Lecture 27: Scientific Networks Prof. Eduardo A. Haddad
Introduction Universities and regional development Importance of universities in a regional innovation system Knowledge spillovers Evidences of spatial localization Scientific landscape in 1990-2010 Accelerated growth of production in emerging scientific nations (China, Brazil and India) Increase of scientific collaboration (over 70% of their production from domestic collaboration) Spatial scientometrics Analyzes spatial patterns of scientific activity The importance of geography in the process of creation and diffusion of knowledge 2
Scientific collaboration in numbers How to measure scientific collaboration? Co-authorship in scientific research Curriculum Lattes (CNPq) Main system of curriculum information for the Brazilian scientific community Information is publicly available But the process of data extraction and mining is complex Scriptlattes (MENA-CHALCO et al, 2013) Summary of database Total of CVs Total of scientific publications 1.131.912 7.351.957 3
Identification of major knowledge areas (Scriptlattes) Major Knowledge Areas (CNPq) Engineering Humanities Applied Social Sciences Linguistics, Letters and Arts Agricultural Sciences Exact and Earth Sciences Health Sciences Biological Sciences Other areas 4
Identification of geographic location (Scriptlattes) 4.616 cities 5
Identification of co-authorship (Scriptlattes) Bibliographical Production selected in CL Articles in scientific journals Expanded abstract published in proceedings of conferences Book published/organized Book chapter published Researcher A Researcher B Author A; Author B. Impacts of... Revista XX, São Paulo. 2007. Author A; Author B. Impacts of... Revista XX, São Paulo. 2007. Similar titles Coauthorship between researchers A and B 6
Full counting method Researcher A Researcher C Researcher D Researcher B Region X Region Y Region Z Region X Region Y Region Z Region X 1 2 2 Region Y 2 0 1 Region Z 2 1 0 7
Full counting method Region X Region Y Region Z............... Region X 1 2 2............... RegionY 2 0 1............... Region Z 2 1 0.......................................................................................................... Contains the total number of interregional co-authorships (symmetric matrix) Main diagonal: contains the total number of intraregional coauthorships Co-authorships associated to 1.347 Brazilian cities (dimension of matrix) 210 matrices of interregional co-authorships by year (1990-2010) and major knowledge areas (9 + total) 8
Scientific production in Brazil: 1992-1994 Scientific production measured by authorship in scientific publications 9
Scientific production in Brazil: 2007-2009 Scientific production measured by authorship in scientific publications 10
Location of public universities 11
Growth rate Increase of scientific production by knowledge areas 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% I-II II-III III-IV IV-V V-VI Engenharias Ciências Humanas Outros Ciências Sociais Aplicadas Linguística, Letras e Artes Ciências Agrárias Ciências Exatas e da Terra Ciências da Saúde Ciências Biológicas Total I - (1992-1994) II - (1995-1997) III - (1998-2000) IV - (2001-2003) V - (2004-2006) VI - (2007-2009) evidence of slowing-down of scientific production in Brazil 12
Proportion of total production Spatial deconcentration 100% 90% 80% 70% 60% 50% 40% 30% 20% 1992-1994 1995-1997 1998-2000 2001-2003 2004-2006 2007-2009 10% 0% 1,0 26,0 51,0 76,0 101,0 126,0 151,0 176,0 Number of cities (ordered by individual production) 90% of scientific production concentrated in 1992-1994 48 cities 2007-2009 102 cities 13
Knowledge flows in Brazil: 2007-2009 14
