he was censored by the then absolutist king and studied in England between 1829 and 1831, where he marveled at the power of steam engines and appreciated the collegiality of Trinity College, Cambridge. His experiences there together with his reading of both British and German liberal thinkers seems to have inspired his subsequent work calling for a new society founded on a common experience of being a free citizen of Denmark, and based on a common history and language instead of faith (see Fukuyama 2015). At a time of transition from absolute monarchy to democracy, he argued for freedom of speech, press, religion, education, trade, assembly, voluntary participation in defense of the country and poor relief, and even voluntary payment of taxes (Damsholt 2015). This philosophy is apparent even today. Denmark has no minimum wage, and relies on voluntary unemployment insurance, for example, with the large welfare state constructed after the Second World War ‘ultimately based on a liberal view of the state’ (Hall and Korsgaard 2015, 21), embodying Grundtvig’s idea of a general responsibility towards the common good as well as individual responsibility. This belief in the importance of personal freedom came to dominate much of Western thinking following the Enlightenment.Footnote 13

Grundtvig’s idea of ‘freedom for all people’, or ‘Freedom for Loki as well as for Thor’, formed the basis of what became the ‘folk high school’ movement. He criticized the state educational system, dominated as it was by Latin and classical teaching, and desired to create free private schools for children and folk high schools for young adults, in line with his idea of ‘education for life’. With Danish defeat in 1864, Grundtvig’s new popular elite (in contrast with the old landowning elite) came to dominate, and between 1864 and 1872, when he himself died, more than fifty new folk high schools were established, and the movement continued to expand (Hall & Korsgaard, 2014). Nearly all of them had two aims: first, to provide a ‘people’s (or folk) education’, with the aim of integrating the populace into a national and democratic community; and second, to provide training in agriculture and housekeeping so that they could improve rural conditions. Korsgaard, (2014, 320) gives an example of a public meeting held in 1865 where three talks were given: on Nordic mythology, on the political situation in southern Jutland, and on horses that contract bone spavin. Thus, as modern agriculture emerged, farmers demanding access to education took in Grundtvigian teachings as well the latest agrarian techniques. Although there is some debate about the importance of Grundtvig for the educational system in Denmark today (Korsgaard & Wiborg, 2006), the continued popularity of free private schools for children and folk high schools for adults notwithstanding, we might expect that this would have a persistent impact through Danish culture.

Denmark and other Nordic countries are often characterized as ‘socialist’ in the American political debate, and Denmark in particular became topical following remarks by Senator Bernie Sanders during his campaign to become Democratic candidate for the Presidency in 2016. This led the then (center right) Danish Prime Minister, Lars Løkke Rasmussen, to attempt to set the record straight in an address to Harvard’s Kennedy School of Government in 2015: ‘I know that some people in the US associate the Nordic model with some sort of socialism. Therefore I would like to make one thing clear. Denmark is far from a socialist planned economy. Denmark is a market economy. The Nordic model is an expanded welfare state which provides a high level of security for its citizens, but it is also a successful market economy with much freedom to pursue your dreams and live your life as you wish.’Footnote 14 In fact, Denmark ranks highly on ‘individualism’ according to Hofstede Insights,Footnote 15 and below we find evidence that areas closer to cooperative butter factories, themselves located close to folk high schools (as measured in 1880) and dominated by Grundtvigian thinking, are associated with a greater belief in democracy (measured by participation rates in the 2015 general election) and a belief in individual freedom (measured by relative lack of participation in the Danish government’s strategy of mass testing for COVID-19).

Grundtvigians, and their Liberal political allies who carried his ideas forward following his death in 1872,Footnote 16 were indeed skeptical about government intervention in health matters. Copenhagen in the mid-nineteenth century was one of the most densely populated cities on the planet, and the medical profession argued strongly in favor of housing and other regulations to combat disease. They met considerable resistance, however, with the important National Liberal politician, Orla Lehmann (1810–1870), one of the heroes of the establishment of the democratic constitution in 1849, speaking out in the upper house of parliament in 1857 against a strengthening of the local health authorities and the establishment of what he called a ‘dictatorship of doctors’, which he saw as an attack on property rights and personal freedom. Some years later, in 1887, liberals succeeded in blocking Denmark’s first epidemic law, although the conservatives succeeded in passing it the following year. In fact, the liberals were still blocking compulsory inspections of creameries in the early twentieth century (Boje, 2020).

