Social capital and knowledge creation: a higher education institution networks

Abstract Social capital concept has become popular over the past decades. What is not known in literature and worth investigating is the network designs and nature that enhance the development of new learning ideas and knowledge creation. The aim of the paper is to examine network structure dimensions, namely centrality (bonding) and structural holes (bridging or looser ties) of social capital formed within higher education networks that stimulate the development of new ideas and knowledge creation of the participants in the networks across multiple domains of science. This study employs negative binomial regression on a sample of participants within a higher education network to predict the development of new ideas and knowledge creation. The results show that the different features of social capital dimensions influence the development of new ideas and knowledge creation of the participants in the networks differently. Specifically, the study reveals that the brokerage network appears to be more relevant than cohesion in the Ghanaian higher education institution networks. When knowledge creation is decomposed into different fields of science, Health Science is found to be the most productive. The decomposing of the knowledge creation into different fields of science remains a value of this study.


Introduction
Social capital concept has become popular over the past decades (Macerinskiene & Aleknaviciute, 2011). According to Putnam (1993), social capital is referred to as features of social organizations for collective gains. Shiri et al. (2013) report on seven distinct aspects of social capital to include social values, social trust, social networks, social cohesion, social participation, social communication and sharing knowledge. A number of studies have shown that social capital can be enhanced through the didactic approaches that apply active systems and collaborative activities (Virtanen and Tynjala, 2019). Previous empirical studies have established that a higher educational setting is considered as an active system and collaborative activities that social capital is more prevalent (Shiri et al., 2013). In higher education institutions that support knowledge creation, network activity that stimulates the development of knowledge creation is a routine practice (Coates and Fraser, 2014). Networks and collaborations in higher institutions are often relied upon by universities to promote extensive network opportunities (Donert, Hay, Theobald, Valiunaite, and Wakefield, 2011).
In Ghana, networks and collaboration among universities is on the rise in recent times. This assertion is premised on a bibliometric analysis of Ghana publications in the science citation index by Boamah and Ho (2017). Boamah and Ho (2017) noted that there has been a significant increase in total number of articles (from 12% in 2014 to 27% in 2015), number of authors (from 32% in 2014 to 43% in 2015), and a number of cited references (from 21% in 2014 to 30% in 2015). The significant increase in networks and collaboration resulting in publication output in Ghana is due to attempts by several governments of finding ways of supporting and strengthening research-based activities in the country. These supports and attempts include the following: (1) the formation of the Ghana Education Trust Fund (GETFund) with its central funding of a minimum of 2.5% of the prevailing rate of national value-added tax deductions (UWN, 2014); (2) the national Science and Technology Research Endowment Fund (STREFund) with initial capitalor seed money of approximately $500,000 to operate as an independent funding mechanism (UWN, 2014), and (3) promotions to higher levels of seniority depending on minimum requirements such as academic qualifications and a certain number of publications, including technical reports (UNCTAD, 2011). These directives and initiatives could explain the recent surge in development of new ideas and knowledge creation in the country (Owusu-Nimo & Boshoff, 2017). According to Burt (2001), people who do better are somehow better connected. Notwithstanding, the network designs and nature in social capital that enhance the development ideas and knowledge creation are yet to be investigated in developing economies. This paper reports on a collaborative partnership formed in higher education settings, such as universities in knowledge creation. Specifically, it employs network structure dimensions, namely, centrality (bonding) and structural holes (bridging or looser ties) of social capital in promoting knowledge creation across multiple domains of science.
The present study contributes to literature in a number of ways: First, the study highlights the controversy between centrality (bonding) and structural holes (bridging or looser ties) of social capital as drivers of knowledge creation. Existing theories behind the importance of networks for knowledge creation have been contradictory. Second, the study contributes to literature on the development of new ideas and knowledge creation by advancing our knowledge of the specific mechanisms that make research collaboration work, and how they influence individual participant's skills and knowledge. Thus, this study considers the relevant aspects of centrality and structural holes of social capital in promoting new ideas and knowledge creation across multiple domains of science. Third, the study assesses the relationships between social capital and knowledge creation across distinct areas of knowledge. This will contribute to the network literature by improving the understanding of policy makers in the drafting of policies tailored to increase knowledge creation across multiple disciplines. Besides, to the best of our knowledge, this is the first study that focuses on Ghana. The study employs current data of all research collaborations and considers all authors that have published with a Ghanaian-affiliated scientists. Also, in terms of knowledge creation measurements, while prior studies have sought to measure knowledge creation using the impact factors (Burt, 2001;Gilsing, Nooteboomb, Vanhaverbekec, Duystersd and Oorda, 2008;Ahuja, 2000), knowledge creation using scientific output is scanty in the Ghanaian academic environment. We therefore anticipate that a more nuanced frame of reference on how one estimates knowledge creation could help contextualize and better explain the influence of the various aspects of network structures. Finally, studies have documented that the nature and designs of the networks produces different effects depending on the performance variable that is taken into consideration (Gonzalez-Brambila et al., 2013). The study dataset allows us to consider different outcomes.
The study is presented as follows: The related studies about social capital are explored in the next section. Subsequently, the study methodology is analyzed and presented. The study results are sequentially presented. The study culminates with conclusion and provides some implications for policy-makers.

