Introduction

Knowledge is deeply connected to individuals and its replication or transfer to others is argued to be much more tacit most of the time (Metcalfe, 2005). While information technologies have presented greater means of communicating information over distances, whether they can help with knowledge flows over distances in each and every case has long been questioned. The first contact among those seeking new knowledge or in need of expertise (Törnqvist, 2004) or differences in types of knowledge (Asheim & Gertler, 2005; Asheim et al., 2007) constitute some of the major dynamics underlining the need for face-to-face contact and importance of close spatial distance.

In this context, geographical proximity has been discussed in terms of its contributions to the co-production of knowledge for the last three decades with growing interest over the years. While some have noted its decreasing influence over the years (Grossetti et al., 2016), its continuing positive role has been identified in multiple settings and scales as well, which range from cooperation among firms (Autant-Bernard et al., 2007; Broekel & Boschma, 2012; Cooke, 2008), industries (Basile et al., 2012), university-industry relations (Balland & Boschma, 2022), universities (He et al., 2020; Olmeda-Gómez et al., 2009), regions (Basile et al., 2012; Capello & Caragliu, 2018; Caragliu & Nijkamp, 2016; Hoekman et al., 2010; Sidone et al., 2016) or cities (Andersson et al., 2014). As also documented by most of these studies, however, aspatial dynamics indeed play no smaller role (Boschma, 2005).

Under such scrutiny, production of scientific knowledge came to be increasingly questioned as a process which consists of some of the most intense knowledge flows and holds distinct dynamics and values (Asheim et al., 2007; Metcalfe, 2005; Törnqvist, 2004). In the formation process of these values, the scientific domain has evolved from a system previously driven by researchers’ spontaneous interactions to interactions increasingly structured by institutions, national policies and international research networks over the years (Adams, 2013). In this process, scientific collaborations have become the main driving force behind the rapid increase in the output of new knowledge as observed from similar performance levels in individual productivity over the same period (Olechnicka et al., 2019).

In this context, the study questions the role of spatial and aspatial proximities and their interaction effects in influencing the quantity of scientific coauthorships among Turkish provinces during the period 2015 to 2019. Nonrationality and context are decisive elements for knowledge flows (Cooke, 2008), which makes analysis of these proximity effects in different cases a point of interest as the Turkish context is a rarely studied example. Beyond the exploration of proximities’ roles in the less studied case of Turkey, the study contributes to the exploration of proximities’ interaction effects as few studies have explored proximities’ interrelations and combined influence (Cao et al., 2019; Capello & Caragliu, 2018; Plotnikova & Rake, 2014). What further differentiates the case from existing studies is its observation of a rapid expansion policy, called “(at least) one university per province” taking place after 2006. Insight into such a rapidly evolving geography of collaborations contributes valuable experiences towards the formulation of expansion policies, especially for developing countries. Towards these aims, the study first reviews the literature on scientific collaborations and the influence of spatial and aspatial proximities on them. Next, a brief evolution of the national scientific system in Turkey is presented in periods with emphasis on the geographical dimension and changes in the final period. In the next section, data, variables and models are presented. Lastly, results are presented and discussed before conclusions are drawn towards future policy decisions.

Literature review

Coauthorships, as ‘real networks’ indicative of scientific collaborations (Autant-Bernard & Massard, 2000; Yan & Ding, 2012), have not only increased quantity of publications (Jones et al., 2008; Olechnicka et al., 2019) but also their quality. This quality improvement can emerge from increased scrutiny and unique combinations of knowledge through the involvement of more co-authors depending on the fields (Franceschet & Costantini, 2010; Thelwall et al., 2023). However, it may also stem from the wider outreach of a study through a larger number of coauthors’ networks, which simply leads to more citations regardless of a study’s actual quality (Thelwall et al., 2023). Nevertheless, their outputs have gained more impact compared to their single authored counterparts (Wuchty et al., 2007). In this context, while science is increasingly being internationalised, national systems still constitute the main grounds for these collaborations (Hoekman et al., 2010). This case is especially stronger for scientifically emerging countries where domestic collaborations constitute the larger share (Maisonobe et al., 2016). Dynamics behind the formation of these coauthorships and geographical proximity have been studied in the last three decades and recently grouped under the term spatial scientometrics (Frenken et al., 2009). The three interests of spatial scientometric studies so far have been to discover the reasons behind the collaborations; to measure the influence of proximity and size effects on chances to collaborate; to observe the evolution of world collaboration networks (Maisonobe et al., 2016). The first two reasons constitute the main interests of this study in this regard.