Scientific collaboration in Brazil Total of Scientific Coauthorships 1992-1994 547.249 2007-2009 9.445.399 Main links in 2007-2009 Interregional links Intraregional links Campinas/SP São Paulo/SP 76.716 São Paulo/SP 1.343.394 Ribeirão Preto/SP São Paulo/SP 74.078 Rio de Janeiro/RJ 630.049 Niterói/RJ Rio de Janeiro/RJ 75.224 Porto Alegre/RS 393.888 Rio de Janeiro/RJ São Paulo/SP 72.500 Belo Horizonte/MG 375.734 Seropédica/RJ Rio de Janeiro/RJ 65.348 Ribeirão Preto/SP 278.218 Porto Alegre/RS São Paulo/SP 47.343 Campinas/SP 212.248 15
Number of cities in the networks Expansion of domestic scientific collaboration 1400 1200 1000 800 600 400 200 0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 ENG HUM SOC LIN AGR EXA SAU BIO TOTAL 16
Co-authorship network: Agricultural Sciences (1990-2010) Capão do Leão/RS Pelotas/RS 53.451 Piracicaba/SP Campinas/SP 30.695 Lavras/MG Viçosa/MG 29.334 Viçosa/MG 370.118 São Paulo/SP 143.982 Lavras/MG 143.559 Individual degree in the network 17
Co-authorship network: Health Sciences (1990-2010) Ribeirão Preto/SP São Paulo/SP 95.120 Campinas/SP São Paulo/SP 78.023 Rio de Janeiro/RJ São Paulo/SP 66.852 São Paulo/SP 2.220.749 Ribeirão Preto/SP 426.694 Rio de Janeiro/RJ 388.479 Individual degree in the network 18
Role of geographical proximity Average Distance (for 105 main cities) 1992-1994 1995-1997 1998-2000 2001-2003 2004-2006 2007-2009 357,8 Km 387,6 Km 425,4 Km 490,5 Km 493,9 Km 465,2 Km Does space still matter for scientific interactions? Did this effect decrease over time? How to measure this effect? Dimensions of proximity Geographical, institutional, cognitive and social proximity Dimensions occur together 19
Spatial interaction modeling: Data scientific collaborations between regions i and j measured by co-authorships in scientific publications and : total of scientific publications in regions i and j measured by authorship in scientific publications : geographical proximity between regions i and j measured by physical distance (Kilometres) : institutional proximity between regions i and j dummy variable (1 if both regions have public universities) 20
Spatial interaction modeling for count data Nature of data Linear regression model Count Data Specification test Is there unobserved heterogeneity (overdispersion)? Negative binomial model Poisson model Specification test Specification test Zero-inflated Negative binomial model (ZINB) Negative binomial model Zero-inflated Poisson model (ZIP) Poisson model 21
Estimation results Poisson 1992-1994 1995-1997 1998-2000 2001-2003 2004-2006 2007-2009 Origem Destino (α 1 = α 2 ) 0,8212714*** 0,7795005*** 0,8220109*** 0,7858146*** 0,7973171*** 0,7885987*** (0,0799382) (0,0678417) (0,0683882) (0,0579629) (0,054490) (0,0578959) Distância Geográfica (β 1 ) -0,0019558*** -0,0019253*** -0,0017452*** -0,0015338*** -0,0015431*** -0,0017769*** (0,0003065) (0,0002531) (0,0002097) (0,0001606) (0,0001556) (0,0001774) Distância Institucional (β 2 ) 0,3644302*** 0,4162181*** 0,1609404*** 0,3136939*** 0,2940121*** 0,4287121*** (0,176917) (0,1436342) (0,1254923) (0,1061115) (0,1068666) (0,1163698) Constante (α 0 ) -7,857116*** -7,251477*** -7,874573*** -7,340407*** -7,597752*** -7,356414*** (1,304543) (1,149595) (1,215009) (1,060104) (1,034923) (1,116736) Binomial Negativo Origem Destino (α 1 = α 2 ) 0,8521451*** 0,8111029*** 0,7492518*** 0,7274424*** 0,7358830*** 0,6437665*** (0,0254626) (0,0215507) (0,0252036) (0,0219370) (0,0219320) (0,279855) Distância Geográfica (β 1 ) -0,0008051*** -0,0008671*** -0,0007708*** -0,000797*** -0,0008311*** -0,0008877*** (0,0000654) (0,0000471) (0,0000387) (0,0000035) (0,000392) (0,000359) Distância Institucional (β 2 ) 0,2046957*** 0,134085*** 0,2595981*** 0,1823897*** 0,0738276*** 0,2052725*** (0,0975923) (0,0765383) (0,0687016) (0,0634148) (0,0598206) (0,0589084) Constante (α 0 ) -8,848608*** -8,256593*** -7,296663*** -6,709063*** -6,811066*** -5,024858*** (0,2523885) (0,2435624) (0,340571) (0,2793517) (0,3146031) (0,4536341) Heterogeneidade (α) 6,08251* 5,08992* 4,562474* 3,848097* 3,618955* 3,750808* (0,1914479) (0,116746) (0,0913702) (0,0664739) (0,0644162) (0,0666601) Notas: i) n 2 =11.025 observações; ii) os erros-padrão estão entre parênteses; iii) ***, ** e * referem-se às estimativas estatisticamente significantes aos níveis de significância de 0,001, 0,01 e 0,05, respectivamente. 22
Interpretation of the results Characteristics of origin and destination regions Positive and statistically significant Institutional proximity Positive and statistically significant Geographical proximity Negative and statistically significant The value -0.0017769 means that for each additional 100 kilometers between two researchers, the probability for collaboration decreases by about 16,3%, holding all other variables constant Effect is not linear: 300 Km (600 Km) 41,3% (65,6%) probability for collaboration - Geographical distance still plays a crucial role 23
Results by knowledge areas The effects of geographical distance on scientific networks are different 24
Conclusion Overview of the evolution of scientific activity in Brazil (1990-2010) Accelerated growth of scientific production and collaboration But there are evidences of slowdown Spatial heterogeneity of scientific production Knowledge flows are concentrated in the Southeast and South regions Evidence of spatial deconcentration of scientific production and collaboration Scientifically less traditional regions gained prominence Role of geographical distance on scientific collaboration networks Distance still plays a crucial role in determining the intensity of knowledge flows Effects varies between knowledge areas 25
Key message
Reference The notes for this lecture were based on the following paper: Scholarly Publication and Collaboration in Brazil: The Role of Geography Sidone, O., Haddad, E. A. and Mena-Chalco, J., forthcoming in Journal of The American Society for Information Science and Technology 27
Questões prováveis 1. Em Tese de Doutorado defendida no IPE-USP no dia 24/11/2015, Ana Maria Bonomi Barufi concluiu que economias de aglomeração reforçam alguns dos mecanismos que geram concentração espacial e desigualdade regional. Discuta a conclusão da Dra. Barufi considerando aspectos relacionados às preferências locacionais de trabalhadores e firmas. 28
Questões prováveis 2. Fábrica fica pronta, mas Honda desiste de inaugurar por causa da crise Frustração. O anúncio do adiamento da fábrica da Honda foi recebido com tristeza e frustração em Itirapina. A expectativa era muito grande, pois a cidade não tem emprego, afirmou Mário Mroczinski, presidente da Associação Comercial. Os grandes empregadores aqui são dois presídios e a prefeitura, disse. Ele esperava que a instalação da Honda mexeria com a economia da cidade. Com ela, também chegariam empresas terceirizadas para atender a montadora. Estado de São Paulo, 30/10/2015 Discuta a frustração dos moradores de Itirapina à luz do modelo de base econômica. 29
Questões prováveis 3. Em reportagem da Revista Amanhã, em que se procurou investigar quais são os lugares mais atrativos do país quando se deixa de lado a guerra fiscal, para focar o que realmente seduz o investidor a competência e a vocação das regiões o Prof. Clélio Campolina Diniz, da UFMG, fez a seguinte observação: se for pela lógica de mercado, São Paulo tem condições de crescer mais que qualquer outro Estado. Você concorda com a afirmação do Prof. Diniz? Justifique sua resposta a partir das definições de economias de aglomeração e áreas de mercado. 30