Theoretical foundations and hypotheses development
Network analysis yields two inconsistent accounts of theories. First, rational choice theory developed and popularized by Coleman (1998) posits that actors in embedded networks perform better because of trust and coordination. Thus, social interdependencies arise among actors because they are interested in events and resources controlled by other actors to maximize their utility by rationally choosing the best solution for them. Coleman differentiates between the kinds of social capital as: relations of mutual trust, authority relations, information potential, effective norms and appropriate social organizations. Coleman further asserts that there are different factors that influence social capital as a whole, namely, closure, stability and ideology. Rational choice theory states that: "Social capital is defined by its function. It is not a single entity, but a variety of different entities having two characteristics in common: They all consist of some aspect of social structure, and they facilitate certain actions of individuals who are within the structure" (Coleman, 1990: 302).
Finally, structural holes theory originally developed by Burt (1992) stipulates that gap between two individuals who have complementary sources to information. Structural holes theory further posits that open social structures with many structural holes, i.e. gaps between two individuals who have related sources of knowledge, enhances productivity because of access to new information. Thus, productivity and innovation are greatly influenced by structural holes (Burt, 1992(Burt, , 2004Hargadon & Sutton, 1997;Hargadon, 2002). Structural holes support the argument that gaps between two actors who have same sources to information can be beneficial in terms of productivity. These benefits are as a result to access to new information (Burt, 1992).
Empirically, the relationship between social capital and knowledge creation is extensive and available in the literature. Notwithstanding, the overall findings have been mixed and inconclusive. After the introduction of scientific productivity by the seminal work of Lotka (1926), a number of empirical works have established a direct link between collaboration and scientific productivity. De Solla Price and Beaver (1966) argue that the researchers who do well are often those who collaborate the most. Zuckerman (1967) demonstrates that most prestigious award winners are often more willing to collaborate. He et al. (2009) point out that university collaboration and international collaboration significantly influence the article's quality and scientist's future productivity. Wuchty et al. (2007) suggest that co-authored papers tend to receive more citations than sole-authored papers. Empirical studies, which have found similar relationships include Narin et al. (1991), Katz and Hicks (1997), and Glanzel and Schubert (2001). Alternatively, McFadden and Cannella (2004) established a negative relationship between social capital and knowledge creation. Their study sampled 173 scientists from 1989 to 1999 and concluded that as connections grow in number, gain to knowledge creation reduces.
Related literature is further reviewed along different dimensions of social capital and distinct domains of knowledge creation. Several studies have also shown that scientific output or knowledge creation is a function of different dimensions of social capital (Nahapiet & Ghoshal, 1998;Singh, 2007;Gonzalez-Brambila et al., 2013). Empirical results for these studies have yielded relevant insights. Notwithstanding, these authors have made the following admissions: First, the authors admitted that they ignored inclusiveness or centrality concerns in their studies. Second, distinct characteristics of social capital influence scientific productivity differently. Again, Gonzalez-Brambila et al. (2013) conclude that since knowledge creation is influenced by driver's inputs of network embeddedness, accounting for various social network dynamics is essential. The study examines and provides new designs of networks that stimulate research productivity across multiple domains of science. The choice of network structures measurements i.e. centrality and structural holes of social capital are as follows: First, studies by Hanneman (2001), Rotolo and Messeni Petruzzelli (2013) suggest that an actor's position in the network is likely to influence his/her opportunity and constraint. Therefore, there is the need to control and account for the most influential actors within the network. Second, productivity and innovation are greatly influenced by structural holes (Burt, 1992(Burt, , 2004Hargadon & Sutton, 1997;Hargadon, 2002). Consequently, the inclusion of structural holes is to support the argument that gaps between two actors who have same sources to information can be beneficial in terms of productivity. These benefits are as a result of access to new information (Burt, 1992). Two propositions are outlined in this study. The first proposition is that different dimensions of social capital affect scientific output (Burt, 1992;Adams et al., 2005;Gonzalez-Brambila and Veloso, 2007). The study, therefore, formulates the hypothesis as: Hypothesis 1. Different dimensions of social capital affect knowledge creation.
Finally, distinct characteristics of social capital influence scientific productivity differently. This proposition has arrived at because studies have shown that collaborations differ according to the fields of science. While others tend to enjoy more collaborations, others do not (Adams et al., 2005;Gonzalez-Brambila and Veloso, 2007). The study, therefore, formulates the hypothesis as: Hypothesis 2. Distinct characteristics of social capital influence knowledge creation differently.