The role of geographical proximity in scientific collaborations’ emergence and continuity has been contested since the early studies. While the early studies highlighted the important role of close geographical proximity in researchers’ collaborations (Beckmann, 1994; Beckmann & Persson, 1998), those in the early 2000s have argued otherwise due to the rapid adaptation of ICT (Rosenblat & Mobius, 2004; Teasley & Wolinsky, 2001). While several studies have highlighted geographical proximity’s ongoing positive contribution, arguments exist that this influence is getting weaker over time (Grossetti et al., 2016). Three prevalent reasons for the spatial concentration of scientific collaborations have been (1) encounters by chance; (2) lower costs for managing face-to-face interactions (which at the same time assumes there is still a need for face-to-face interaction); (3) operating under a common institutional environment, specifically to further enable cross-country collaborations (Frenken et al., 2009).

Beyond geographical proximity, cognitive, social, institutional and organizational proximities have been used to explain aspatial frictions that can further hamper or benefit the chances of collaborations (Boschma, 2005). Beyond their clear conceptual boundaries, several studies have also felt the need to use additional ones such as economic, technological, relational, cultural and others (Caragliu & Nijkamp, 2016). Identification of these aspatial characteristics has allowed a more accurate estimation of the role space plays in forming collaborations. As part of that, spatial and aspatial proximities’ relations have also been questioned in which they can show substitutive or complementary characteristics (Boschma, 2005; Capello & Caragliu, 2018; Mattes, 2012). In this study, cognitive, relational (as a subset of social proximity) and institutional proximities will be detailed in the extent of scientific collaborations.

In the transfer or coproduction of knowledge, similarity in knowledge bases is envisioned to ease the process cognitively among the actors or entities involved (Nooteboom, 2000). At the same time, cognitively too proximate connections suffer from redundancy as interacting parties suffer from a lack of novelty as they get bounded by their highly overlapping knowledge bases (Lambooy & Boschma, 2001). In the extent of scientific research, cognitive proximity has been found to complement its geographical counterpart despite differences in variables used in studies (Cao et al., 2019). For a better interpretation of this relation, disciplinary bias should also be noted in terms of different fields’ responses to the spatial separation, despite differences in the wide variety of classifications preferred in studies ranging from international ones such as FORD or ASJC to national classifications.

Relational proximity can be considered to be a subset of social proximity. In the context of knowledge co-production, it explains the potential to collaborate and the capabilities to accommodate it (Basile et al., 2012). In collaboration networks, both weakness and strength of ties carry negative and positive implications. While strong ties can cover distances by reducing communication costs (Breschi & Lissoni, 2009), dense connections can act as barriers to newcomers and, thus, reduce creativity (Molina-Morales and Martinez-Fernandez, 2009; Molina-Morales et al., 2011). As for the concept’s relation to geographical proximity, collaborations increase in intensity when high social/relational proximity is coupled with closer geographical proximity (Cao et al., 2019; Capello & Caragliu, 2018).

Lastly, institutional environment is conceptualised in four levels ranging from informal institutions, constitution and laws to bylaws and policies, and, lastly, resource allocation processes (North, 1990; Williamson, 2000). The first level, informal institutions, have been embedded in social proximity (Boschma, 2005) therefore, the remaining three stages of formal institutions are understood from the use of this concept in the spatial scientometrics context. In multiple cases, it has been observed to substitute geographical proximity: The effects of resource allocation level are most visible in the example of China, where greater availability of resources to Beijing universities increases the centrality of this region among all regions (Andersson et al., 2014). In another case, regional inequality is found to influence collaborations as more prosperous coastal regions have greater chances of collaboration among them compared to inland regions (Scherngell & Hu, 2011). In this case, the restriction of citizens’ (including researchers) long-term mobility with a unique institutional barrier (hukou) is argued to intensify this trend. In the case of the European Union, institutional frameworks such as Framework Programmes fostered co-publications despite language barriers (Acosta et al., 2011; Hoekman et al., 2010) although segregation in collaboration patterns still exists based on complexity of knowledge (Balland et al., 2019). However, institutional environments and interventions are highly context-dependent and can significantly vary from one implementation to another. Along with this variety, its influence can change significantly as well.

