Regional inequalities in a learning economy: the case of Brazilian bioethanol production

This paper aims to analyze the regional inequalities of bioethanol production, focusing on learning and knowledge creation. The papers have two central assumptions: first, learning is the key resource in modern capitalism; second, it is unevenly distributed in the space. For capturing this phenomenon in bioethanol production, we divided the Center-South region into three: the traditional regions of production in the State of São Paulo; the new expansion regions in the western of São Paulo; and the complementary regions in other federative units. We divided the knowledge and learning in three dimensions: the worker experience, the worker educational degree, and the academic research about sugarcane. The results show the growth of the accumulation of experience and the improvement of labor education in the western of São Paulo and the complementary regions, but a concentration of academic publications in the traditional regions.

the production process" (LUNDVALL; JOHNSON, 2016, p. 134). We argue that identifying the geography of knowledge is a key-point for contemporary analysis of regional development insofar as different regions have different competencies.
This paper aims to analyze the regional inequalities of bioethanol production, focusing on learning and knowledge creation. Is there a regional division of labor where the knowledge is generated in the traditional regions of São Paulo, and only the production moves into the countryside? Alternatively, is there a growth of both production and knowledge in the new regions of expansion?
The paper focuses on the Center-South region of Brazil 5 . For the analysis, we divided this region into three parts: the traditional regions of productions; the expansion regions in the state of São Paulo; and the complementary regions in other federative units. To shed light into the questions raised, the paper makes use of two Brazilian databases: the Annual Relation of Social Information (RAIS) to obtain data related to the workers qualification in these regions; and the Brazilian Lattes Platform to obtain regional data about the number of publications and research efforts and build proxies for knowledge generation.
This study is in line with the idea that "Brazil's success with sugarcane cannot be understood solely as deriving from a natural comparative advantage. Above all, it is the result of a virtuous trajectory of technological learning strongly based on incremental innovations" (FURTADO; SCANDIFFIO; CORTEZ, 2011, p. 156). Our analysis can contribute to understanding the spatial dimension of the innovation process in the sugar-energy sector and provide ideas that can support investigations about the limits and potential that bioethanol has for regional development.
The paper is divided into more four parts besides the introduction. The first one is a theoretical discussion about the idea of regional development based on agricultural production in a learning economy. The second one presents the methodology to justify the regionalization and the data we used. The third shows the results. Finally, the last section presented our conclusions.