Finally, one might ask why it took until 1882 for the first cooperative butter factory to emerge. Although Danish estate owners had market contacts since the beginning, merchants, who sought to establish marketing channels abroad, especially to industrializing Britain, were initially held back by the uneven quality of the produce of peasants, who seem to have held cows mainly because they produced fertilizers for grain production. Marketing butter made by peasants widely proved difficult until the introduction of the automatic cream separator in the late 1870s. The principle that cream could be separated using centrifugal force was discovered in Germany in 1864, but the crucial refinements were made in the Duchy of Holstein in 1876. Separators based on this design were then launched by rival Danish and Swedish firms in 1878/9 (Pedersen, 1999, 51). With this innovation, it became possible to separate the cream from transported milk using centrifugal force. Until then, separation was a time-intensive process involving waiting for the cream to rise to the surface of the milk after which it could be skimmed. Although this could be expedited by cooling the milk using water or ice, transporting the milk to a central facility (usually by horse and cart along uneven country roads) meant that the milk would become completely homogenized making the gravity method of separation impractical. This was not an issue for the cream separator which thus finally allowed peasants to enjoy the benefits of centralized production and marketing pioneered by the elites and their hollænderier more than a century before, this time largely in the form of cooperative, steam-powered butter factories.Footnote 17 They did not enjoy an easy start, however. The first coops in southwestern Jutland met with skepticism from our elites, who commissioned a report hoping to demonstrate their inferiority, but which ultimately reached the opposite conclusion. Thus, there can be little doubt that the estates themselves were not promoting the cooperative form as such. Nevertheless, the cooperatives spread rapidly through the 1880s, and it has been suggested that this led to rapid development in the countryside, something we provide more evidence of below through data on wealth taxes for 1904/05.

3 Persistence and the spread of the cooperatives

To test for the influence of the elites on the location of cooperative butter factories a century later, we examine whether areas closer to estates using the Holstein System were more likely to have cooperative butter factories nearby after the first wave of cooperatives ended around 1890. Specifically, we consider Denmark at the level of the smallest administrative unit available, the parish, of which there are 1675 outside the cities.Footnote 18 In an alternative specification we also provide results where we divide Denmark into 38,370 1 × 1 km grid cells to pick up the very local geographical variation in the location of cooperative butter factories and the estates of the elites and be able to account for potential confounding factors at a very detailed level.Footnote 19 In practice, we calculate measures of the degree to which a parish or cell is near hollænderier and cooperative butter factories. We digitized and processed a myriad of published historical sources and maps based on them to be able to control for every remotely related relationship that comes up in the existing, extensive historical literature. Apart from the estate list by Roholt and the clover map by Kjærgaard, none of this existed in digital form before. Table 1 provides a brief description of the variables included in the regressions below, as well as summary statistics and sources.

Table 1 Summary statistics and variable descriptions

Our main proxy for the spillovers from elites to peasants is the variable elites 1782, constructed in the spirit of Harris’ (1954) ‘market potential’ (mp) measure as the sum of the tax assessed land quality adjusted sizes of all estates with hollænderier weighted by their distance. We calculate this mp for all parishes and grid cells. The reasoning behind this strategy is that the influence of the elites increases with the estate size, but decreases with distance. In other words, the closer and larger the estates, the more influence they would have had in terms of knowledge spillovers. When assessing the relationship in this way, however, one needs to bear in mind that any surviving hollænderier on estates would have been competitors to the first cooperatives (see Henriksen, 1999).Footnote 20 The estates with hollænderier are given in Fig. 3, together with all other estates in 1782.

Fig. 3
figure 3

Location of hollænderier among all estates in 1782, Source: own work based on Andersen (1963), Christensen (1886) and Roholt (2012)

To identify the historical innovating elites, we used a three-step procedure. We started with the geo-referenced list of historical landed estates described in Roholt (2012) for the purpose of mapping the cultural heritage of estates. Since this might be subject to survivorship bias, we also digitized (by hand) a list of all estates included in a list compiled in 1770 and reported in Christensen (1886, Online Appendix B) and matched it with the Roholt list,Footnote 21 and added coordinates for the additional estates based on current and historical maps. We then matched this list of all estates to a list of estates with hollænderier compiled by local officials for the Danish treasury in 1782 and published by Andersen (1963). This is the only comprehensive list of estate dairy units available for the agricultural reform period. We also went through the owner lists on the website underlying Roholt (2012) and other historical sources to determine whether the crown or the government had owned an estate for at least 10 years. We also digitized maps from the land-use studies by Frandsen (1983) and other sources to identify, among others, the demesnes of estates and the historical cultivation systems and in-kind rent payments, the distance to the historical ox road, and the prevalence of clover cultivation for every parish and grid cell.

For the main outcome variable, we identified cooperatives in 1890 by digitizing a published map from Bjørn (1988), a leading Danish agricultural historian who dedicated decades of his life to tracing the cooperative movement. To measure the extent to which a parish or grid cell is exposed to cooperative butter factories, we also construct a market potential measure for dairies in a similar fashion to for the hollænderier and construct the variable cooperative creamery exposure as the sum of cooperative butter factories weighted by the inverse distance from the parish centroid to the cooperative, as measured in 1890 (although we consider 1914 as well below, as well as alternative specifications using the grid cells and estates as alternative units of analysis). We prefer this specification, since it provides a continuous measure of access to butter factories. Thus, a parish or grid cell which is surrounded by butter factories would certainly have been associated with industrialized dairying – and would be assigned a high value in our specification – while an alternative specification such as a dummy for the presence of a creamery in a parish or grid cell would simply be assigned a zero. We nevertheless include this specification in a robustness check.