As observed from the literature above, interaction effects between spatial and aspatial proximities have been explored in a few studies beyond the main effects. Cognitive and relational proximities may be expected to be parallel to those previous findings as they are based on the way research is carried out in general. Institutional settings, on the other hand, can differ based on scale as well as national policies and interventions which can lead to changes in its relation with the geographical proximity as well. Therefore, the first hypothesis is as follows:

Hypothesis 1

Scientific collaboration chances among provinces are influenced by spatial and aspatial proximities’ interaction effects.

While the aspatial proximities’ relation to geographical proximity has drawn some attention over the years as explored above, their interrelations with each other have been less studied. One such example has addressed this gap in the extent of pharmaceutical studies in which the interaction effects observed have been between cultural proximity and other proximity variables (Plotnikova & Rake, 2014). Therefore, interrelations among cognitive, relational and institutional dimensions remain to be explored in the case of more aggregated fields. This case leads to the second hypothesis:

Hypothesis 2

Scientific collaboration chances among provinces are influenced by interaction effects between aspatial proximities.

The case of Turkey

In the last three decades, international scientific community has increasingly been growing with scientists from developing countries. While China has become a driving force in this process, others like Brazil and India have also carved out specialization areas in which they make significant contributions (UNESCO, 2021). The expansion of their scientific systems and increasing collaboration trends have played a key role in these countries’ emergence in science (Meneghini, 1996; He, 2009). Similarities can be found between them and the Turkish case. However, Turkish collaborations in general as well as the effects of the country’s rapid expansion strategy have not been studied extensively from a spatial scientometric perspective, which is a huge gap for a country whose collaborations are primarily domestic. Studying the expansion and collaboration dynamics of Turkey contributes to global experiences and provides roadmaps for others in their catch-up process.

As a European Research Area (ERA) country, Turkey’s composition of collaborations is dominated by national collaborations (61%) in comparison to international collaborations (26.8%) with single-authored publications having the lowest share (12.2%) (Scival, 2021). However, international collaborations have been on an increasing trend since 1996 when they constituted 16% of total collaborations. From 1995 to 2010, the country was the fastest growing country in terms of degree centrality, being one of the two leading countries along with South Korea towards decentralization of science internationally (Choi, 2012). Increasing participation by Turkish researchers in international conferences through greater funding support has been influential in this rise (Aytac, 2010). More notably, significant policy changes in science and its geography took place through founding new universities under the “(at least) One university per province” approach of the government in 2006. From 2006 to 2010, provinces without universities were prioritized (Fig. 1) and, in the process, university numbers almost doubled in four years with the addition of 70 new ones to the existing 76 (Fig. 2). By 2010, 37 out of 146 universities in total were located in Istanbul, which was followed by 13 in Ankara and 6 in İzmir. This increase in university numbers was accompanied by an increase in student numbers for both old and new universities, which has been detrimental to the scientific productivity, most notably in public universities (Akçiğit & Özcan-Tok, 2020).

Fig.1
figure 1

Geographical distribution and number of universities per province in periods

Full size image
Fig.2
figure 2

Number of Universities founded by years in Turkey

Full size image

Beyond geographical expansion, however, fragmentation in the organizational structures of universities in Turkey is observed in this post-2006 higher education landscape (Kocatürk & Karadağ, 2021). Behind-the-scenes political conflicts in the mid-2000s over the autonomy of higher education resulted in greater government control over universities’ internal processes (Tekeli, 2019). Rectors, as the primary influence in the organizational structure of universities, became key pieces in this process (Kocatürk & Karadağ, 2021, p.244). This had implications for scientific knowledge production as observed by the low performance of universities with rectors who have lower academic competence and greater affiliation/interaction with political bodies (Karadag, 2021). In addition, while the post-2006 universities were found to follow rules and regulations set by the central authority Council of Higher Education (CoHE), older universities have had an organizational culture in which internally shared values shape decision-making processes and possess greater scientific productivity (Karataş Acer & Güçlü, 2017).