The geography of knowledge in agricultural and agroindustrial systems
There is a classical discussion in the economic literature about the potentialities of agriculture to induce regional development. In the historical debate, there are both a group of authors who associate agricultural growth with underdeveloped structures (FRANK, 1969;FURTADO, 1972;MYRDAL, 1957) and authors who defended that, under certain conditions, agriculture can be a prime influence for regional economic growth, inducing external economies, urbanization, and industrial development (NORTH, 1959). Walker (2004) presented a contemporary perspective of this discussion in studies about the agricultural zone of California. Page & Walker (1991) analyzed the agro-industrialization as a motor of regional development in the American Mid-West region (PAGE; WALKER, 1991). These studies showed how agricultural growth could create a complex social division of labor. These are cases where the region economy increased as a result of a combination of local interactions of farming production, processing and input industries, and universities that develop scientific research according to the local demands. In this context, according to Walker (2004, p. 15), modern agriculture "requires, above all, human knowledge and practical labor skills, and advances in technology are commonly made through human learning." The sentence is in line with the perspective that knowledge is a crucial resource in modern capitalism (LUNDVALL, 2016). Sustained growth depends on the creation of local capabilities. As Lee (2009, p. 1) pointed out: "without a certain critical degree of capabilities, growth, which is based on lower wage rates or simple price competitiveness, tends to be short-lived." We follow the idea of Lundvall and Johnson (2016) that more than a knowledge-based economy, we live in a leaning and dynamic economy.
The learning economy is a characteristic of a period of capitalist development that emerged in the 1970s with the information and communication technological revolution 6 (PEREZ, 2009). Lundvall and Johnson (2016) define the changes at this moment with three interconnected phenomena. The first is related to the radical innovations of information paradigm -personal computers, software services, telecoms, and the internet (PEREZ, 2009, p. 189). On the one hand, it "has drastically reduced the costs of handling, storing and moving information" (LUNDVALL; JOHNSON, 2016, p. 109). Nevertheless, the efficient use of information remains a costly process that depends on learning. The second is the change from a Fordism to a flexible specialization model. It increases the possibilities of the division of labor. Therefore, it also increases the necessities for communication and cooperation among firms, universities, and even competitors. The third phenomena are the changes in the innovation process that demand a continuing and incremental advance that emanate more interactive learning (LUNDVALL, 1988) for the survival of firms.
A combination of long-distance and proximity relations defines the geography of knowledge and learning in the technological paradigm of information (STORPER, 2013). Proximity relations are associated with the tacit dimension of scientific knowledge (POLANYI, 1966) and the necessity of faceto-face interaction (STORPER; VENABLES, 2004). It leads to a spatial concentration of sophisticated knowledge (BALLAND; RIGBY, 2017). Long-distance relations, on the other hand, are complements to absorb external and codified knowledge. It is a mechanism to diversify the knowledge base and avoid the risk of local lock-in (BOSCHMA, 2005).
The approach of regional innovation systems combines interactive learning and proximity relation (ASHEIM; GERTLER, 2006; ASHEIM; SMITH; OUGHTON, 2011; COOKE; URANGA; ETXEBARRIA, 1997). According to Asheim & Gertler (2006, p. 299), "the regional innovation system can be thought of as the institutional infrastructure supporting innovation within the production structure of a region.". Among the main interactions able to create and induce the abortion of knowledge in regions, we highlight the relationship between firms, universities, and research centers (LESTER, 2005). These interactions are central in the creation of human capital and to transform traditional agriculture into a modern activity (SCHULTZ, 1965).
According to Perez (2010), one of the effects of a technological paradigm is its "capacity to transform profoundly the rest of the economy (and eventually society)." Therefore, it is essential to understand agriculture in the information era and within a learning economic context as changes in the techno-economic paradigm enter in different spheres of agricultural production. The use of information technologies in agriculture dates from the 1980s. Tractors using Global Positioning System (GPS) became available at the beginning of the 1990s, making possible the use of precision agriculture. This method was fundamental to manage the use of fertilizers, pesticides, and water in agriculture (BŐGEL, 2017). The adoption of this technology is almost simultaneous in Brazil (CASTILLO, 1999). Besides the use of technology to improve the knowledge about natural conditions, information technologies created the possibility of processing and storage data, also enhancing the management and technical decisions of farmers.
In order to discuss the agriculture and agro-industrial systems in the learning economy, we use the distinction of Lundvall and Johnson (2016) of four types of knowledge: know-what, know-why, know-how, and know-who. Table 1 presents the original definition of Lundvall (2016) and our proposal to apply this perspective on agriculture.
Chart 1: Types of knowledge in agricultural and agroindustrial systems

Know-why
Knowledge about principles and laws of nature, the human mind, and society. Science-based knowledge.
Agrarian sciences knowledge. General university valuable knowledge to agriculture.

Know-how
Refers to skills. Experience-based learning, Learningby-doing.
The experience acquired on the farmingspecialized knowledge created about a specific crop in the region.

Know-who
Information about who knows what and who knows what to do. Ability to cooperate and communicate.
The capability of identifying the leading agents and their competences in the agricultural innovation system.

Source: Org. by the authors
As we may conclude from Chart 1, although all types of knowledge are combined in the process of innovation, we argue that some of them are more critical to regional development. The know-what and the know-who are more equivalents to reproducible information. Nevertheless, the use of information and the capacity of regions to create and absorb scientific knowledge depends on the know-why and know-how. It is essential for more autonomy of a region to execute innovation efforts. In agriculture, know-why is central to adapt the production to the regional particularities regarding both natural and economic conditions. The know-how complements this capacity. It shows how a region develops its abilities and expertise in agricultural production through new methods, practices, and organizational dynamics. It is an essential dimension for the creation of tacit knowledge in the regions.
Thus, in order shed light on the question raised in the Introduction about knowledge generation and learning capabilities and how they unevenly distributed in the territory, we will focus on the know-why and know-how associated with sugar-energy sector.