As an instrument for the influence of the elites, we use the distance to Sofiendal, the first estate in Denmark to establish a hollænderi. As described earlier, ideas spread from here throughout the country. As well as our instrument for distance to Sofiendal, we also present results with an alternative (‘threefield (mp)’). We discuss this alternative instrument, alternative outcome variables, as well as the variables used for analyzing potential mechanisms (see the last two panels of Table 1) in the relevant sections below (Sect. 4 on robustness checks and Sect. 5 on mechanisms). Our controls are listed under four classifications. The distance to the first cooperative creamery is listed as an alternative origin, as the cooperatives might conceivably have spread from there, and if that were the case, we would expect a negative coefficient. Similarly, a negative coefficient on distance to Copenhagen might mean that Denmark’s largest city, located, like Sofiendal, in the east of the country, is driving our results. We do not include this in our regressions, since it is almost perfectly correlated with distance to Sofiendal (at 96 percent) and in fact simply picks up 0.3 of the coefficient on distance to Sofiendal. In a horse race specification, we find that Sofiendal has by far the larger explanatory power.

Under ‘elite controls’, we first control for the proximity of all estates (Estates (mp) 1770) to capture the general political and social effects of the presence of large landowners on peasant cooperation, which might be negative or positive, following the various results in the literature discussed above. This measure is also calculated by the sum of the inverse distances to the estates, i.e. a market potential measure as explained above and as indicated by (mp). Second, we control for the share of a parish that is occupied by estate demesnes (with or without hollænderier) in the late seventeenth century (Demesne share 1680 s). Since this land was directly cultivated by estates, it is much less likely that (peasant) cooperatives would be founded there and hence we expect a negative coefficient. Third, we separately control for exposure (market potential) to (pre-privatization) crown ownership of estates (Crown estates (mp)) as these estates were typically larger and more centrally located and hence might have had an independent and perhaps stronger impact on the surrounding peasant society, so we expect a positive coefficient. Finally, we control for a second wave of innovators’ locations (Late innovators (mp)), again using the sum of the inverse distance to each of them. This latter identifies estate owners who participated in a commission set up by the Danish Royal Agricultural Society in 1858 and were particularly influential, so we might expect a positive coefficient, and goes some way to addressing the idea that the use of the distribution of elites in 1782 might not reflect the later developments and impact of the estate owners.

We then consider various ‘agricultural controls.’ One might further argue that the Holstein System was simply established in areas with previous dairy know-how and cowherds, and we thus introduce a number of variables to take account of this possibility. First, Frandsen (1983) gives direct information on dairying, that is, the amount of in-kind rent payments in butter (in barrels) made by peasants in 1662 per unit of land (Butter production, 1662).Footnote 22 Although this says little about demesne production of butter on estates, it might be positively related to the later establishment of cooperatives if there is persistence in dairying patterns among the peasantry (which hollænderier might just have taken advantage of). Second, as clover was an important part of Koppelwirtschaft and the whole Holstein system, we control for the share cultivated with clover in 1805 as a further control for pre-existing conditions for dairying (Clover share, 1805), although it is also used for feeding cattle, again making the expected sign ambiguous. Third, barley represents the main alternative use of the land,Footnote 23 with an expectation that areas more suitable for cultivation of this crop might be associated with less dairying, and thus a negative coefficient. We therefore control for the average potential yield (rain-fed)Footnote 24 of parish area for barley (Barley suitability), as given by FAO/IIASA (2002). Finally, Frandsen (1983) reports information on the prevalence of field-grass-systems in agriculture in the 1680 s at the time of the large land quality (and use) assessment for a new land-based taxation system. We control for the share of parish area using the field-grass-system (Field-grass-system, 1682)–this might be positively related to dairying, but it might more likely proxy for the use of cattle-raising making our expectation for the sign of the coefficient ambiguous.

We then include ‘market access controls.’ First, we calculate distance to the coast (Distance coast) as a simple measure for openness in the form of market access given that the cooperatives were heavily export oriented and hence, we would expect a negative sign of the coefficient as the exporting potential would be lower for areas further from the coast. Second, we include parish level population density in 1787 (Population density, 1787), proximity to market towns (the most important economic centers outside Copenhagen) (Market town), distance to a railroad in 1890 (Distance railroad, 1890), and finally distance to the Ox Road (Distance Ox Road). This was until well into the nineteenth century the main export route for cattle, the main alternative use of similar resources.Footnote 25

Before we turn to the formal analysis, Table B1 in the online appendix gives the results of balance tests which reveal, based on the significant correlations between the various potential historical predictors of distance to Sofiendal, that most estates were close to Sofiendal (the density of estates is indeed greater on the island of Zealand), and that most crown estates were close to Sofiendal (consistent with our story about the purchase of these by the elites from Holstein). Importantly, there is no correlation with historical butter production, and while there is a positive correlation with field-grass-share, we have no clear expectation as to the sign there for the above stated reasons. The correlation with the distance to the Ox Road is negative, which is to be expected given that it is in the west of the country, whereas Sofiendal is in the east. In short, we see no clear objections to proceeding, and as a first step to disentangle whether the knowledge of the elites spread to the peasantry over time, Table 2 provides baseline OLS results from estimating the following regression equation:

$$Cooperative\,creamery_{i} = \alpha + \beta elites1782_{i} + \varvec{X_i^\prime} \varvec{\delta} + \epsilon_{i} .$$
(1)

where \(i\) is a parish, \(Cooperative\,creamery\) is cooperative creamery exposure in 1890, \(elites1782\) is our proxy for the influence of the elite on parish \(i\) as explained above, \({{\varvec{X}}}_{{\varvec{i}}}\) is a vector of control variables described above, \({\varvec{\delta}}\) is the associated vector of coefficients and \({\epsilon }_{i}\) is the error term. We also compute and report Conley standard errors to account for potential spatial autocorrelation.Footnote 26 In most specifications, \({{\varvec{X}}}_{{\varvec{i}}}\) contains region fixed effects.