In regard to studies on Turkish collaborations, a rare inquiry into the role of space on the national scale as well as the nature of collaborations was carried out by Gossart and Özman (2009) in the extent of social sciences. Universities were divided according to their publication patterns in international and national journals. Their collaborations excluded one another on this basis as well, as those focusing on international journals did not collaborate with the ones publishing in national journals. This pointed out the lack of dissemination of international knowledge to the rest of the country. Another recent study which analyses scientific collaborations between universities through publicly funded projects was carried out by Unutulmaz-Gürlek and Dulupçu (2022), whose findings on proximities were parallel to the literature on the positive role of geographical proximity in fostering collaborations.

Data and methodology

Data

The data used in the study were obtained from productivity metrics and collaboration metrics of Scival service, connected to Scopus database. An important limitation of the data stems from the international nature of the database. Turkish international publications only account for around 30% of the total publications in the country, therefore, there is a large amount of domestic publications and collaborations missing from the analysis. However, the lack of accessible presentation of data from national databases has led to the preference for Scopus data.

Coauthorship data covers a period from 2015 to 2019. This period allows analysis of collaborations in the pre-pandemic era, which gives insight into collaborations before external factors forcefully made working at a distance more popular. Moreover, as the study seeks to differentiate the case of universities founded in the post-2006, the period selection can account for these universities’ (1) growth of their knowledge stock and (2) collaboration formations based on this stock accumulated. Towards the measurement of the chances of coauthorships, publication data is also gathered. This data covers the 2010–2014 period to avoid simultaneity with the co-authored publications. These data are obtained in six aggregate fields as well as in 42 sub-fields (only for publications) of FORD classification outlined in the Frascati Manual of OECD (OECD, 2015). The fields consist of agricultural and veterinary studies, engineering and technology, humanities and arts, medical sciences, natural sciences and social sciences. In comparison to other options, such as ASJC, this classification presents greater interpretation opportunities due to its more compact aggregate field definitions.

The co-publications include all types of publications ranging from indexed publications and conference proceedings, to book chapters and reviews. There are two reasons behind this preference. Firstly, different patterns exist in publication preferences for each field. Indexed scientific journals are more preferred in natural sciences, whereas monographs, conferences and books are more popular in social sciences and humanities and arts fields (Archambault & Larivière, 2010; Hicks, 2004). Also, these fields may be more spatially bounded and involve publications smaller in scale, leading to a small number of coauthorships, as observed in the example of the humanities and arts field, (Nederhof et al., 2001). In comparison, output of fields such as natural sciences may involve a large number of participants in highly international settings, such as publications resulting from large-scale projects (i.e. CERN-led projects).

Data is obtained at the institutional level, which is then aggregated to the provincial level (NUTS3 regions). To measure the propensity to collaborate based on coauthorships between 2015 to 2019, only universities founded before 2010 were included in the analyses. As a result, 146 universities’ coauthorship linkages are aggregated to 81 provinces during the process. Out of these 146 universities, 70 of them were founded between 2006 and 2010. In the aggregation process of these pairs, 10,585 pairs of coauthorship linkages between universities are aggregated into 3321 (3240 inter- and 81 intra-) province pairs. As these pairs represent linkages, their values represent the number of coauthorships between a province pair.

Model and variables

In the context of scientific interactions between pairs of provinces, gravity model is useful as a spatial interaction model. Undirected knowledge flows in the form of coauthorship linkages represent the gravitational force created between province pairs. This interaction intensity, \(Pij\), is explained by the productivity of province pairs i and j represented by the number of their publication sizes, M, under the friction of a spatial separation measure, \(d\), represented by distances between provincial centres. Additionally, while coauthorship data covers the 2015–2019 period, publication data is obtained for the 2010–2014 period to address the problem of simultaneity. This way, publications of previous periods, as the size effect, are measured based on whether they lead to coauthorships in the next period. A single weight parameter is used for the mass variables (Hoekman et al., 2010; Ponds et al., 2007). As coauthorships are undirected interactions between two provinces, the parameters of two masses cannot be distinct but rather be equal (Ponds et al., 2007, p.437). As a result, the size is represented by the product of two masses and their weight, \({\beta }_{1}\).

$$P_{{ij}} = \frac{{K(M_{i} M_{j} )^{{\beta _{1} }} }}{{d_{{ij}}^{{\beta _{2} }} }}$$
(1)

In order to obtain a linear form, log-normal specification is used.

$$\ln P_{{ij}} = \alpha + \beta _{1} \ln Size + \beta _{2} \ln Dist + \varepsilon _{{ij}}$$
(2)