Sample and procedure
The study sample consists of research scientists involved in higher education institutes. These institutes include nine (9) public universities, one (1) teaching hospital, six (6) government-based research centers, and two others, namely, Noguchi Memorial Institute Medical Research and Council for Scientific and Research Institute (CSIR). While the study employs STATA 16 software to do data mining, cleaning and estimations, open software called VOSviewer was used to generate Network visualization (NV) and Overlay visualization (OV) analysis. Web of Science (WoS) provides the data source for this study. This data includes at least one author from Ghana with a paper published from 1 January 2010 to 14 September 2019. The justification for the data span is not far-fetched. Within these periods, governments have made several attempts in finding ways of supporting and strengthening research-based activities in the country. The preliminary investigation on the publication record includes the following: (1) Date of publication, (2) Name of authors, (3) Address information, (4) Number of co-authors, (5) Field of knowledge, (6) Institutions of authors, (7) Country of authors, and (8) Continents of authors. These are connected in Boolean logic (OR and AND) to build the research strings. The strings are then used in the Web of Science online database.
The search for collaboration is also limited to two authors. More than two authors are dropped from the network variables. This threshold is essential because, per the data perused, about 85% of articles in the WOS database are by one or two authors. This produced a population size of 765 articles with 1,530 authors. About 256 full-text articles with 512 authors are assessed for eligibility after removing and eliminating all the duplicates. To avoid duplications of authors, affiliation and countries, various identification numbers are assigned to all authors. The study further captured single authors. The inclusion of the single authors would not only be used as a control variable but it can help account for the overall impact of social capital on the development of new learning ideas and knowledge creation.

Dependent variable
In this study, peer-reviewed papers or articles are used as proxies for knowledge creation. These are either single authored publications or publications by two authors. Studies that have used similar proxies for knowledge creation include Singh (2007) and Gonzalez-Brambila et al. (2013).