Table 2 Main results (OLS)–cooperative creamery exposure 1890 and Elites 1782

In all specifications in Table 2, the coefficient on \(elites1782\) is positive and statistically significant. This implies a positive effect on the likelihood that cooperative butter factories were established in proximity to hollænderier established by elites. The introduction of regional fixed effects in column 2 has very little impact on the estimated coefficient.Footnote 27 In column 3, we control for the distance to the first cooperative creamery, and find that the coefficient is largely unchanged as compared to columns 1 and 2. As controls for estates without hollænderier and the presence of an estate demesne in the parish are included in column 4, the size of the effect is more than cut in half but is still highly significant. This is perhaps not surprising, given that it is very unlikely that a cooperative would be located in or around such locations. Controlling for crown estates and late innovators again reduces the coefficient on \(elites1782\) in column 5. The same is true when we include barley suitability, historical butter production, the share of the area growing clover, historical presence of the grass field system, and the distance to the coast in the set of control variables in column 6. Finally, when we add control variables for market access (Population density, presence of a market town, Distance to the railway and Distance to the Ox road) in column 7, the estimated coefficient on \(elites1782\) remains strongly significant. The estimated coefficient in column 7 of Table 2 implies that an increase of one standard deviation in the elite influence increases the likelihood for a parish to be exposed to cooperative butter factories by (0.0018 * 91.04) = 0.164, or 8 percent of a standard deviation in the cooperative creamery exposure using the most conservative estimate. This indicates that the effect of the elites is not only statistically, but also economically, significant.

The results also produce some interesting findings for the control variables. Being closer to the first cooperative creamery is positively associated with the emergence of cooperatives as one might expect. The presence of estates or crown estates nearby is also positively associated with the emergence of cooperatives but not if the location is on an estate demesne. Further, the presence of late innovators is also positively associated with the presence of cooperative butter factories as expected. All market access and geographical control variables enter with the expected signs except for distance to the coast. Cooperatives were more intensely established further away from the coast. One potential explanation is that the distance to the sea had become less important with the arrival of the railroad, which itself could have been an important locational determinant for many cooperatives. Cooperatives were also more intensely established in areas without a grass-field-system in the 1680 s. This result may be interpreted as a legacy of the historical optimal use of land. In the seventeenth century those parts of the country that were most fertile were cultivated using the traditional three-field-system whereas the less fertile sandy soils especially found in western Jutland were primarily used for grazing. With the advent of the cooperative movement, the fertile parts of the country could support more cows and in turn more cooperative butter factories.

Even though we include an extensive set of control variables, the concern that omitted variables determine both the location of hollænderier and cooperatives could remain. To address this, we propose an instrumental variable identification strategy, where we instrument \(elites1782\) by the distance to Moltke’s estate, Sofiendal, where the Holstein System was first established. This is consistent with our story that the system spread through Denmark inspired by Moltke (and his administrator, Völckers) as well as the historical literature cited above. Hence, we estimate the following instrumental variables model:

$$Cooperative\,\;creamery_{i} = \alpha + \beta elites1782_{i} + \varvec{X_i^\prime} \varvec{\delta} + \epsilon_{i} .$$
(2)
$$elites{1782}_{i}=\theta +\gamma DistanceSofienda{l}_{i}+\varvec{X_i^\prime} \varvec{\Gamma}+{\mu }_{i}.$$
(3)

where we include the great circle distanceFootnote 28 to Sofiendal in our first stage (3) as our excluded instrument in (2).

In Table 3 we present the results of the first stage. The table follows the same structure as Table 2 and we control for the same variables as above. We notice that the coefficient on the distance to Sofiendal is always negative and strongly significant. Moreover, as shown at the bottom of Table 4, the F-test of instrument relevance is always well above 10, as per the usual rule of thumb. Thus, the instrument is highly relevant and there is no sign that distance to Sofiendal is a weak instrument. In column 4 in Table 3, it is seen that hollænderier, among other things, were established in places nearer estates in general but away from the estates of the influential owners. Proximity to crown estates does not influence the location of hollænderier. Higher suitability for barley production increases the likelihood of being close to hollænderier as do places further away from the coast and closer to the historical Ox Road as seen in columns 6 and 7. Importantly, the inclusion of the control variables does not change the statistical significance of the relation between the location of the hollænderier and Sofiendal.