Aspatial dimensions are then used to expand this basic model. Cognitive proximity is measured by the calculation of cosine similarity among provinces’ publication stocks, as employed in several instances in the past (Gui et al., 2018; He et al., 2020; Hoekman et al., 2010). In order to construct this for each of the six aggregate fields of FORD, publications in the 42 sub-fields of these fields are obtained for the period 2010–2014. Subsequently, these publications are used to construct a research profile in each of the six fields. Each province pairs are then represented by vectors x and y. The cosine angle between these vectors is measured to obtain the similarity between two provinces’ knowledge similarity. The value obtained is in a range from 0 to 1. If the provinces’ knowledge base is similar the value is closer to 1, with dissimilarity leads to values closer to 0 (Han et al., 2012). As the humanities and arts field has lower publication frequencies, there are notably higher amounts of 0 values.

$${\text{sim}}\left( {x,y} \right) = \cos \left( \theta \right) = \frac{{x \times y}}{{\left| x \right|\left| y \right|}}$$
(3)

Unlike cognitive proximity, there have been a variety of methods and indicators used for social/relational proximity. Focusing on those which used network properties, centrality-based indicators have been utilised such as degree centrality, betweenness centrality or structural holes (Cao et al., 2019; Gui et al., 2018) to discover the advantages of network positions. On the other hand, structural indicators representative of interactions/structures have only recently been given highlight (Dai et al., 2022). Focusing on these interactions, the study addresses the relational dimension by measuring the tie similarities of regions. Pearson coefficient correlation is preferred as a structural equivalence method to obtain these similarities. In this process, the expected number of common neighbours is subtracted from the actual number. If the score is between 0 and 1, there are more common neighbours than expected, while there are fewer than expected if the score ends up between − 1 and 0.

In measuring institutional proximity, the effects of the post-2006 rapid expansion policy and consecutive changes in the institutional environment on Turkish universities’ collaborations are observed. Whether this institutional divide extends to their scientific collaborations is aimed to be observed. To do so, the presence of post-2006 universities on the provincial level is calculated by the publication stock of these universities in their provinces. In this case, if they are the only university in a province, the value equals 1. If no university from this period exists in a province, the value is equal to 0. Then, product of collaborating provinces’ scores is used as the institutional proximity variable in the model.

After these additions, the final form of the model is as displayed in the equation below. All variables are centred. In their log-normal form, a constant (+ 1) is added to the dependent and independent variables. The complete list of indicators is as presented in Table 1.

Table 1 Indicators used in the estimation
Full size table
$$\ln P_{{ij}} = \alpha + \beta _{1} \ln {\text{Size}} + \beta _{2} \ln {\text{Distance}} + \beta _{3} Cog_{{field}} + \beta _{4} \text{Re} l_{{field}} + \beta _{5} Inst_{{field}} + \varepsilon _{{ij}}$$
(4)

Two additional models for each field are also constructed with two different groups of interaction terms to gain a greater insight into the role of both spatial and aspatial proximities. First, the combinations of aspatial proximities (cognitive, relational and institutional) have formed three pairs of interaction terms to expand the base model. In the second model, these aspatial proximities are coupled with the geographical proximity variable to form three spatial-aspatial proximity pairs of interaction terms.

Results

Descriptives in Table 2 show the highest number of coauthorships are observed in natural sciences, followed by medical sciences and engineering and technology fields. This is an expected result as per these fields’ publication and co-authorship frequency patterns. Correlation analysis is also presented in Table 3 below.

Table 2 Descriptive statistics by FORD fields
Full size table
Table 3 Correlation table of variables for each field
Full size table

Coauthorship shares of provinces are observed in the six fields in Fig. 3, represented by columns while the line displays those provinces’ cumulative share. The centrality of Istanbul and Ankara can be observed in copublication numbers, represented by the first two columns in the graphs separated from the rest. Despite Istanbul having three times more universities than Ankara, these two provinces’ shares are closer in most fields except agricultural and veterinary studies (Ankara being the top co-publishing province) and natural sciences (Istanbul being the top co-publishing province). At the same time, these two fields also indicate different patterns. While agricultural and veterinary studies display more spatially decentralized collaboration links around the country, natural sciences represent the other end of the spectrum as a large share of copublications is concentrated in a few provinces in this field. For the former, regional economic structure may be influential in the scientific growth as high performers in this field, such as Erzurum and Konya, are agricultural powerhouses in the country. On the other hand, knowledge in natural sciences is less spatially bounded which enables provinces to reach beyond spatial distances. Analysis of the data highlights participation in large-scale international projects, specifically the CERN-led one which has been the cause of this spike for multiple provinces.