Independent variables
The independent variables used are centrality and structural holes as they relate to social capital. Centrality is one of the most studied concepts in social network analysis (S. Borgatti, 1997). Centrality refers to an actor's position in a network (Zheng, 2010). Lin (2001) refers to centrality as position strength, implying that those higher in the network's hierarchy have more access to knowledge about network structure (actors, ties, and resources) and so are better able to access and use social capital effectively. Numerous measures have been developed, including degree centrality, closeness, betweenness, eigenvector centrality, the influence measures of Katz (1953), Hubbell (1965), Hoede (1978), and Taylor's (1969) measure. This study employs centrality dimensions such as closeness centrality and degree centrality. These are used and discussed as follows: Closeness centrality is about how close an individual is (on average) to all other individuals. This variable is proxied by farness. Thus, how many steps (on average) does it take an individual to reach all other individuals? Farness is calculated as: where dist_ik is the length of the shortest path from node i to node k.
Another centrality measure is the degree centrality. This calculates degree centrality for each node of a network or list of networks. It divides the centrality scores by N-1, where N is the total nodes in a network. This standardization makes sure that centrality scores always range from 0 to 1. It estimates the number of incoming (in-degree centrality) or outgoing (out-degree centrality) ties. In-degree (effect) with high in-degree is impacted by multiple other variables. An in-degree of 0 means that a variable is not influenced by others in the system. The in-degree is calculated as: where yji is the total shortest path. Out degree (cause) with high out-degree variables have an ability to change many others in the system. Variables with an out-degree of 0 do not directly influence others (Zhang & Luo, 2017).
where yji is the total shortest path.
Structural holes, defined as gaps between two individuals who have related sources of knowledge, enhance knowledge creation because of their connections to current information. The structural holes measure focuses on redundancy and constraints in the network. This measure calculates Burt's (1992) effective size, efficiency, constraint and hierarchy. All these measures are used to identify structural holes. They all build on the idea of redundancy. Structural holes proxies used in this study include effective size, efficiency, constraint and hierarchy. Effective size is the total altars that ego has, minus the average number of ties that each alter has to other alters (Burt's, 1992). This measure focuses on the ego's total impact. The efficient size is estimated as: where p iq ¼ y iq=si ¼ sum over i½y ij� is the proportion of actor i's relations that are spent with q. m jq is the marginal strength of contact j's relation with contact q. Which is j's interaction with q divided by j's strongest interaction with anyone. For a binary network, the strongest link is always 1 and thus m_ jq reduces to 0 or 1 (S. P. Borgatti, 1995). Efficiency also measures what proportion of ego's ties to its neighborhood are "non-redundant." An actor can be effective without being efficient; and an actor can be efficient without being effective (Burt, 1992). The next proxy for structural holes is the constraint dimension. This dimension refers to how much room one has to negotiate or exploit potential structural holes in their network. This estimates the extent to which a manager's time and energy are concentrated in a single group of interconnected colleagueswhich means no access to structural holes (Burt, 1992). This is calculated as: where Ci is network constraint on manager i, and cij is a measure of i's dependence on contact j.
The final aspect of structural holes is the Hierarchy. Network hierarchy examines the nature of the constraint on ego and does not estimate the degree of constraint. It is often traced to a single contact in the network. The hierarchy measure is assumed to be high when the total constraint on the ego is centered in a single other actor. The opposite is the case (Burt, 1992).

Statistical methods
In this study, several estimating models are considered. Other studies have used OLS regression, Zero-inflated regression, Model Poisson regression and Negative binomial regression. Initial diagnostic tests, as shown in Table A, reveal the prevalence of heteroscedasticity in the observations. This means that the variance of the error terms differs across observations.
The study employs a negative binomial model as an estimating method. The choice of the negative binomial model is informed as follows: First, negative binomial regression is for modeling count variables, usually for over-dispersed count outcome variables. 1 Secondly, the negative binomial model is employed as a functional form that relaxes the equidispersion restriction of the Poisson model. 2 Finally, a useful way to motivate the model is through the introduction of latent heterogeneity in the conditional mean of the Poisson model (Greene, 2007). Thus, the study specifies: Where h i ¼ exp ε i ð Þ is assumed to have a one parameter gamma distribution, G θ; θ ð Þ with mean 1 and variance 1=θ ¼ k. Hence; After integrating h i out of the joint distribution, we obtain the marginal negative binomial (NB) distribution as follows: y i ¼ 0; 1; :::; θ>0; The latent heterogeneity induces over dispersion while preserving the conditional mean. Thus;  (2011) proposed in recent times, it is expected that this trend will increase in subsequent years. Table 1 presents the decomposed knowledge creation of the various higher education networks by year. It further expresses the output in percentages for all research higher education networks across all the years. During the years under review, public universities contribute about 93% to new ideas and knowledge creation. This is followed by government-based research centers (5%) and others (2%). The huge impact from the public universities to knowledge creation is understandable because the majority of these higher education institutions have been in existence since the colonial era, and they are still significant in promoting knowledge and technology (UNCTAD, 2011). Besides, since 1957 when the country gained its independence, these institutions have continued to function and have taken on more responsibilities.