Table 3 IV first stage, Parish level results
Table 4 IV second stage, Parish level results

We present the instrumental variable (or two-stage-least-squares) estimates of Eq. (2) in Table 4. Compared to Table 2, the coefficient on the \(elites1782\) variable is now larger and more stable as control variables are added. Thus, when we rely on the plausibly exogenous component of \(elites1782\), we obtain large and significant effects that can be interpreted as the causal impact of the landed elites and the associated spread of knowledge on the emergence of the cooperative creamery movement. To a large extent the coefficients on the control variables are like those reported in Table 2. It is, however, worth noticing that when using the instrumental variable strategy \(elites1782\) remain significant while other estates now have a negative impact and crown estates are only borderline significant.Footnote 29

We note that there is no significant relationship between butter production in the 1600 s and the location of hollænderier, which would indicate serious endogeneity problems. Nevertheless, even though we have included this and many other control variables, there may still be remaining concerns as to whether the exclusion restriction is violated. To investigate further the robustness of our results to this threat we use the ‘plausibly exogenous’ technique of Conley et al. (2012). Instead of claiming complete exogeneity of instruments the framework tests the degree to which the instrument is endogenous and tests the 2SLS results given the potential level of endogeneity. In our case we test the degree to which our IV, Distance to Sofiendal, directly explains the geographical distribution of cooperative butter factories. If this is low, we can arguably claim that our results are robust to the potential endogeneity. We use the ‘Union of Confidence Intervals’ (UCI) approach and assume that the direct impact of our IV is somewhere between zero and the lower 95% confidence interval value from the reduced form estimate with the full set of controls (δ ∈ [−0.024715;0]). Figure 4 demonstrates how large a potential direct effect of the instrument (Distance to Sofiendal) needs to be to render the IV estimate on \(elites1782\) insignificant.

Fig. 4
figure 4

Testing the plausible exogeneity

For the distance to Sofiendal to directly explain away our results, delta would have to be − 0.0218 or below as indicated in the figure (blue line). As the reduced form estimate of distance Sofiendal is − 0.0230 (green line) the direct channel should be 95 percent of the entire reduced form estimate. We conclude that while it is possible that the direct effect is of this order of magnitude, this seems highly unlikely and hence we believe our results are robust to the potential level of endogeneity of our instrument.Footnote 30

To evaluate the importance of the elite based explanation relative to other potential explanations we calculate standardized coefficients in Table 5. When compared to all the potential cofounders, we see that \(elites1782\) has much larger explanatory power. In fact, the beta coefficient on \(elites1782\) explains more than all other 16 covariates combined. In column 7, a one standard deviation increase in the \(elites1782\) variable leads to a 2.24 standard deviation increase in the cooperative creamery exposure variable, or 2.24 * 2.14 = 4.79 which is 56 percent of the mean exposure. This is our preferred estimate as the instrumental variables estimate arguably measures the causal effect, given the potential for issues with confounding and measurement error for the OLS estimate.

Table 5 Main results (IV, Second stage)–with standardized coefficients (“Beta Coefficients”)

If the presence of hollænderier had a persistent effect for a century before the first cooperatives, we would expect that this meant a spread of the ideas used on the estates to the wider peasant population, due to the traditional links between the estates and the surrounding peasantry, and this is reflected by our reading of the contemporary literature. We consider two potential channels: specialization in dairying in imitation of the elites, and demand for education through folk high schools, in order to be able to exploit the fruits of modern techniques, including the use of the centrifuge. Regarding the former, we can consider the number of milch cows around the country in the intervening period. In 1760 there were 270,000 milch cows in Denmark, increasing to 335,000 in 1774, and 450,000 in 1810 (Drejer, 1962, 22, Jensen, 1998). Moreover, in 1837 we have parish level data from the first (surviving) animal census, which puts the total level at 578,000 in 1837. In 1861, there were 756,834 milch cows in the animal census. By 1881, the year before the first cooperative creamery was founded, there were 898,790. If we are to believe the persistence story, the relative local density of cows should have remained fairly constant before 1882: with no additional spread of elites, the areas specializing in dairying should have remained the same. Consistent with this, the correlation coefficients between the densities in 1837 and 1861 and 1881 are all around 0.9. The vast majority of these cows would have been outside the estates of the elites, since their demesnes only covered around 10 percent of the cultivated land at the start of the period (Lampe & Sharp, 2018), and it was thus the peasantry which largely contributed to this. It then remains to demonstrate that the location of the hollænderier also explains the pattern we observe in the cow densities. To do this, we employ the same empirical strategy as above, but with the cow densities in 1837 at the parish level as the outcome variable. Again, the relationship is very strong and robust–a greater influence from the elites implies greater cow densities, as shown in panel A of Table 6, with panels B and C showing similar results for 1861 and 1881. This result is consistent with the work of Henriksen (1999), who shows that the share of cows supplying a cooperative correlates with cow density, which implies that the peasantry was also turning to dairying in areas close to hollænderier, imitating the elites as the narrative above implies.

Table 6 Channel (IV, Second stage)–cow density 1837, 1861, and 1881, school absence 1857, and folk high schools 1880 and elites 1782

Looking across the table, we note that adding control variables does not change the significance of \(elites 1782\). Importantly, columns 6 and 7 show that the relationship between cow densities and \(elites 1782\) is not driven by past specialization in butter production by the peasantry.Footnote 31 Note that the number of parishes falls from 1675 to 1641 due to missing data on cows for those parishes.