Regionalization of the sugar-energy industry in Brazilian Centre-South
The first step to analyze the regional capabilities of knowledge creation in the sugar-energy productive system was to propose a regionalization of sugarcane production in the Brazilian Centre-South area. We used a group of meso-regions, defined by the Brazilian Institute of Geography and Statistics (IBGE), to create our regionalization.
We considered three groups of regions using as the main condition the historical moment of sugarcane growth. The region 1 is the traditional regions of São Paulo, being the group of Piracicaba, Ribeirão Preto, Araraquara, Bauru, and Assis. Region 2 is the western region of São Paulo, being the group of Araçatuba, São José do Rio Preto, Presidente Prudente, and Marília. Region 3 is the complementary region of sugarcane expansion. It means the group of the South of Goiano, in the state of Goiás; the Triângulo Mineiro/Alto Paranaíba, in the state of Minas Gerais; East de Mato Grosso do Sul, and the Southwest of Mato Grosso do Sul in the state of Mato Grosso do Sul; and the Northwest Paranaense, Center-North Paranaense, and Pioneer North Paranaense in the State of Paraná. Map 1 presents the regionalization used for the data analysis.
Map. 1: Regional division of sugarcane production Source: Org. by the authors

The analysis of the sugar-energy sector: measuring the know-why and know-how
Based on Belik, Perosa & Fredo (2014), we focused our attention on the sugar-energy sector as the sum of four sectors at a 4-digit level of disaggregation from the National Classification of Economic Activities (CNAE 2.1) 7 : sugarcane crop (0113), sugar production (1071), sugar refining (1072), and ethanol production (1913). The first is agricultural production. The others refer to the milling. This choice allows us to capture direct and indirect effects on the regional occupational structures.
We used the public version of the Annual Social Information Report (RAIS). This database considers microdata on formal employment, but, as the levels of formality in this sector are increasing, we can consider that our sample is representative enough.
In order to analyze the know-how in the sugar-cane sector using RAIS, we will use the worker experience in months within the firm. In the sugar-energy sector, it shows the creation of specific skills and capabilities of producing sugarcane according to local conditions. Each local has its particularities in terms of climate, soil, and the interaction of sugarcane variety with the environment (MESQUITA, 2016). Therefore, if the workers are keeping more time in the mill, it is assumed that this type of knowledge, essential for sugarcane productivity, is increasing.
To analyze the know-why in sugar-energy sector, we aim to discuss how scientific knowledge is created and applied to the sector. One of the central agents for this is universities (FURTADO; SCANDIFFIO; CORTEZ, 2011). As Garcia et al. (2014) pointed out, universities are essential agents for the innovation process due to two fundamental roles: the training of the labor force and the creation of new knowledge in the research activity. Therefore, we divided the know-why capacity into these two categories.
For the labor qualification, we analyze, from RAIS, the evolution of schooling or level of formal education. We divided the data into five groups: illiterate, up to elementary school completed, up to high school completed, college incomplete, and at least college completed (which include workers with MSc. and Ph.d. titles).
To analyze research knowledge created about sugarcane, we use the Lattes Platform. To obtain all papers about sugarcane, we filtered the papers that have the following words in the title: cana-de-acucar or cana de acucar or sugarcane.
At first, it showed a total of 6,340 paper and 7,229 authors (including also co-authors). However, excluding the authors who did not identify their professional address, we reached a total of 5,369 papers and 4,193 authors. This data was combined with our regional division using the researchers' professional address. We also arranged the top universities in sugarcane research. The

General characteristics of region 1, 2 and 3
Before analyzing the regional differences related to knowledge production, this item aims to present a general view of the sugar-energy sector in Regions 1, 2, and 3. Table 1 shows different data related to these regions in the years of 2003, 2010, and 2017.  Considering the sum of region 1, 2 and 3, Table 1 also indicates geographical deconcentration of sugar-energy sector in these regions both in employment and sugarcane area. It shows that the traditional region is reducing its share compared with the new regions of production in west São Paulo and the frontier regions of the state. Map 2 illustrates this process considering the total employment within the sugar-energy sector by the municipality. Although this data indicates a dispersion of production and employment, the knowledge and learning process follows a different geography.

