Group farming in France: Why do some regions have more cooperative ventures than others?

The global debate on food security and the kinds of farming systems that could prove economically and ecologically sustainable has focused overwhelmingly on small family farms versus large commercial farms, with little attention being given to alternative models based on farmer cooperation. France offers a significant but under-researched and internationally little-recognized model of group farming – the GAEC (Groupement Agricole d’Exploitation en Commun) – based on farmers pooling land, labour and capital. This model is of considerable contemporary interest for both France and other countries. Catalysed by a 1962 law, GAECs accounted for 7.6% of farms and 15% of agricultural adult work units in 2010, but their incidence varied greatly across regions. Using data from the French agricultural census and other sources, this paper identifies the factors – economic, ecological, social and demographic – underlying this regionally uneven development of GAECs (and comparatively of EARLs – Exploitations Agricoles à Responsabilité Limitée – another type of group farm introduced in 1985). Regions with a higher incidence of group farms are found to be those that were historically dominated by middle-sized farms, had a local ecology favouring labour-intensive animal breeding, especially pastures, a higher proportion of agricultural graduates, greater economic equality and social institutions that promote community cohesion, among other factors. These results illuminate not only the conditions favourable to the emergence of group farming in France, but also the conditions under which such farmer cooperation could take root in other (including developing) countries, subject to context-specific modifications of the French model.


Introduction
In recent years, an increasing weight is being placed internationally on small family farms as vehicles for enhancing food security, equity and ecological sustainability (FAO, 2014;HLPE, 2013). This emphasis brings together two dominant features of farms today: most are family run -globally at least 88% of all farms are assessed to be family farms (FAO, 2014); and most are small in size -an estimated 84% of farms across 111 countries operate under 2 hectares (FAO, 2014;see also HLPE, 2013). 1 Critics who see such farms as nonviable present large commercial farms as the alternative (Collier and Dercon, 2014). Both models, however, leave serious concerns unaddressed. On the one hand, small family farms face significant constraints linked with scale, fragmentation and resource access, which are difficult to overcome. On the other hand, large commercial farms have low prospects of providing jobs either to the vast numbers of people still dependent on agriculture, or to new entrants to the rural labour force (Dorin, 2017;Dorin et al., 2013;Imai et al., 2017). Clearly, we need to think about other models of farming. In this context, group farming, wherein farmers pool their land, labour and capital and share costs and profits, could be of particular interest. This model, which has received rather little attention either in policy or research, is based on voluntary cooperation, without farmers forfeiting their private property rights, in sharp contrast to the collective farms created through forced collectivization under socialist regimes.
The lack of attention to voluntary group farming is surprising, given that there are contemporary examples in many countries. In France, for instance, group farming has existed at least since the 1960s, and today there are over 100,000 group farms. Elsewhere, joint farm enterprises began to emerge in the 1980s and 1990s, such as for cattle upkeep and milk production in Norway (Almas, 2010) and Ireland (Macken-Walsh and Roche, 2012), and crop production, especially rice, in Japan (Sarkar and Itoh, 2001). It also emerged in many post-socialist countries after de-collectivization of agriculture in the 1990s, including in East Germany, Kyrgyzstan, Romania and Nicaragua. 2 Since the early 2000s, it has been promoted successfully by state governments and civil society in several parts of India . Reasons for the neglect of this alternative model are likely to lie in the adverse experience of collectivization under socialist regimes; 3 the poorly designed and hence largely unsuccessful efforts to promote cooperative farming in the 1950s and 1960s in the post-colonial, newly independent developing countries of Asia, Africa and Latin America; 4 and, most of all, in the pessimism embedded in economic theory on the possibility of people cooperating, given tendencies to free-ride (see Olsen, 1965, among others).
In recent decades, however, a substantial body of empirical work has demonstrated that communities are successfully cooperating for governing common-pool resources, leading to improved conservation outcomes (Agarwal, 2010b;Ostrom, 1990). Service cooperatives, especially for marketing, sharing machines, or procuring inputs, are also widespread globally. 5 But cooperation for governing common-pool resources differs from cooperation for agricultural production, which involves complex forms of daily interaction. Similarly, marketing cooperatives do not require everyday cooperation in the production process itself. For instance, family farms can produce milk, meat or crops individually, while using cooperatives only for selling these items. In contrast, pooling land, labour and capital for agricultural production requires what Agarwal (2014) terms 'multipurpose comprehensive cooperation'. This is much less common, and even less studied.
A study of group farming, based on such multipurpose cooperation, acquires particular relevance today, given that people in many countries (both developing and developed) are seeking diverse pathways to carve out viable livelihoods within agriculture. This includes not only existing farmers looking for more lucrative and sustainable farming options, 6 but also new job seekers who (as noted) have limited non-farm outlets, as well as those who want to make agriculture a lifestyle choice. 7 In this search for alternatives, it would be especially timely to examine farming models based on cooperation in production.