Another mechanism would be through a demand for education, in particular from the folk high schools. Panels D and E of Table 6 thus investigate the impact of the elites on this. The latter considers the presence of folk high schools, the idea being that they would be more likely to be established in locations where demand was greater. We use the location of the 75 folk high schools active by 1880 (after the ending of the main wave of establishment). Their mean age was approximately 11 years in 1880 and we used their age (rather than size, although they are probably positively correlated) as a way to adjust for their impact and calculate the ‘years since establishment’-adjusted mp for folk high schools. Again, the evidence for this is clear, and we consider the potential cultural implications of this below, in particular since they were linked to Grundtvigian ideology. One might also expect to see a similar pattern of demand for schooling in earlier decades, however. We thus supplement this with data from 64 deaneries (Danish: provstier, somewhere between a parish and a diocese) on absenteeism (days of school missed per pupil) from primary schools in 1857. This seems a reasonable proxy for the support from parents regarding the importance of education. Particularly in rural areas, although primary education was compulsory, farmers would sometimes keep their children at home to work on the farm. Despite the modest number of observations, the sign of the coefficient is consistent with the idea that areas impacted more by the elites had more support for schooling. Finally, note that for both mechanisms, specialization and education, we can claim causality when using the distance to Sofiendal instrument.

4 Robustness

In this section, we investigate the robustness of our results in several ways. First, we test the robustness of our results in the temporal dimension. Hitherto we have focused on the time around the ending of the first wave of the cooperative creamery movement around 1890. One might argue that the results are sensitive to this specific year and therefore we select another point in time to test the hypothesis. Specifically, we select 1914, just before the First World War changed the landscape. Columns (1) and (2) in Table 7 presents the results from this robustness test. Focusing on the standardized coefficients in column (2) we see they are statistically significant, though slightly smaller than those for 1890, again emphasizing the persistence of the impact of the eighteenth-century elites, although intuitively, we would expect that new cooperatives were increasingly inspired by other cooperatives.

Table 7 Robustness (IV, second stage)

Next, we show that the relationship between the elites and the cooperative movement is present also at the grid level. We use the same strategy as presented above but use grid cells as the unit of observation instead of parishes. Column (3) in Table 7 shows the second stage results. We see that the results are robust to this alternative unit of observation. Reassuringly, the results are very similar to the parish level results in terms of both magnitude and significance.

Then, we consider estates as an alternative unit of observation. We show that the relationship between the elites and the cooperative movement is also present if we apply a similar strategy at the estate level. Using the estate as the unit of observation confers the advantage that we now compare estates with estates which are likely to be more similar units than e.g. parishes. In this way, we reduce heterogeneity between the units of observations, but of course end up with fewer observations. We estimate the model using a similar outcome variable as before, but the distance to the cooperative butter factories now being measured from each estate. We code a dummy for whether an estate used the Holstein System in 1782 and construct similar control variables as in the setups described above. Thus, the variation comes from the plausibly exogenous component in the establishment of a hollænderi on an estate. Column (4) in Table 7 shows the second stage results. We see a positive and significant relation between the presence of a hollænderi on an estate and the proximity of cooperative butter factories. The F-statistic is close to 9 and thus below the conventional rule of thumb. However, as noted by Cameron and Trivedi (2009), an F-statistic above 5 is also sometimes applied, and besides, for specifications with fewer controls, corresponding to columns 1–6 in Table 4, it is well above 10.Footnote 32 As a further test we added the Anderson-Rubin 95% confidence interval for weak IV in squared brackets. We found distance to Sofiendal to be a relevant instrument at the grid and parish levels above, so the weaker relationship here is likely due to the lower number of observations.

Further, we test the importance of regional fixed effects in our two stage least squares estimates as they could potentially inflate the estimated effects. Hence, we re-estimate our main results in Table 4 but exclude regional level fixed effects in all specifications. Table 8 presents the results of this test. Panel A shows the results without fixed effects while Panel B repeats the results from Table 4 for convenience. We see that while there is a tendency for the fixed effects to inflate our estimates they are very stable as covariates are subsequently included and importantly, all results remain highly significant at conventional levels. Thus, we conclude that our instrumental variable results are not driven by the inclusion of fixed effects for regions.

Table 8 Main results [T4] (IV, 2. stage) without fixed effects–cooperative creamery exposure 1890 and elites 1782

Additionally, we also consider an alternative instrument. Kjærgaard (1980) notes that Koppelwirtschaft was more likely to emerge in areas where the three-field crop rotation system was used. We digitize a map of historical field use (Frandsen, 1983), which indicates the use of two-, three-, four- or five-field crop rotation in Denmark for the years 1682–1683 (see Figure B1 in the online appendix). We then calculate the exposure of every parish to the use of three-field crop rotation by taking the sum of the inverse distances from the parish to every indicated use of the three-field system. We then employ this measure (Threefield (mp)) as an alternative instrument to the distance to Sofiendal. Table 9 presents the first- and second-stage results in columns (1) and (2). The instrument is relatively strong and confirms the positive effect on the emergence of cooperatives.

Table 9 Robustness, three-field system and cooperative creamery indicator

A valid concern might be that having market potential measures based on distances on both the left and the right-hand side of the estimation equation might lead to spuriously high correlations and significance. Column (3) in Table 9 thus shows a specification using an indicator variable for whether a parish had a cooperative creamery on the left-hand side instead of the market potential measure. Clearly, introducing a distance-based measure only on one side of the equation does not alter the results, although we prefer our main results for the reasons stated above.

