Collaborative Knowledge Production in China: Regional Evidence from a Gravity Model Approach

Abstract

Scherngell T. and Hu Y. Collaborative knowledge production in China: regional evidence from a gravity model approach, Regional Studies. This study investigates collaborative knowledge production in China from a regional perspective. The objective is to illustrate spatial patterns of research collaborations between thirty-one Chinese regions, and to estimate the impact of geographical, technological, and economic factors on the variation of cross-region collaboration activities within a negative binomial gravity model framework. Data are used on Chinese scientific publications from 2007 with multiple author addresses coming from the China National Knowledge Infrastructure (CNKI) database. The results provide evidence that geographical space impedes cross-region research collaborations in China. Technological proximity matters more than geography, while economic effects only play a minor role.

Scherngell T. et Hu Y. La production en collaboration de la connaissance en Chine; des preuves régionales provenant d'un modèle de gravité, Regional Studies. Cette étude examine la production en collaboration de la connaissance en Chine d'un point de vue régional. On cherche à illustrer les tendances géographiques de la recherche en collaboration pour trente et une régions chinoises et à estimer l'impact des facteurs à la fois géographiques, technologiques et économiques sur la variation des activités de collaboration interrégionales au sein d'un modèle de gravité du type binomial négatif. On emploie des données sur les publications scientifiques chinoises de 2007 dont les adresses à auteur multiples proviennent de la base de données China National Knowledge Infrastructure (CNKI). Les résultats laissent voir que l'espace géographique fait obstacle à la recherche interrégionale en collaboration en Chine. La proximité de la technologie l'emporte sur la géographie, tandis que les retombées économiques ne jouent qu'un rôle secondaire.

Publication en collaboration Production de la connaissance en collaboration Modèle de gravité régional binomial négatif Régions chinoises

Scherngell T. und Hu Y. Kollaborative Wissensproduktion in China: eine empirische Analyse mit räumlichen Interaktionsmodellen, Regional Studies. Die vorliegende Studie untersucht kollaborative Wissensproduktion in China aus einer regionalen Perspektive. Zielsetzung ist es, räumliche Muster kollaborativer Wissensproduktion zwischen 31 chinesischen Regionen zu beschreiben und den Einfluss von geographischen, technologischen und ökonomischen Determinanten auf die Variation interregionaler Kollaborationsaktiviäten zu messen. Die Studie verwendet neue Daten aus der China National Knowledge Infrastructure (CNKI) Datenbank über chinesische Ko-Publikationen mit mindestens zwei Autoren aus dem Jahr 2007. Die Ergebnisse zeigen, dass die Kollaborationswahrscheinlichkeit signifikant mit zunehmender geographischer Distanz abnimmt. Der Einfluss von technologischer Nähe ist jedoch wichtiger als geographische Distanzeffekte, während ökonomische Unterschiede eine geringere Rolle spielen.

Ko-Publikationen Kollaborative Wissensproduktion Negatives binomiales Schwerkraftmodell Chinesische Regionen

Scherngell T. y Hu Y. La producción de conocimiento colaborador en China: ejemplo regional de un modelo de gravedad, Regional Studies. En este estudio investigamos la producción de conocimiento colaborador en China desde una perspectiva regional. Nuestro objetivo es ilustrar los patrones espaciales de las colaboraciones de investigación entre treinta y una regiones de China y calcular el impacto de determinantes geográficos, tecnológicos y económicos sobre la variación de las actividades de colaboración interregional según una estructura de modelo de gravedad binomial negativa. Los datos utilizados proceden de publicaciones científicas chinas de varios autores de 2007 que proceden de la base de datos de la Infraestructura de Conocimiento Nacional de China (CNKI). Los resultados demuestran que el espacio geográfico obstaculiza las colaboraciones de investigación interregionales en China. La proximidad tecnológica importa más que la geografía mientras que los efectos económicos desempeñan solamente un papel menor.

Copublicaciones Producción de conocimiento colaborador Modelo binomial negativo de gravedad regional Regiones chinas

Keywords:

• Co-publications
• Collaborative knowledge production
• Negative binomial regional gravity model
• Chinese regions

JEL classifications:

• L14
• O38
• R15

Acknowledgements

The authors gratefully acknowledge Manfred Paier (AIT, Foresight and Policy Development Department) and Bernhard Dachs (AIT, Foresight and Policy Development Department) for valuable comments that helped to improve this work; and Changyu Liang (ZTE) for assisting in developing the co-publication database and the collaboration matrices. Yuanjia Hu gratefully acknowledges Josef Fröhlich (AIT, Foresight and Policy Development Department), Yitao Wang (University of Macau), and Eurasia Pacific Uninet (EPU) in Austria for supporting this research.