Here the experience of France is of particular relevance. It has perhaps the oldest and numerically the most important example of voluntarily constituted and legally enabled group farms, as embodied in GAECs (Groupements Agricoles d'Exploitation en Commun) and, to some extent, in EARLs (Exploitations Agricoles a`Responsabilite´Limite´e). France can thus provide particular insights on the characteristic features of these group farming models and the contexts in which they emerge. This is not only because of their long existence and numerical strength, but also because of their regional clustering: GAECs are much more concentrated in north-western and central-eastern France, than elsewhere (see Figures 1 and 2). This begs the question: do regions with more GAECs have features  especially conducive to cooperation, such as the local economy, ecology, demography, culture or a mix of these and other factors? An analysis of these factors is of interest both in itself and to help us understand which underlying factors are more conducive to the formation of group farms. This understanding could also provide lessons for the potential adaptability of such models beyond France.
The questions we pose have been little researched, especially the factors underlying the regional clustering of group farms. Our paper thus breaks fresh ground. It will also contribute to ongoing academic and policy debates on economically viable agriculture, by its focus on farming models based on cooperation. The paper is divided into six sections. The next section provides a background to GAEC and EARL formation and an overview of existing studies, following which there is a discussion of the broad characteristics of GAECs and EARLs based on agricultural censuses from 1988 to 2010. We then present our model and hypotheses regarding the factors that could explain the greater incidence of group farms in certain regions over others in 2010. The penultimate section presents our regression results, followed by concluding reflections and policy pointers in the closing section.

Background and existing studies
Legally, GAECs were catalysed in France by a law passed in 1962, which became fully effective in 1965 (Raup, 1975). The law specified a legal minimum of 2 partners/associates and a maximum of 10, with the requirement that all partners work full-time on the farm. The law also incorporated a 'transparency principle' (Article L323-13 of the French Rural Code) under which the state, for its agricultural support programmes, would treat each GAEC partner as an individual entity, while recognizing too a GAEC's collective identity. This principle enabled GAEC partners to benefit from public incentives on the same basis as individual farmers, including when the Common Agricultural Policy (CAP) of the European Union (EU) introduced direct income support 'decoupled' from price support in 1992. 8 Also in 2013, the EU officially recognized the transparency principle for GAECs. 9 Socially, however, GAECs were propelled especially by the Jeunesse Agricole Catholique (JAC), 10 an association of young Catholic farmers, 'guided by the conviction that a "third road" was needed between what was regarded as the abuses of capitalism and the excesses of Marxian collectivism' (Raup, 1975: 3; see also Boussard, 1991). The farmers, supported by some intellectuals and high-level civil servants, are said to have convinced the French government that an enabling law was needed. Their efforts fell on fertile ground, since the government, led by key figures such as Edgard Pisani (Minister of Agriculture, 1961-1966, 11 was seeking to modernize family-based agriculture, and the GAEC was seen as an institutional innovation close in structure to a family farm, in contrast to a corporate farm dependent largely on hired employees. Forming a GAEC was expected to help individual family farms improve their managerial efficiency, productivity, and work conditions (GAEC & Socie´tie´s, 2010a). 12 For instance, by pooling their resources, even farmers with limited means could modernize their farming techniques and organization; experiment with new technologies that needed too much capital for an individual farmer to afford; take advantage of scale economies; and free children or spouses from agricultural work to seek higher education or non-farm jobs. Moreover, in recognizing sons and fathers as equal partners, GAECs enhanced the status of sons within the family, and thus their incentive to work harder and increase productivity. Farmers' wives, however, remained unequal, both economically and socially (Darque, 2008). Under the law, married couples could not become associates without the presence of a third associate, until a change in law in 2010 allowed spouses to constitute couple GAECs.
In 1985, another type of group farm -the EARL -was legally instituted, basically as a limited-liability society, subject to different laws and procedures than GAECs. 13 For example, unlike GAECs, which required at least two associates, EARLs could also be formed by one person as well as by married couples. Moreover, in EARLs only shareholders who individually or as a group hold the majority capital are required to work on the farm, while minority shareholders need not. GAEC associates, in contrast, cannot take up any significant income-earning activity outside the GAEC. 14 Notwithstanding these differences, it is important to study EARLs, since those with 2 or more associates (the maximum can be 10) constitute a type of group farm, requiring the associates to make capital investments and cooperate. In fact, sometimes EARLs change into GAECs or vice versa (GAEC & Socie´te´s, 2010a). Today, many GAECs have sustained for long years (one with six associates is over 50 years old: Copex, 2012), and have moved on to another generation of associates.