Regional inequalities: the "know-how" dimension
We used the worker experience as a proxy for absorptive capability and tacit knowledge. While the first one is necessary for learning, the second one is the output of the learning-by-doing process, and time plays a central role in this process of experience accumulation. Figure 1 shows a kernel density estimation of the distributions of workers experience in months, for every macro-region (red) and the sugar-canesectors (blue). Median values are represented by the vertical lines, respectively. Horizontal axes are in log10 for a better visualization.
As observed for 2003, median time experience is around 24-25 months (10 1,37 − 10 1,39 ), with a higher concentration of workers to the left for region 1 and to the right for region 3, which also presents a strong bimodality. Mean values are also close among regions, with approximately 60, 56, and 46 months on average for regions 1, 2, and 3, respectively. Thus, the skewness of the distributions and central statistics put in evidence the presence of workers with very high experience 8 .
In 2010, the sector in region 1 presented a shift to the right, which implies a higher survival rate of workers, which is also reflected in the central moments of the distribution. A bimodal pattern also appears in regions 1 and 2 although there was no positive change on the means and averages of the sectoral distributions of regions 2 and 3. On the contrary, dispersion raised in region 1 and decreased in regions 2 and 3. With all this and considering that we are working only with what we defined as permanent workers, we can infer the emergence of two workers profiles: the first as the one settling at the sector and the second one as the most experienced.
In 2017, we observed a shift to the right of the sectoral distributions and the weakness of the bimodal pattern. Means and medians had a significant increase in all the regions, with region 1 leading (mean 86 and median 58), followed by region 2 (mean 66 and median 49) and region 3 (mean 59 and median 47), which also puts in evidence a path-dependence effect in the sense that regions 2 and 3 are the "new" ones in the expansion process. Together with better qualification and a demand for less manual activities, we expect a positive impact on sectoral performance. The horizontal axis is in log10 for better visualization of the tails. It is also important to notice the higher medians of the sugar-energy industry if compared to the median worker experience within the regions. Overall and differently from the sectoral dynamic, region 2 is the one who presents a higher mean and median worker experience among the three regions.  Table 2 shows the composition of the labor force within the regions and the sugar-energy sector by the level of formal education (schooling). The percentage of illiterates is diminishing consistently in region 3, although the sector presents a higher share of illiterates if compared with the regions. From the three regions, it is region 2 that appears with the higher percentages of formal workers that at least completed high school or have at least college in every year analyzed. On the other hand, the share of more qualified workers within the sector is lower than the ones observed in the respective regions. Notwithstanding, there is a change in the sectoral occupational structure once there is a constant increase in the participation of higher qualified workers. It certainly brings benefits to the performance of the sector given that there is a whole set of capabilities that can be acquired only by formal education and that are also central in the process of knowledge generation, the development of absorptive capacities and innovation.  Table 3 reinforces what we observed in Table 2, in the sense that it highlights the changes in the demand for more qualified occupations. More specifically, it shows a higher demand for professionals within the Exact and Biological Sciences (that include engineers and agronomists, for example). Indeed, even though the shares appear to remain equal or even diminishing over time, the increase in absolute values is significant and higher than the increasing pattern observed within the regions. For example, in region 1, while the raise on the demand of these occupations is about 87% from 2003 to 2017, the rise observed within the sector is about 249% for the same period. Region 2 presents a raise of 83% while the sector within that region presents a raise on this demand of around 452%. Finally, the change in the same demand within the sugar-energy industry is around 684% against the 160% observed within the region. Finally, we believe that it brings evidence in favor of a structural change within the sugar-cane industry. However, if this change is positive for the sector and the region, it should be reflected in the sectoral performance. For capturing this, we observe the average wage per hour worked, which is presented in Figure 2. The kernel density estimations for regional labor productivity is depicted in blue, whereas the sectoral estimates are in red. Horizontal axes are in log 10 for better visualization of the tails. Region 1 has the higher average wage per hour worked (45, 53 and 62 approx. in 2003, 2010 and 2017, respectively), followed by region 2 (40, 48 and 57) and 3 (35, 45 and 56), although variability appears higher in regions 2 and 3.