Trends in knowledge creation
The distribution of knowledge creation or scientific output of a higher education by fields of science is presented in Table 2. Such segmentation is to help show the publication designs along     In Table 3, distribution of papers by fields of science and location (continents) of co-authors is further examined continentally (Africa, Asia, Europe, Australia and North America). Such decomposition will help policy makers in finding ways to improve knowledge generation by the various disciplines. In Health Science, while collaborations with scientists within Africa yield 38% of output, 29% is recorded for European collaborations. Collaborations with Australian scientists contribute 67% of scientific output in the Humanities. Scientists within Africa and Europe contribute more in Interdisciplinary  Studies than any other continents. North American collaboration produces more knowledge generation in the area of Natural Science. Interestingly, knowledge output in terms of Physical Science comes from Europe and North American-based scientists. This is not surprising as Europe and North America have been major sources of scientific and technological research and new discoveries over the centuries. Table 4 presents the network summary and dyad census of the study. Network density, which estimates the possible relations in a network that are actual relations, is also considered. This shows the proportion of possible relations in the network that are actually present. The density network value ranges from 0 (sparse) to 1 (dense). As shown in Table 4, network density of 0.0106, 0.0619 and 0.3500 for institutional, country and regional networks, respectively, suggest that higher education networks in Ghana are not more cohesive. The study also estimates the network reciprocity indicator. This measures the degree to which ties among a group of nodes are mutual. It is the degree to which scientists in Ghana follow each other or enjoy reciprocal collaboration. The value ranges between 0 and 1, with 0 indicating the absence of mutual ties in the network and 1 indicating that all the relations are mutual in the network. As a result, scientific collaboration in Ghana is not mutual. Network transitivity also follows a similar trend in the analysis apart from the regional network. Betweenness centralization, known as the bridge as shown in Table 4, demonstrates how many shortest paths go through an individual. A higher value of betweenness centralization implies that more scientists in Ghana depend on an actor to make connections with other scientists. The cause and effect, as measured by the in-degree and out-degree, indicate that with the exception of institutional networks, scientists found in country and regional networks are likely to have the ability to influence and change others in the systems. When dyad census, which describes a pair of nodes connected through a link, is estimated, the null outweighs asymmetric connections. This suggests that all the networks are stable. The list of institutions, countries, regions and their respective codes are shown in Tables B and C as contained in the appendix.

Collaboration analysis
Collaboration analysis for the study is presented in terms of co-authorship and citations analysis. As shown in Figures 2-11, the edges have direction, and those directions of travel on the lines are defined by those arrows. This implies that collaborations of Ghanaian scientists have a direction. Figures 2-10 seek to analyze co-authorship using as units of analysis authors, institutions, countries and continents along two visualizations, namely, network visualization and the overlay visualization. While network visualization considers the relatedness of coauthorships, overlay visualization focuses on developments over time. In general, the closer two authors are located to each other, the stronger their relatedness. As shown in Figures 2 and 3 Figure 11 reports a graphical representation of citations using countries as the unit of analysis. It shows the number of citations received by all documents published by the country. The result shows that knowledge creation or scientific publications and collaborations from Ghana, China, Kenya and Nigeria receive more citations.

Descriptive statistics
Summary statistics of the study are presented in Table 5. Generally, the results show some level of variations of variables across the sample. This implies that most of the variables have their observations not clustered around the means. Different knowledge generation across fields of science could be a result of the existence of heterogeneity. Specifically, scientists found in Health Science, Physical Science and Interdisciplinary Studies tend to have a higher knowledge creation and enjoy more collaborations than scientists in the Humanities, Natural Science and Social Science. This result supports the findings of Gonzalez-Brambila (2014). Interestingly, scientists in Ghana publish more single-authored papers. Across the fields of Science, the Social Sciences enjoy the highest number of single-authored papers with the least being the Humanities. In Table 5, the study reports lower values for the networks regardless of the field of science. This indicates that scientists in Ghana observe close connectivity and they are able to access information directly or indirectly. The lower average values of in-degree (cause) and outdegree (effect) reported suggest that scientists in the networks have the ability to influence and change others in the systems. In terms of structural holes, results are not different. The study established open social structures with many structural holes. Thus, gaps between two individuals who have related sources of knowledge can be beneficial because of their connections to current information. The presence of structural holes facilitates the development of innovative