These farming models, based on close cooperation among associates (especially in GAECs), are of considerable contemporary interest for France and other countries. Apart from the potential advantages of group farming already listed, there are others, such as the ability of group farms to provide viable livelihoods to those (especially the young) who lack adequate land or capital to farm alone, or who like a rural life and are willing to cooperate on a daily basis for the less strenuous work life possible in a GAEC compared with individual farming. In a GAEC, for example, associates can substitute for each other in labour-intensive or monotonous tasks and take holiday breaks. Resource pooling can also allow associates to invest in larger capital-intensive machines, expand farm size, diversify farm activities or specialize in labour-intensive work such as animal breeding, and enlarge the range of skills and knowledge beyond those found in one farmer or one family. The small number of associates who know each other well can also help overcome the classic problems of free-riding and work shirking through peer-vigilance and mechanisms for enforcing accountability, such as weekly meetings and management committees.
Notwithstanding their relevance, long duration and numerical strength, however, there is rather little systematic and rigorous research on GAECs. On the quantitative side, although the French agricultural census provides periodic data on the numbers, locations, composition and characteristics of GAECs, these data (which are comprehensive from 1988) appear to have been little-used analytically, beyond the Ministry of Agriculture's own descriptive briefs (see various issues of Agreste Primeur and also . The studies that do use the census -such as those investigating farm size inequality across departments (Piet et al., 2012), or those preparing a typology of farm size and type of labour used (Bignebat et al., 2015) -fail to separate individual and group farms, thus conflating a crucial characteristic of French farms. In fact, most existing research on GAECs is qualitative, and even this is limited in scope. Between 1965 and 1988, for example, the archives of GAEC & Socie´te´s list 29 master's theses on the subject, but most are sociological studies of single GAECs or writings on GAEC law. In this period, there was in fact a fascination with the history of GAEC formation. For instance, a doctoral dissertation by an American anthropologist who researched 42 GAECs (not selected systematically) spread across France in the mid-1970s provides historical insights on how GAECs emerged and functioned (Murphy, 1977). There are also occasional research papers on the administrative, legal and incentive provisions that encouraged GAEC creation (Raup, 1975), or on how family farms evolved into GAECs (Bazile and Viallon, 1985;Madec, 1983;Reboul, 1977), as well as some specialized monographs, such as on the involvement of GAEC associates in the social movements of the 1970s (Alland and Alland, 2002). In fact, this preoccupation with the legal, philosophic, historical or social aspects of GAECs continues, with empirical exploration being limited to a few GAECs, as evidenced by more recent journal articles (e.g. Barthez, 2007;Chandellier et al., 2012;Foyer et al., 2012), as well as two special issues of Revue de Droit Rural (2012).
These writings are insightful, but they leave many questions unanswered, not only about the characteristics of France's group farms and changes in them over time, but particularly about the geographic variations in their incidence across France. Explaining this variation through a rigorous empirical analysis, in order to better understand the conditions that would be conducive to cooperation in production, is the central concern of this paper, and one which has not been addressed so far in the existing literature.

Characteristics of farm structures and changes over time
To map the range of French farming enterprises over time, we use department-wise data from the French agricultural censuses of 2010, 2000(Ministry of Agriculture, 2014. 15 We complement this with data from other sources, to quantify the explanatory variables outlined in the next section. Our analysis is confined to 92 of the 96 departments of metropolitan France: the four excluded departments (Paris, Hauts-de-Seine, Seine-Saint-Denis and Val-de-Marne) are highly urbanized and have few farms. In this section, we describe the broad characteristics of France's group farms and their evolution, before examining their regional variation in the following sections.

Farm types and shifts over time
In 2010, individual farms, GAECs and EARLs respectively constituted 69.4%, 7.6% and 16.0% of all farms (Table 1). Notably though, almost half of the EARLs had only one associate, making them effectively similar to individual farms except for their legal status. Since our interest is in understanding farmer cooperation, we will concentrate on GAECs, with some comparison with EARLs that have two or more associates (herein called EARL ! 2).
In the 22 years between 1988 and 2010, the number of all farm enterprises fell by more than 50%, mainly due to the dramatic decline in individual farms (Table 1). In contrast, between 1988 and 2000, GAECs and EARLs increased notably, in both numbers and proportions, after which EARLs continued to increase numerically, but GAECs showed a slight decline. Since 2010, however, even GAEC numbers have been rising. According to GAEC & Socie´te´s, 2296 new GAECs were registered in 2014 alone, many of them being couple GAECs which were by then legally permitted. 16 In proportional terms, the numbers of both GAECs and EARLs rose between 1988 and 2010, with GAECs doubling from 3.7 to 7.6 as a percentage of all farm enterprises. According to GAEC & Socie´te´s (2010a: 10), there is a strong correlation between the increase in GAEC numbers and government incentives to encourage young people to settle in agriculture (DJA or Dotation Jeunes Agriculteurs). 17 But other factors, which cannot be explored here, may also underlie these shifts.