4.3.Regional inequalities: the know-why dimension in the labor force
The sugar-energy sector has a higher average wage per hour worked in regions 1 and 2 if compared with the whole region. Asymmetric distributions show the heterogeneity, with a right fat tail indicating that there is a significant part of workers year with productivity lower than the average and some with higher productivity. For example, if we consider 2017, and a confidence interval of plus/minus one standard deviation, we notice that the higher productivity observed is 247, 10 and 17 times higher than the lower one observed within the sector. Also, in 2017, the sector presents its median productivity shifted to the right in all the regions. Social and economic benefits will be higher in region 3 given that we are working with wages that affect the welfare of the families directly and that it was that region that presented the major increase on average values, going from 28 reais per hour to 62 on average. 9 To finish our analysis let us observe another dimension of the know-why, this time more related to the knowledge generation and research activities.

4.4.Regional inequalities: the know-why dimension in research activities
The publications evolution about sugarcane, under regional and temporal perspectives, allows a stratified understanding of the generation of knowledge in the sector. According to Table 4, of the total of 5,639 publications on sugarcane, approximately 60% of the publications occurred between 2011 and 2017, which indicates a significant expansion of the publications in the second period. Even though the distribution of publications between regions still shows small changes in the share of each region over the total published over the years. Evidence suggests that there were no significant changes in the participation of each region in the generation of knowledge about sugarcane in Brazil.  The concentration of the scientific production presented in Table 4 is a result of the size of the universities where papers authors work. As Table 5 shows, the universities of the state of São Paulo appear among the institutions with the most significant number of publications, reinforcing the importance of the state for the generation of knowledge about sugarcane. In region 1, it is important to highlight the role of Escola Superior de Agricultura Luiz de Queiroz (ESALQ) located in Piracicaba, the Universidade Federal de São Carlos -headquarter of the (Inter-University Network for the Development of Sugarcane Industry -Ridesa, in São Paulo, located in Araras, and the Instituto Agronômico de Campinas, mainly because of the Center of Cana located in Ribeirão Preto. Region 1 also benefits for its geographical proximity with the Universidade de São Paulo, Universidade de Campinas and the Centro Nacional de Pesquisa em Energia e Materiais (CNPEM). The data of Universidade Estadual Paulista Julio de Mesquita Filho is aggregated, which makes it difficult to identify the precise location of publications. Nevertheless, considering that this university has courses of Agrarian Sciences in Jaboticabal and Botucatu, both in region 1, it also shows a concentration process. Regarding region 2, we did not find any representative institution publishing papers about sugarcane. The exception is the campus of Universidade Estadual Paulista Julio de Mesquita Filho in Dracena, that has a course of Agrarian Science. Our hypothesis is that that region 2 benefits from the knowledge infrastructure of region 1. In region 3, there are three representative institutions: Universidade Federal de Uberlândia (16º in the ranking), Universidade Estadual de Maringá (18º) and Universidade Estadual de Londrina (20º).

Conclusions
The paper analyzed the regional inequalities in the sugar-energy sector, considering the growth of bioethanol production after the 2000s and the territorial dispersion of mills and sugarcane production. The theoretical approach is associated with the idea that modern capitalism is based on a learning and knowledge economy. For capturing this, we divided the Centre-South region into three regions. We also divided the knowledge effects into three dimensions: the know-how, the know-why related to labor qualification and the know-why related to research activities.
The know-how shows evidence of regional structural changes in the sugar-cane industry, reflected in the occupational structure. At this point, the most important gain resulted from worker experience is in region 2. Regarding the know-why related to labor qualification, region 2 also appears with the higher percentages of formal workers that completed high school or have at least college. In this part, both region 2 and 3 have more percentage of workers with a superior education than region 1. The know-why related to publications show an inverse situation. Region 1 has by large the most representative knowledge infrastructure, which reflects in its dominance of publications.