Figure 5. Overlays visualization for co-authorship in terms of institutions.
products (Burt, 1992(Burt, , 2004Hargadon & Sutton, 1997;Hargadon, 2002). In terms of the field of science, Health Science appears to be rich in structural holes. Table 6 reports the results of the tests performed in the study. A main thrust of the study is Hypothesis 1. It tests for the effects of different aspects of social capital on knowledge generation. The two main aspects are centrality and structural holes. Model 1 in Table 6 presents the results of the tests. The centrality dimension of social capital, proxied by farness, out-degree and in-degree, is found to negatively affect knowledge creation or scientific output at the 1% significance level. Thus, per this result, centrality inhibits knowledge creation or scientific productivity. This result contradicts the findings of Rotolo and Messeni Petruzzelli (2013) but finds support in the work of Li et al. (2013) who assert that centrality significantly affects publications negatively. Structural holes were proxied by effective size, efficiency, hierarchy and constraint. Results from Model 1 show that only efficiency positively affects knowledge creation at the 1% significance level. Thus, while centrality negatively affects knowledge creation, the effect of structural holes is not conclusively determined from the results in Model 1. This result runs parallel to the empirical studies of Hargadon (2002) and Burt (2004). This result suggests that brokerage in networks (represented by efficiency) appears to be more relevant than cohesion and that good ideas often emanate from networks that are rich in structural holes. These results find support in the empirical work of Burt (2014) and Gonzalez-Brambila et al. (2013). Gonzalez-Brambila et al. (2013) found a prevalence of structural holes over cohesion. The same inferences cannot be made if structural holes are proxied by effective size, hierarchy and constraint.

Different dimensions of social capital affect knowledge creation
Also in Model 1, the study further combines structural holes proxies and single authored papers as control variables to estimate their impact on knowledge generation. This is done to ensure Africa Asia Australia Europe North_America mynet_1_2 = 0 mynet_1_2 = 21 Regional Collaboration Network Figure 8. Graphical representation of regional networks.

Figure 9. Network visualization for institutional citations.
some level of consistency with the findings. These variables include effective size and efficiency, effective size and constraint, efficiency and hierarchy, and single author papers. Of the four control variables, effective size & constraint, efficiency & hierarchy and single author papers produced significant results although varied. The combined effect of effective size and constraint positively and significantly affect new learning and knowledge creation, albeit at the 10% significance level. This implies that the combined dimension of social capital, i.e. efficiency and the nature of constraint is important for quality, but not for output (Gonzalez-Brambila et al., 2013). The negative relationship between the combination of efficiency, hierarchy and knowledge output suggests that the combinations of ego's total impact and their capacity to negotiate or utilize potential structural holes in networks are less relevant to new learning and knowledge creation. Single-authored papers, which appear to be common in Ghana, negatively affect new learning and knowledge creation at the 1% significance level. Thus, in Ghana single authored papers do not contribute significantly to new learning and knowledge creation. Overall, in Model 1, varied results are found for the two aspects of social capital. Thus, Hypothesis 1 is validated.

Distinct characteristics of social capital influence knowledge creation differently
Hypothesis 2 tests how the distinct features of social capital affect new learning and knowledge creation differently. The results are represented in Models 2 to 7 of Table 6. In Model 2, centrality proxies such as closeness and farness positively and significantly affect new learning and knowledge creation in the Health Sciences. This implies that centrality appears to be more relevant among researchers in Health Sciences. Structural holes, measured by effective size, efficiency, hierarchy and constraint insignificantly affect knowledge generation. This result contradicts the findings of Gonzalez-Brambila et al. (2013). Gonzalez-Brambila et al. (2013) asserts that the richest networks in structural holes are seen among researchers in Health Sciences. Single authored papers negatively affect new learning and knowledge creation, although at the 10% significance level. This suggests that in terms of knowledge generation diversity, single authored papers in Health Sciences are less important in terms of new learning and knowledge creation.
In Model 3, both centrality and structural holes dimensions proxied by in-degree and efficiency respectively affect Humanities negatively. This indicates that as far as social capital is concerned, strength of position in the network and open social structures do not promote productivity in the Humanities. Single authored papers also do not significantly enhance the Humanities as an area of knowledge generation. In Model 4, social capital dimensions, i.e., centrality and structural holes, insignificantly affect Interdisciplinary Studies. This could be as a result of the nature of research Interdisciplinary Studies develop. Model 5 reports varied results for the Natural Sciences. In terms of centrality dimensions as proxied by in-degree and out-degree, centrality positively affects Natural Sciences at 1% and 5% significance levels. Although structural holes showed significant a relationship between effective size, hierarchy and Natural Sciences, its impact is inconclusive. For instance, while effective size negatively and significantly affects Natural Science at the 1% significance level, hierarchy positively affects Natural Science at the 1% significance level. The mixed results under Model 5 could mean that while hierarchical clustering, defined as dissimilarities between all actors in the network are relevant to Natural Science, reduction of nonredundancy elements in the network measured as effective size is not beneficial to Natural Science. Single authored papers negatively affect Natural Science's productivity at the 1% significance level.
Model 6 reports the results for Physical Science as an area of knowledge generation. At the 1% and 5% significance levels, centrality and structural holes dimensions proxied by in-degree and effective size respectively affect Physical Science negatively. Thus, ego's total impact and ability to influence others in the network do not enhance new learning and knowledge creation as measured by Physical Science. Among the control variables, single authored papers only revealed a negative relationship with Physical Science output. Finally, Model 7 concludes and reports the results for knowledge generation diversity with particular focus on Social Sciences. The results show that while centrality indicators such as closeness and farness respond negatively to Social Sciences' knowledge output, structural holes indicators measured as efficiency positively influence Social Sciences knowledge output. From Models 2 to 7, the study concludes that distinct natures of social capital influence new learning and knowledge creation differently. Thus, Hypothesis 2 is validated.