Social composition of group farms
Although the majority of GAECs continue to be constituted only of family members (84% in 2010), proportionally family GAECs have been declining and non-family or mixed GAECs have been increasing ( Table 2). Among family GAECs, those constituted by father and son fell from 59% in 1988 to 28% in 2010, while same-generation GAECstypically of brothers or cousins -doubled. This suggests that GAECs are moving towards equality among associates (since father-son GAECs contain an implicit social hierarchy, notwithstanding their equal legal status as associates).
Group farms are also becoming important in terms of annual work units (AWUs) in agriculture, with 1 AWU being equivalent to one adult working full-time for a year on the farm (termed 'Unite´de Travail Annuel' or UTA in France). 18 In 2010 (the year of the last French agricultural census), GAECs accounted for 15% of AWUs and EARL ! 2 for 12.7% of AWUs, together providing 27.7% of total AWUs (Table 3). This is a marked rise from 8% in 1988, while the figures for individual farms fell from 82.6% to 44.1% during  (Table 3). Moreover, if we take average land area, measured here in terms of 'Surface Agricole Utile' (SAU) or utilized agricultural area, 19 GAECs are substantially larger in area than other types of farms and grew faster than them between 1988 and 2010. In 2010, GAECs managed 147.6 ha on average, relative to 95.2 ha managed by EARLs and only 34.0 by individual farms (Table 3). 20 Notably too, we found that multi-generational family GAECs and GAECs of both family and non-family associates tended to be the largest, going up to 180 ha on average. A larger number of associates enables farms to diversify their activities and expand in area, while also making such expansion necessary for an adequate income per associate.

Activity specialization
Most important for our discussion, group and individual farms differ notably in their product specialization. In 2010, almost one-quarter of all farms specialized in seasonal crops (mostly cereals and oilseeds), 42% in animals, and 21% in horticulture or gardens, plantations and vineyards, with the rest doing mixed farming (Table 4). However, these proportions varied significantly by farm type. Over 65% of GAECs were involved in animal rearing (with 51% breeding meat and/or milk cattle), and only 10% were producing seasonal crops. EARL ! 2 came in between these, with the main activity again being animal farming, practised by 47%, followed by crops (22%). 21 This activity pattern suggests a link between GAEC formation and farm specialization. Rearing animals tends to be much more labour-intensive than growing crops, since animals need daily care (feeding, milking, etc.) even with some mechanization. Moreover, this intensity continues throughout the year and, unlike crops, is not limited to seasonal peaks. Hence we would expect farmers who want to rear animals to veer towards group farming. Also, we might surmise that one of the factors underlying the concentration of GAECs in some regions over others could be ecology, which favours a particular type of specialization. For instance, animal farming is often the best option in many mountainous or semi-mountainous areas (as, say, in parts of central-eastern France), where crop cultivation tends to be less profitable. Here, difficult work conditions could also encourage cooperation. 22 Similarly, regions with more permanent pastures would provide fertile ground for forming GAECs, although, technically, intensive livestock breeding is also possible through stallfeeding. Consider, for instance, Figures 3 and 4. Figure 3, which gives the regional spread of all farms specializing in animal breeding, 23 reveals an interesting overlap between animal farming regions and the GAEC concentrations noted from Figures 1 and 2: both show concentrations in western and eastern France. In turn, we would expect animal breeding to be linked to permanent pastures. Figure 4 gives us permanent pasture land as a percentage of SAU. Here, the overlap with GAEC incidence and animal farming is broadly consistent in the eastern parts but not in the western. In western France, as in Bretagne, there is little pasture land, and here the animals are largely stall-fed, although many still grow fodder for feed.
We also explored and found an interesting relationship between the regional concentration of pastures and regional changes in GAEC numbers over time. In Table 1, we had noted that in absolute numbers GAECs increased considerably between 1988 and 2000, and then decreased between 2000 and 2010, returning broadly to the 1988 figure by 2010. To see if this held across departments, we disaggregated these changes further, mapping departments where GAECs had increased in both periods; decreased in both periods; or had moved in divergent directions between the two periods. Figure 5 shows a notable similarity to      Table 1 shows that the GAECs specializing in animal breeding are the ones that have increased in absolute numbers between both periods (with the exception of stall-fed breeding), while the GAECs specializing in crops and horticulture/plantations have decreased in numbers in both periods, suggesting a particular link between GAEC formation and pasture-dependent animal farming. This relationship between the regional incidence of GAECs and the regional concentration of permanent pastures (and hence animal farming) is further borne out by our regression results.