Notes

Liang and Zhu (2002) used co-publication data to analyse the effect of geographical proximity by means of some simple correlations. They found that geographical proximity is one of the most important factors of cross-region knowledge flows. However, they did not control for other factors that might affect collaboration intensities. For instance, geographical distance may just be a proxy for technological distance (Fischer et al., 2006).

Research input, as captured by R&D expenditures, and research output, as captured by publications, in China show a smooth upward trend over time from 1997 to 2007 (National Bureau of Statistics/Ministry ofScience and Technology (NBS/MOST), 2008).

By focusing on publications with financial support, the scope of the study is delimited to a special form of co-publications that is, in the authors' view, a more suitable proxy for cross-region collaboration intensities since policy initiatives aimed at supporting research collaborations are taken into account. By this the indicator comes closer to recently used indicators for knowledge flows between European regions as captured by joint R&D projects in the European Framework Programmes (Scherngell and Barber, 2009, 2010).

A full counting procedure is followed, that is, a publication with n authors produces n(n – 1)/2 links that can be counted. For instance, for an article with three authors in three different regions, three links are counted: from region a to region b, from region b to region c, and from region a to region c. When all three authors are located in one region, three intraregional links are counted. Authors from the same institution are included when counting collaborations. They enter the calculations as intra-regional observations. However, bias in this direction does not play a role since the share of ‘same-institution’ collaborations is similar across all regions (between 27% and 31%).

Note that the n × n matrix is symmetric by construction (y _ij  = y _ji ).

Note that equation (1) is equally applicable to the sector-specific networks. They are not distinguished in the formal presentation.

The index of specialization calculated in this study is defined as:

1.

where y _ik is the share of publications in the scientific field k in region i; and is the respective average of China (for example, Hallet, 2000). The average specialization index is 0.09, that is, regions are highly diversified with respect to their publication intensity in different disciplines. The Herfindahl concentration index shows values between 0.05 and 0.07 for the six fields, that is, they are relatively equal distributed across Chinese regions.

The model described in this section is equally applicable to Y ^(agr), Y ^(eco), Y ^(it), Y ^(med), Y ^(nse), and Y ^(soc). They are not distinguished in the formal presentation.

Data for researchers, including those employed at any institution, are derived from the China Statistical Yearbook 2007 (National Bureau of Statistics (NBS), 2007).

Note that due to symmetry of the origin and destination variables, there is a special case with [α₁ = α₂].

According to Bröcker (1989), the intra-regional distance is calculated as:

1.

where A _i denotes the area of region i, that is, the intra-regional distance is two-thirds the radius of an presumed circular area. Note that the authors refrained from adding an intra-regional dummy variable to the model due to high correlation with the logarithm of the geographical distance variable .

Physical neighbours are defined as those sharing a common border. Note that intra-regional observations are not part of this reference group, that is, a region cannot be a neighbour to itself.

Gross regional product (GRP) data come from the China Statistical Yearbook 2007 (NBS, 2007).

These disparities – reported in a large number of empirical studies (for example, Liu and Jia, 2008; and Xu et al., 2005) – cannot be captured by the other covariates. The differences refer not only to economic disparities, but also to cultural, educational, social, and institutional ones. For instance, Walsh (2007) provides empirical evidence that the vast majority of foreign invested R&D centres are located along the eastern coast. This might attract researchers to collaborate with other researchers located in these regions. Thus, it is strongly assumed that it is important to account for a coastal area and central area effect in the model in order to avoid an omitted variable problem.

The coastal area includes the regions of Beijing, Fujian, Guangdong, Guangxi, Hainan, Hebei, Jiangsu, Liaoning, Shandong, Shanghai, Tianjin, and Zhejiang. The central area is composed of the regions of Anhui, Heilongjiang, Henan, Hubei, Hunan, Jiangxi, Jilin, Nei Mongol, and Shanxi.

The technological classes used correspond to the second-digit level of the IPC systems.

The additional parameter γ changes assumption (10) by:

1.

which is a natural form of over-dispersion in that the over-dispersion rate is:

1.

Note that when γ = 0, model (12) collapses to the standard Poisson specification as given by equation (8).

Another way to deal with the problem of unobserved heterogeneity is to use quasi-maximum likelihood estimation strategies. The authors prefer the negative binomial solution as it allows the likelihood ratio and other standard maximum likelihood tests to be implemented (Ismail and Jemain, 2007).

The Five-Year Plans of China are a series of economic development initiatives, for example, The First Five-Year Plan (1953–1957), The Tenth Five-Year Plan (2001–2005), and The Eleventh Five-Year Plan (2006–2010). The Grand Western Development Program for ‘the Tenth Five-Year’ (2001–2005) has been completed.