Fig.3
figure 3

Coauthorship shares of provinces accorSding to research fields from 2015 to 2019

Full size image

The geographical dimension of these distributions can be observed in Fig. 4. Only the 5th quintile of coauthorships is mapped for ease of interpretation, thus, only high-intensity coauthorships are visible. The sparse network is observable in the case of humanities and arts and social sciences, and the central role of both Ankara and Istanbul is visible in both of these fields. Especially for the flows in social sciences, Ankara is still central in connecting to provinces around the country since the study of Gossart and Özman (2009). On the other end, the network in the natural sciences field is dense with relatively wider connectivity beyond Istanbul and Ankara. This case may stem from large-scale participation in previously noted international projects. As for the engineering and technology and medical sciences, several centres are visible after the core two provinces. However, relatively more flows are concentrated between the core and periphery for the former. İzmir, Konya, Erzurum and Adana provinces are in the top 10 in these fields, which are located in western, middle, eastern and southern Anatolia respectively.

Fig.4
figure 4

Coauthorship flows among Turkish provinces in the 2015–2019 period

Full size image

As the coauthorship data includes a high number of zero value linkages, it fails to fulfill normality assumptions. Count models are suitable for estimation purposes in their case. Due to the overdispersion, which is frequently encountered in coauthorship data, the Pearson χ2 statistic is checked for all six fields’ models. Together with better AIC scores, the generalized Poisson model was found to be the better fit as there were no significant improvements in the negative binomial alternative. No high correlation is found among the variables of the models as can be observed in Table 3 below with VIF scores also below 2. Infrequent publication and co-publication patterns in the humanities and arts field have led to non-significant results for aspatial proximities and their interaction terms in all three models of this field.

Starting with the models with main effects on coauthorships in Table 4, the relevance of geographical proximity as a positive influence is observed to be continuing for all fields. Its relatively stronger influence in engineering can be attributed to the limited availability of research infrastructure such as laboratories and equipment as well as greater tacit knowledge requirements in the field. In the extent of cognitive similarity, engineering and social sciences coauthorships are observed to be not influenced while medical sciences have the most positive association. The relatively low number of regional centres for a long time, which had provided medical services to their immediate neighbouring regions, can be argued to have created all−round expertise in these provinces leading to convergence and high overlap in their knowledge bases. Relational proximity also leads to an increase in most fields. On the other hand, the publication activity of post-2006 universities does not contribute to higher chances of interprovincial coauthorships for the analysis period.

Table 4 Generalized poisson model
Full size table

In the event of including aspatial-aspatial interaction terms in the base model, the main effects with significant results do not change. Only the cognitive and relational proximities’ interaction effect exists among the terms included which is significant in all the fields except the humanities and arts. This means similarity in cognitive dimension extends positive influence on collaboration chances when network ties of provinces are dissimilar and vice versa. This indicates a substitutive relation between the two proximities. This case also highlights a notable change in the results for engineering and social sciences fields. Cognitive proximity is only influential in these fields when there is less overlap in collaboration networks.

As for the spatial-aspatial interaction effects, closer the distance higher the influences of both cognitive and relational proximities are, which points towards a complementary relationship. This finding is parallel to the previous literature (Cao et al., 2019; Capello & Caragliu, 2018). Notably, the post-2006 universities’ publication activity was also found to contribute to the formation of coauthorship chances at a closer distance in the extent of the engineering and natural sciences fields.

As noted previously, other social network analysis metrics have also been used for social/relational proximity in past spatial scientometric studies, although structural metrics have been less observed. For a robustness check of this variable, two approaches are followed: (1) Use of different metrics for structural equivalence and (2) use of a different structural metric. For the first, different similarity metrics are employed. These similarity metrics are set-theoretic similarity measures, from which cosine similarity and the Jaccard Index are used (Newman, 2010; van Eck & Waltman, 2009). Despite their differences from probabilistic similarity measures, such as the Pearson correlation coefficient used in this study, the results obtained are found to be robust in most cases. In the second approach, local clustering coefficient is used as the different structural metric, which indicates the degree of control a node exerts on its immediate neighbours (Newman, 2010). While little change was observed in the base models, aspatial proximity pairs’ interaction effects for the engineering and technology field yielded non-significant results, unlike the original model. Other than this case, most results have been robust once again. Additionally, cognitive proximity is also checked considering the similarities that may be observed across the fields. The Pearson correlation coefficient is used for this purpose (Moreno et al., 2005; Scherngell & Hu, 2011) and the results were found to be robust again.