One additional point needs to be mentioned here, which is difficult to capture statistically but is likely to have a bearing on changes in GAEC incidence over time, namely a snowballing effect. We would expect that in regions where GAECs have been historically successful, a demonstration effect would lead more farmers to see GAECs as alternatives to individual farms and form one themselves. Also, over time, such regions would stimulate investment in support services such as for refrigeration, milk processing and storage, animal slaughter and marketing, which newcomers forming GAECs could take advantage of. Similarly, farmers who have been exposed directly to GAECs formed by their parents or relatives or neighbours are more likely to form or join GAECs themselves than are those totally unexposed to such enterprises. All these factors could intensify a regional clustering of GAECs over time. We get some support for this phenomenon in the observed strengthening over time of regional clusters around animal breeding, for which the support services mentioned above are likely to prove much more important than for crop cultivation. In other words, a snowballing effect could vary across activities. This could be one among other factors underlying the changes in group farm numbers between 1988 and 2010, but exploring these factors further is beyond the scope of this paper.

Explaining regional variability: hypotheses
We draw on our conceptual and historical understanding of French farming to identify potential explanatory factors for the observed regional variability of GAECs and EARL ! 2 in 2010. Some factors could be economic, others ecological, social/cultural or demographic. We use an ordinary least squares (OLS) regression model (specified below) to test for the significance of the identified factors. The dependent variable Y is the percentage of GAECs (or EARL ! 2) to total farm enterprises in a department: where d represents departments (there are 92), X i denotes the explanatory variables (detailed below and in Appendix Table 2), {b 0 , b i } are the parameters to be estimated, and R represents the 13 regions we have controlled for.
The explanatory variables are the following: • economic ( Below we outline our hypotheses relating to these variables, first for GAECs and then briefly for EARL ! 2.

Economic factors
We examine two economic variables to explain the regional (departmental) incidence of GAECs: (1) farm size categories dominant in the department in 1970 (that is, fairly soon after the GAEC law became effective); and (2) income inequality as measured by the Gini index. 24 In relation to pre-existing farm size categories in a department, normally we would expect the presence of a large number of small farms to encourage group farming, since resource pooling would help non-viable farms to create a viable economic unit. This trajectory may not play out for GAECs, however, since GAEC associates are legally required not only to pool their resources (each has to bring a share of capital or land), but also to work full-time on the farm without seeking outside employment, or running individual farms on the side. This means that the GAEC must provide associates with at least a basic income. Moreover, the idea that all farms (not just GAECs) should be large enough to yield a minimum income for the family was embedded in the French government's 1975 decree, under its Rural Code (article 188), which specified the minimum surface area (Surface Minimum d'Installation (SMI)) for settling in agriculture. 25 These factors are likely to reduce the chances of very small farms combining to form GAECs. Rather, we would expect departments with a high percentage of small farms in 1970 (the earliest year for which there are data after the GAEC law became effective) to be less likely to have formed GAECs.
Departments with a high percentage of very large farms in 1970 would also be less likely to form GAECs, since they would not need to do so to tap economies of scale, although they may form one for other reasons, such as engaging in multiple economic activities. In short, departments with a high percentage of farms in the very small and very large categories in the early period of GAEC formation may be expected to have lower proportions of GAECs today. It also means that the relationship between farm size categories and GAEC incidence is likely to be non-linear. In this context, Raup's (1975: 21) observations for the 1970s are especially notable: he observed that GAECs tended to be concentrated in areas dominated by medium-sized farms, and were rarer in areas dominated by either very large or very small farms. He did not, however, statistically test this observation, as we will be doing.
To test the effect of a pre-existing farm size distribution on GAEC formation, we used data from the 1970 agricultural census. Although it does not give the size of each farm, it gives us the proportion of farms in different farm size categories (in SAU units) in the department. Our hypothesis is that the higher the proportion of farms in a department in 1970, in either the <20 ha category or the !50 ha category, the lower is likely to be the proportion of GAECs to total farms in 2010 in that department. In contrast, the higher the proportion of farms in the 20-49 ha category in 1970, the higher is likely to be the proportion of GAECs to total farms in 2010.
With regard to our second economic variable, we would expect GAECs to thrive more in regions of relative economic equality, since GAEC formation is likely to need the prevalence of a spirit of cooperation. Indeed, equality among associates is an important principle in GAEC formation and functioning. Also, group homogeneity is often noted to be more conducive to cooperation than heterogeneity (Baland and Platteau, 1996). Hence regions with high inequality could adversely affect GAEC formation. We measure inequality by using the Gini index (Gini coefficient Â 100) of the department's per household taxable income.

Ecological factors
We would expect local ecology to play an important intermediary role in GAEC formation. As noted earlier, some regions of France are less suited to crops, such as ecological zones with poor land quality, marginal pasture areas and mountain terrain characterized by low crop yields, leaving animal breeding on pastures as the main livelihood option. In addition, the CAP has regulations restricting the conversion of permanent pastures to crop land. 26 For example, the CAP reform of 2003 linked the maintenance/use of permanent pastures to subsidies, which would be reduced if the pastures were converted to cropland (see e.g. Beaufoy et al., 2011). Moreover, the reform specified that land under permanent pasture in a given reference year be maintained. In France, the reference year was 2003 and the regulation was enforced from 2005 (Desjeux et al., 2007: 19). 27 In many regions, therefore, ecological and related conditions encourage milk or meat farming over crops or other activity. Livestock breeding, in turn, needs high labour inputs throughout the year, which would encourage GAEC formation, since GAECs can induct more adults to share the work. There would therefore be a greater likelihood of GAEC formation in regions with more permanent pastures (and hence greater potential for livestock farming). We use the percentage of SAU land under permanent meadows and pastures in a department as an explanatory variable, expecting it to be related positively to the incidence of GAECs in the department.