To further take account of this concern, we test the relevance or the predictive power of distance to Sofiendal in explaining the geographical distribution of hollænderier across estates by randomly allocating the 224 hollænderier to the 791 estates for which we have full data coverage. We repeat this procedure 10,000 times and regress the random allocation of hollænderier on distance to Sofiendal and the full set of covariates to check whether the true distribution of hollænderier is better explained by distance to Sofiendal than a given random allocation. In Fig. 5a we plot the t values of distance to Sofiendal from these 10,000 regressions. The figure shows that distance to Sofiendal does indeed explain the true location of hollænderier significantly better. Only in 13 cases (p = 0.0013) did the random allocations outperform the true allocation. Thus, the instrument is not just capturing spatial noise. We see this as a further piece of evidence that the hollænderier did in fact spread in the way we suggest from Sofiendal to the rest of the country. Similarly, we randomly allocated the cooperative creameries 10,000 times between the parishes using the same distribution of creameries across parishes (one parish with three coops, 54 parishes with two coops etc.). Based on each random allocation of cooperative creameries we calculated the market potential for cooperatives and estimated the t value of elites 1782 in the second stage instrumental variable regression. Figure 5b shows that only in a few cases (p = 7/10,000 = 0.0007) were the randomly allocated cooperative butter factories better explained by hollænderier than the true allocation of creameries. We see this as an indication that the t-values were not just inflated by spatial autocorrelation.

Fig. 5
figure 5

Testing the relation between distance Sofiendal and the distribution of hollænderier a The gray bars show frequencies of t values of Distance Sofiendal from regressing (randomly allocated) elites1782 on Distance Sofiendal and the full set of covariates. The random allocation of the 224 hollænderier to the 791 estates was repeated 10,000 times. The vertical black line indicates the t value from the true distribution of hollænderier (t value =  − 2.98). Share of runs with |t^(random Elite) | >|t^(Elite 1782) |= 2.98: 0,0013. b The gray bars show frequencies of t-values of elites1782 from regressing (randomly allocated) MP for cooperative creameries on elites1782 and the full set of covariates. The random allocation was repeated 10,000 times and was created to mimic the true distribution of creameries across parishes (i.e. same mean and standard deviation). The vertical black line indicates the t value from the true distribution of elites1782 (t value = 2.98). Share of runs with |t^(random AM1890) | >|t^AM1890 |= 2.98: 0,0007

5 The cooperative butter factories, local economic development, and the persistence of cultural norms

The historical narrative indicates that the cooperative butter factories were an important factor in the growth trajectory of the Danish economy in the late nineteenth century, and that their link with Grundtvigianism, in association with folk high schools, played an important role for the development of Danish culture. We have stressed above that the cooperative butter factories were able to capture the UK market for butter, and the traditional narrative (e.g. Henriksen, 1993; Hyldtoft, 1999; O’Rourke 2006) also describes how agriculture stimulated demand for machinery for producing, among other things, centrifuges. Moreover, increasing demand from the peasantry as they became richer stimulated the development of a market for consumption goods. Services also benefitted due to increased demand for trade, transportation, and financial services. Finally, the cooperative butter factories used physical capital, such as automatic cream separators and steam engines, and arguably built social capital as they required local cooperation, both of which are regarded as proximate causes of economic growth (e.g. Acemoglu et al., 2005). Thus, there are good reasons to believe that the cooperative butter factories stimulated aggregate macroeconomic development in Denmark.

To further substantiate this, we have collected data on wealth tax revenue per capita (Statistics Denmark, 1905). These taxes were introduced in 1903 and first collected in the tax year 1904/1905. The tax data have been matched to the parish level, and we present OLS regression results for the association between our measure of the cooperative butter factories and the tax revenue per capita variable. The wealth tax was a flat rate of 0.6 per thousand without any allowance, but those earning below a certain threshold were exempt (Philip, 1955, 209), implying that that taxable wealth is arguably closer to actual wealth, and in fact we note that wealth per capita performs stronger in terms of statistical significance than data on income taxes and thus focus on it here.Footnote 33 The association between cooperative butter factories and wealth tax revenue per capita is displayed in Panel A of Table 10. Although the presence of the large estates might naturally correlate with the wealth measure in particular, the effect would be tiny, given that they were a miniscule proportion of the total population.Footnote 34

Table 10 Local economic development and cooperative creamery exposure 1890

Thus, as economic historians have long surmised, there is a strong association between the cooperatives and local economic development. The coefficient on the cooperative creamery market potential variable implies that log wealth per capita increases by 0.06 when the cooperative creamery variable increases by one standard deviation (Panel A, column 7 of Table 10). As the mean of log wealth is -0.16, this is arguably a large effect. Further analysis of this relationship is beyond the scope of the present work, and we do not claim a causal effect. Our instrument, distance to Sofiendal, would naturally violate the exclusion restriction, since it might have a direct impact on income and wealth which is independent of that of the cooperatives. Moreover, although it might be tempting to test for significance with income or wealth measures today, modern Denmark is a highly urbanized society, and it would be misleading to focus only on the rural economy, which anyway rapidly transitioned from a dependence on small cooperatives over the course of the twentieth century. In fact, as Panels B-E of Table 10 demonstrate, there is if anything a negative relationship between the location of cooperatives in 1890 and income per capita (and perhaps wealth) by 1937, and no clear relationship by 2019. Such results are clearly consistent with the work of Franck and Galor (2021), who found if anything a negative long run impact of industrialization in France on prosperity, and deserve a more detailed analysis.