Conclusions, implications and limitations
The paper discusses collaborative partnership formed within higher education networks that provide insight into network designs that stimulate the development of new ideas and knowledge creation of the participants in the networks. This study has generated the following relevant results. First, Ghana has witnessed a significant growth in knowledge generation and scientific production for the last six years and public universities, research centers and others have been active participants. Second, Health Science, Humanities and Natural Science contribute more to knowledge generation in the country. Third, Health Science enjoys knowledge generation and output from scientists within Africa and Europe. In the Humanities, scientists within Africa and Australia contribute more to knowledge creation. Knowledge generation in Interdisciplinary Studies is dominated by scientists within Africa and Europe than in any other continents. North American collaboration with Ghanaian scientists produces more knowledge generation in the area of Natural Science. Interestingly, knowledge output in terms of Physical Science comes from collaboration between African, European and North American based scientists. Fourth, at the institutional, country and regional levels, social networks depict directed networks.
These findings suggest that collaborations with Ghanaian scientists have a defined direction. Network density, however, is not cohesive. While all networks are stable, scientific collaboration in Ghana is not mutual. Furthermore, single-authored papers, although common in Ghana, do not benefit knowledge creation. Consistent results are produced for all areas of knowledge. In addition, relevant design of networks stimulates knowledge creation differently. For instance, while the centrality dimension of social capital negatively affects knowledge creation, the structural holes dimension, proxied by efficiency, positively and significantly affects knowledge creation. These results suggest that the brokerage network appears to be more relevant than cohesion. Varying results are produced when the data is decomposed into different fields of science. Again, the study results confirmed the negative impact of centrality networks to knowledge creation. However, the impact varies. Finally, although the structural holes dimension of social networks enhance knowledge creation, the combinations of ego's total impact and their capacity to negotiate or exploit potential structural holes in networks are found to be more relevant than clustering based on dissimilarities between all actors in the network.

Policy implications
The study offers the following practical implications. First, initiatives and policies by successive governments to boost knowledge generation and research collaborations should be sustained and improved. Second, a higher percentage of funding from the government should be given to public universities and research centers. This is because they have been in existence since the colonial era and are still significant in promoting knowledge. Also, policy makers should be aware that as they try to promote knowledge creation, relevant designs of networks stimulate knowledge creation differently. Finally, as more attention is given to Humanities and Natural Science courses and curricula, policies that would strengthen and support institutions, countries and regional collaborations should be pursued by these higher education networks.

Practical implications
The innovations in knowledge creation could be enhanced when management of these higher education networks encourage and prefer mutual networks to asymmetric networks. Furthermore, since centrality dimensions of social networks are inferior to knowledge creation, pro-structural whole dimensions of social networks should be pursued.

Limitations
As is the case with any other research work, this study has its limitations. First, results from the study are based on cross-sectional data. Authors are aware of the dynamic nature of social networks and that several analyses could be conducted from time to time using the crosssectional data. Second, the study ignored other measures of creation of knowledge such as books, citations, and patents and only used the number of publications. These measures are also vital when it comes to knowledge creation. However, the approach used in this study has been appropriately used in other prior studies. Again, the study only considered centrality and the structural holes dimensions of social networks. Measures such as relational, reputation, externalinternal index were not considered. Furthermore, the threshold used dropped more than two authors from the calculations of network variables. While co-authorship includes more than two authors, the use of only one or two authors still captured 85% of articles in the WOS database used for the study. Finally, the Web of Science provides the data source for this study. The database generates some natural bias since it favors English-speaking countries. These limitations notwithstanding, the study offers practical value to both policy makers and the research community.