A region's irrigation potential can also affect activity choices. Crop cultivation is more dependent on irrigation than other farm activities, and one crop -maize -alone accounts for half of all surface area irrigated in France (Barraque´et al., 2010). Farmers in regions with a higher percentage of irrigable land are thus more likely to cultivate crops, and hence less likely to form GAECs. We use irrigable area as a percentage of total SAU to test this effect. Notably, only 9.8% of SAU land in France was irrigable in 2007 and only 5.8% was actually irrigated around that time. 28

Social factors
The importance of social norms and relations of trust and reciprocity in laying ground conditions conducive to cooperation is now widely recognized in collective action theory. In the early years of GAEC formation, catholic priests and religious associations, especially the JAC, played an important role in establishing social norms that emphasized community, cooperation and benefit sharing (Murphy, 1977). Rene´Colson, a French farmer, became the general secretary of JAC in the early 1940s and mooted the idea of small farmers pooling resources to invest in machinery, and practice modern agriculture through cooperation. 29 In 1951, he founded an organization with his colleagues, which evolved into GAEC & Socie´te´s 40 years later (GAEC & Socie´tie´s, 2012: 5).
Catholic priests worked with JAC to encourage young farmers to form GAECs (Murphy, 1977;Raup, 1975). To capture these social effects on GAEC formation, we used the number of active (non-retired) Catholic priests in a department in the early 2010s as a proxy measure to explain the regional incidence of GAECs. In today's France, of course, religion is a less cohesive force, and other types of institutions -political bodies, farmers' unions, and similar associations -can generate social capital, cohesion and support. But we lack data to empirically measure the impact of these diverse institutions, which may also pull in different directions.

Demographic factors
We would expect at least two demographic factors to affect the geographic incidence of GAECs: availability of educated youths specialized in agriculture and the gender composition of the agricultural workforce. The first is likely to matter especially due to the link between having an agricultural degree and getting state support. For instance, formal training in agriculture at the secondary level or above has been essential (at least since the early 1990s: Rogers, 1991: 150) for getting the young farmer subsidy (DJA) that is offered by the French government and the EU (see also endnote 17). Although the DJA is not linked to any type of farm (group or individual), a young farmer could use it more effectively by becoming a partner in a GAEC than by setting up an individual farm on his/her own, since this could require additional investments. The importance of formal training is strengthened further by the necessity (barring exceptional circumstances) for farming establishments to have professional agricultural skills in order to receive national and European incentives. 30 A degree can also help farmers acquire the skills to manage bigger farms, which need more complex accounting procedures. Moreover, without an agricultural qualification, farmers face greater restrictions on buying or renting farm land. 31 Overcoming such restrictions matters more to GAECs, since they have a greater need than individual farms to expand farm size in order to provide for all the associates. Overall, therefore, in regions where a larger percentage of students are specializing in agriculture, the potential for constituting GAECs is likely to be greater.
The gender composition of the agricultural workforce is also likely to matter, given the noted legal restriction until 2010 on wives becoming GAEC associates with husbands, unless there was a third associate. This restriction was much debated over the years. Some argued (unsuccessfully) that excluding spouses was neither pragmatic nor conducive to the efficient organization of a farm (e.g. Foyer et al., 2012). Others felt that the restriction would free wives from hard agricultural labour and allow family holidays, which would prove difficult if both spouses worked on the same farm. In 2010, this restriction was legally removed (GAEC & Socie´te´s, 2010b), but since we are using 2010 data it is too early to catch the impact of this change. For our analysis, we expect that the larger the percentage of female farm workers in a department, the lower would be the proportion of GAECs.
What about EARL ! 2? We may expect the effects to be similar to GAECs on some counts but different on others. In terms of divergence, first, since not all EARL associates work on the farm, EARLs are less likely to be affected by the labour intensity of an activity (in particular cattle versus crop farming), and therefore by whether a department is dominated by permanent pasture or crop land. In fact, EARLs may even favour crops since they need less labour than animal upkeep (in other words, the relationship could even be negative). Second, related to crops, we would expect the availability of irrigable land to be positively related to the incidence of EARLs. Third, since EARL ! 2 could be formed by couples even before 2010, and given that associate wives are not required to work in an EARL, we do not expect the percentage of women working in agriculture to notably affect the incidence of EARL ! 2.