Discussions and conclusions

Science is increasingly carried out in teams which is the trending force behind both productivity and novelty (Jones et al., 2008; Olechnicka et al., 2019). In this context, this study sought to identify the dynamics and interrelations of spatial and aspatial forces behind the co-production of scientific knowledge. The study confirms previous findings as coauthorship chances increase when the interacting provinces are closer to one another spatially (Cao et al., 2019; Hoekman et al., 2010; Sidone et al., 2016). On the other hand, similar knowledge bases and networks increase the chances of coauthorships in most fields. In spatial-aspatial pairs, complementary dynamics with geographical proximity were visible for the cognitive and relational proximities. Their positive influence in closer proximity should be regarded in parallel to these aspatial proximities’ interaction with one another as well. When these two aspatial effects were combined, they had substitutive effects on one another and had a negative association with coauthorship chances. Thus, too much overlap in collaborating regions’ networks and knowledge bases is found to reduce opportunities to co-produce knowledge as well. This can be regarded as indicative of a lock-in: If two provinces’ networks are too similar, their knowledge bases should differ to explore new knowledge combination opportunities.

In the case of institutional proximity, post-2006 universities’ academic production in their provinces has had a negative association with coauthorship chances. In other words, collaborations among provinces with universities before the 2006 policy have higher chances of collaboration. In contrast, however, coauthorship chances increase for the provinces with new universities when they are closer to one another as the interaction term with the geographical proximity shows in the extent of engineering studies and natural sciences. While this complementary relation may contrast past examples in the literature, which point towards a substitutive one (Cao et al., 2019), it is not so unexpected in Turkey’s case. The policy decision to establish universities in all provinces can be argued to have enabled collaboration opportunities among provinces to be directed by these fields’ inherent conditions which entail geographical proximity effects. The requirement of physical access to laboratories and other equipment may be argued to have pushed researchers in neighbouring provinces to establish bonds with one another to make use of their equipment. Moreover, teams’ compositions with large numbers of participants in case of these fields may present greater collaboration opportunities which would transfer knowledge and expertise to new institutions to improve their research base and contributions. Furthermore, both fields entail codified and globally footloose knowledge (more for natural sciences) which may increase their ease of access to novel knowledge either by themselves or through their neighbouring provinces with high international connectivity. In these ways, geographical proximity can be argued to be a positive force in the integration of new universities and scientifically emerging provinces to the national scientific community in these fields.

Policy implications of these findings present important spatial strategies in the event of rapid expansion of higher education. The 2006 policy decision of Turkey was a sweeping attempt and did not prioritize any provinces in its resource allocation. Under such conditions, the increasingly fragmented public funding negatively affected both the older and newly founded universities in sustaining their research activity. In this context, the findings of the study support arguments in favour of spatial concentration at the regional level for the growth of new universities at least in the extent of engineering and natural sciences, which can be springboards for a quicker development of research activity. Moreover, this would also allow older universities to form new networks and access different knowledge bases, which are also effective in increasing co-authored publication opportunities. Under straining resources, utilizing collaborations in such a way would empower the emerging provinces and ultimately contribute to a wider strategy for improving national scientific visibility.

In future studies, the growth of new universities’ collaborations may also be observed from a researcher level of analysis as new universities’ academic personnel have initially consisted of staff hired from previously existing universities which provide these universities organic ties to some pre-existing knowledge networks and resources. Thus, whether their ties form on this basis, persist and how their diversity is affected as a result constitute crucial points for further observation in understanding knowledge flows in a rapidly transforming example such as Turkey. In addition, the exploration of variety in knowledge may also be carried out by analysing institutions or authors to delve deeper into the role of cognitive proximity in future studies. As Turkey is also a rapidly growing international actor in scientific collaborations (Choi, 2012), its international collaborations are also required to be given more emphasis by future studies.