Nevertheless, to the extent that the cooperatives represent the birth of modern Denmark, we believe that our analysis demonstrates the important role of agricultural elites for this process. Moreover, we can test for their impact today on at least two values associated with Grundtvig: a belief in democracy, and individualism. Such values might persist in the countryside, even after economic activity has moved to the urban areas. For the former, we use the obvious indicator, electoral participation, which we measure at the level of the polling station (valgsted), matched to the parish level for the last general election in 2015.Footnote 35 Panel A of Table 11 presents our results. We see clear evidence that people are more likely to vote closer to the areas which were mostly associated with cooperation in 1890. In fact, the cooperatives were almost always organized along Grundtvigian lines on the basis of one member one vote, and thus not based on the relative importance of the members, for example in terms of the amount of milk they delivered.

Table 11 Cultural persistence and cooperative creamery exposure 1890

We then turn to individualism, which in normal times might be difficult to measure on the very local level, with survey data such as the European Value Survey only disaggregated for the five regions of Denmark. Fortunately (in this respect at least), we have an alternative: participation in the Danish government’s mass COVID-19 testing strategy.Footnote 36 From early in the pandemic, the Danish government made free and easily accessible testing a major part of how it sought to ensure the fewest possible restrictions, allowing society to remain relatively open. According to calculations presented by Aburto et al. (2021), the strategy in general has to date been a success, with Denmark (together with Norway) being only two countries in their sample of 29 which did not experience drops in life expectancy. Recently, some studies have praised the Danish record during the COVID-19 outbreak, with the Economist ranking it first out of 23 rich countries.Footnote 37 At the time of writing, every person (man, woman and child) in Denmark has been tested on average more than 20 times, far in excess of any other country,Footnote 38 and moreover at central testing facilities rather than the home testing relied on in many other countries, such as the UK.Footnote 39 The vast majority of the population supported these efforts, but nevertheless, there is local variation. In contrast to vaccination,Footnote 40 testing confers no private benefit on the individual, and someone skeptical of government advice might be less likely to attend a test center regularly. For our analysis we rely on a period from the first week of November 2021 at which point the level of infections was relatively low, and there were no societal restrictions based on testing. Parish-level data for this period is published online by the State Serum Institute.Footnote 41 Controlling for the obvious confounders–distance to a test center, the infection incidence, and vaccination ratesFootnote 42–we show in Panel C of Table 11 a clear correspondence between the presence of cooperative butter factories in 1890 and participation in this program. Panels B and D give the same results, but without including the major cities, where such values are likely to be less persistent, given for example internal and foreign migration. As expected, similar regressions for vaccination rates show no clear relationship with cooperatives.Footnote 43 We interpret this as tentative evidence for a continuing Grundtvigian spirit and a belief in individual freedom in the Danish countryside.Footnote 44

6 Conclusion

How did Denmark ‘get to Denmark’? We have demonstrated that a reason for the extremely rapid spread of industrialized dairying in Denmark between 1882 and 1890 is the spread of innovations, starting with the introduction of the Holstein System, by large landowning elites from Schleswig and Holstein over the preceding century. We have described based on the contemporary literature how these innovations spread throughout the country and trickled-down to farmers beyond the large estates. Moreover, we have demonstrated empirically that areas with more hollænderier developed greater cow densities and were associated with folk high schools, revealing the spread of dairying and demand for education around the country, and that the initial wave of cooperation was in areas which had been so treated. We also show that the historical persistence of the elites still played a role in 1914 and that our results are stable under a number of robustness tests. Moreover, there is a clear correlation between areas with more cooperatives and higher local wealth by the early twentieth century. Finally, we demonstrate that the association between the butter factories and the Grundtvigian folk high school movement provided a cultural underpinning to modern Denmark, one founded on a belief in individual freedom and democracy.

On a less optimistic note, however, it should also be remembered that the process as a whole took well over a century. The institutions, technology, schools, etc. did not appear overnight, or within the first decade of cooperation. Farmers would not have known that their comparative advantage lay in dairying in the 1880s, and they would not even have had the cow densities for this to be the case, if the hollænderier had never existed. This has implications for understanding the reason why the attempt to transfer Danish-style cooperatives to other countries, such as Ireland in the 1890s (see e.g. Henriksen et al., 2015) and Iceland around the turn of the twentieth century (Jónsson, 2012), as well as to developing countries more recently, were relative failures. For more than a century, elites were initiating a whole package of reforms which eventually allowed the cooperatives to emerge and prosper. There has been a tendency to see cooperation as the solution to agricultural poverty, but this work suggests that this must be in combination with other reforms, which, at least initially, elites are probably better placed to implement than the peasants themselves. Finally, it should also be remembered that cooperative Denmark enjoyed easy access to a nearby, large and industrializing country, which prior to the First World War embraced a strong free trade ideology. Present day poor, agricultural countries, even if they were able to transform their agriculture through some of the mechanisms described here, would anyway find it difficult to export their produce to the protectionist rich world of the twenty-first century.