Regression results
In our regressions, we examine the impact of the above-mentioned explanatory variables on the two dependent variables at the department level: (1) the percentage of GAECs out of total farm enterprises; and (2) the percentage of EARL ! 2 farms out of total farm enterprises. The analysis is based on 2010 data, unless another year is mentioned. Exact definitions of the explanatory variables are given in Appendix Table 2. All equations are adjusted for regional fixed effects based on the administrative classification of France into 13 regions in the early 2010s. The regression results are presented in Table 5 and the  summary statistics in Appendix Table 3.
First consider the results for GAECs (Table 5, equation 1). Of the two economic variables, the outcomes for farm size are particularly interesting. We find that the higher the proportion of 20-49 ha farms and the lower the proportion of !50 ha farms in 1970, the greater the share of GAECs to all farms in 2010. Departments with a 1 percentage point higher proportion of 20-40 ha farms (relative to <20 ha farms) in 1970 have a 0.15 percentage point higher share of GAECs in 2010. In contrast, departments with a 1 percentage point higher proportion of !50 ha farms (relative to <20 ha farms) in 1970 have a 0.11 percentage point lower share of GAEC farms in 2010. In other words, the regional incidence of GAECs in 2010 rises with an increase in the proportion of 20-49 ha farms but falls with an increase in the proportion of !50 farms in 1970, while departments dominated by <20 ha farms come in between. In short, it is in departments with a predominance of farms which were neither too small nor too big in 1970 that we find the highest proportions of GAECs four decades later (as was observed for an earlier period by Raup, 1975); and the relationship between farm size and GAEC formation is non-linear. 32 Our second economic variable, the Gini index of household taxable income, is again consistently and negatively significant, as hypothesized. Departments with higher levels of income inequality have a significantly lower proportion of GAECs relative to all farms. On average, departments with a one point higher Gini index have a 0.4 percentage point lower proportion of GAECs. Of the two ecological variables, however, although the signs of the coefficients for both variables are in the directions hypothesized, only the pasture variable is significant: the higher the percentage of permanent pasture to total farm land in the department, the greater the proportion of GAECs to other farms, while irrigable area is negative but insignificant. Our regression results for pastures thus support and supplement the observations in the previous section.
The social variable is again significant. The greater the number of active priests in a department, the greater the percentage of GAECs to other farms, although the coefficient is small. The demographic variables have an important impact, however. Departments with a higher proportion of students specializing in agriculture have a significantly higher percentage of GAECs, and those with proportionately more female farm workers have a significantly lower percentage of GAECs. A 1 percentage point rise in the proportion of students in agriculture raises the share of GAEC farms by 0.35 percentage points, while a 1 percentage point rise in the proportion of women among farm workers lowers the share of GAEC farms by almost 1 percentage point (equation 1). Overall, our model explains 82% of the regional variations in GAEC proportions. Now consider the EARL ! 2 results (Table 5, equation 2). Here we see interesting differences with GAECs. To begin with, the effect of farm size distribution in 1970 on the incidence of EARL ! 2 in 2010 is (as with GAECs) positively significant for the 20-49 ha category, but unlike GAECs it remains positively significant even for the !50 ha category. Departments with a 1 percentage point higher proportion of 20-49 ha or !50 ha farms in 1970 (relative to <20 ha farms) have a 0.05 percentage point higher share of EARL ! 2 in 2010. The difference between the coefficients of the 20-49 ha size group and the !50 size group is also significant. In other words, it is departments dominated earlier by medium to large-sized farms that have the higher shares of EARL ! 2 in 2010, while a dominance of <20 ha farms in a department in 1970 was a deterrent to EARL ! 2 formation. 33 It is likely that the differences between EARL ! 2 and GAECs in the farm size results arise because EARLs from the start could be managed as large corporate entities with a few associates, whereas with GAECs large farms required several associates, especially with labour-intensive animal production, and this could make coordination more difficult. In recent years, however, mechanization has moved apace even in milk farming, with the introduction of robots for milking, and GAECs engaged in animal breeding too can grow to quite large sizes.
The results for GAECs and EARL ! 2 also differ on other counts. For instance, for EARL ! 2 although the Gini index is negative, it is not statistically significant, nor is the number of active priests. The percentage of SAU under pasture is negatively significant for EARL ! 2 (while for GAECs it was positively significant), and the percentage of SAU irrigated is positively significant for EARL ! 2, whereas it was negative and insignificant for GAECs. These results are in keeping with our earlier observations that EARL ! 2 farms are more likely to undertake crop cultivation than GAECs, while animal farming (linked with pastures), with its high labour intensity, is less likely to be favoured by EARL ! 2 farms, understandably since (unlike GAECs) they cannot depend on the guaranteed labour of associates.
In addition, for EARL ! 2 (in contrast to GAECs), the impact is insignificant for both proportion of students in agriculture and proportion of females among farm workers. That gender is not significant is not surprising, since unlike GAECs (as noted earlier) married couples could form EARLs even before 2010. Across all the variables, the coefficients of the explanatory variables for the EARL ! 2 equations are also much smaller than for the GAEC equations. However, our model explains 81% of the regional variations for EARL ! 2.
Overall, therefore, the departments that tend to have a larger percentage of GAECs are those that were dominated by middle-sized farms historically (rather than very small or large farms); whose ecological conditions are more favourable to animal farming (especially the percentage of land under permanent pastures); whose economic and social conditions are conducive to cooperation (less economic inequality and a greater presence of institutions that promote community cooperation); and which have particular demographic characteristics (a greater presence of students in agriculture and a lower presence of women among farm workers). For EARL ! 2, economic inequality matters less, and although farm size distribution is important, the effect is different from GAECs at the higher ranges. Also, while the ecological variables are significant, it is in the direction opposite to that for GAECs, and the demographic variables are insignificant. On gender, given the change in GAEC laws, the difference between GAECs and EARL ! 2 farms is likely to disappear in time.

Concluding reflections and policy pointers
What lessons can we draw from this analysis for France, and for regions beyond France? For France, our analysis suggests, first, that group farming will tend to find more fertile ground in regions that have less overall economic inequality, a larger percentage of farms historically in the lower to middle size range (rather than very small or very big) and more social institutions that promote community cohesion/cooperation.
Second, cooperative ventures are more likely to be sought for those agricultural activities that require intensive labour inputs on an everyday basis, such as cattle breeding (for milk or meat). Hence group farms are more likely to emerge (or take root more easily if promoted) in ecological zones that have a high incidence of permanent pastures, or where other types of farming are less profitable or less possible, as in harsher mountain areas and in zones with poor-quality land. The reverse is likely in regions that have favourable conditions for crop cultivation due to, say, soil type or access to irrigation. Since crops require less intensive labour inputs on a regular basis than animal breeding, they can also be managed effectively by individual families (or associations like EARLs, which have few working associates), with peak requirements being covered by hired labour and machines. Here there would be less incentive to undertake group farming.
Third, demographic factors matter, such as the incidence of agricultural graduates or of women farm workers in a region, but these factors can also be subject to legal conditions (as in France), such as whether an agricultural degree is needed to access farm subsidies, or whether spouses alone can form an association.
The incidence of farm enterprises needing lower levels of cooperation than a GAEC, such as EARL ! 2, are likely to be affected by somewhat different factors. Economic inequality and social and demographic factors appear not to matter in the same way, but ecological variables such as pastures and irrigation do matter, depending on the choice of farm specialization and hence intensive labour needs.
These observations stem from our results based on existing models of group farming. Variations on these models may emerge, however, if some of the legal requirements for forming GAECs were relaxed. For instance, GAEC associates need to work full-time on the farm and pool all their productive resources. This requirement implicitly dictates a certain minimum farm size. If associates were allowed to undertake supplementary income-earning activities at an individual level, such as managing their own farms in addition to the group farm, or allowed to seek part-time non-farm work, then even small farms could pool their resources and farm collectively, while also pursuing other earning opportunities. Of course, in such models it would be especially important to set in place mechanisms for monitoring each associate's work contribution, in order to ensure that the work is equitably shared and the likelihood of free-riding is minimized.
Beyond France, say for other parts of the EU, the factors found significant in our studyespecially economic and ecological -are general enough to be relevant. Also, EU laws (such as those pertaining to permanent pastures or subsidies for young farmers) which have played an enabling role for group farming in France would be applicable to other countries in the Union.
For developing countries, however, it is more difficult to extrapolate directly from our results, since their laws governing land use and tenancy and the structure of subsidies tends to be quite different. Even so, it could be argued that efforts to promote group farming are more likely to be received favourably in regions of lower economic inequality, and where the local ecology and economy favour labour-intensive farm activity, such as animal breeding. Many of the other advantages of group farming mentioned earlier, such as labour sharing, viable farm size, resource pooling for capital-intensive investments and skill diversification would apply as well. The GAEC model also points to a potential pathway for women farmers and junior family members in the Global South to move away from simply being unpaid workers on family farms, since it allows spouses, siblings and adult children to become equal business partners in a group farm. At the same time, models that allow a combination of group farming and individual activity (rather than those which only allow group activity) are likely to be more relevant for developing countries, where most farmers cultivate very small farms of a hectare or less, and therefore need to diversify their livelihood portfolios to earn a decent income.
Finally, looking ahead, group farming can provide an important third model of farming, beyond family farms and corporate agriculture, not only for farmers who are seeking better economic and ecological outcomes, but also for newcomers interested in community-based options, agroecological practices and building social and solidarity economies around food and farming systems. GAECs emerged in France as a result of young farmers seeking a third path, away from the negative features of both capitalist and socialist farming, and a government seeking to modernize agriculture within the ambit of families and communities. These motivations are as relevant today, and not just in one country or region but increasingly more globally.