Bodong Chen

Crisscross Landscapes

Notes: Lungeanu & Contractor. (2015). The effects of diversity and network ties on innovations

2017-08-14


References

Citekey: @Lungeanu2015-mb

Lungeanu, A., & Contractor, N. S. (2015). The effects of diversity and network ties on innovations: The emergence of a new scientific field. The American Behavioral Scientist, 59(5), 548–564. https://doi.org/10.1177/0002764214556804

Notes

Summarize: Very interesting study investigating factors impacting (reflected in) coauthorship relations in a new field. SIENA was used for the modeling of network dynamics. Key finding:

Taken together with net findings about cognitive dissimilarity, this study suggests that innovation in networked research is most likely when teams draw on individuals with diverse knowledge resources, but who also have prior collaborations with each other or with common others. The key to sustaining innovation in such teams is to reduce the possibility that close collaboration results in a reduction of the diversity in their knowledge expertise

Reflect: Useful for ongoing work on coauthorship in ET and MOOC.

Wondering how to model interdependencies between the citation and coauthorship networks. Also wondering that publications have substantial latency, happening way after the actual collaboration. Wondering whether it would be helpful to see more real-time data (Slack?). It depends on how one conceptualizes collaboration. I guess publication is the most direct way to do it.

Highlights

Abstract This study examines the influence of different types of diversity, both observable and unobservable, on the creation of innovative ideas. Our framework draws on theory and research on information processing, social categorization, coordination, and homophily to posit the influence of cognitive, gender, and country diversity on innovation. Our longitudinal model is based on a unique data set of 1,354 researchers who helped create the new scientific field of oncofertility, by collaborating on 469 publications over a 4-year period. We capture the differences among researchers along cognitive, country, and gender dimensions, as well as examine how the resulting diversity or homophily influences the formation of collaborative innovation networks. We find that innovation, operationalized as publishing in a new scientific discipline, benefits from both homophily and diversity. Homophily in country of residence and working with prior collaborators help reduce uncertainty in the interactions associated with innovation, while diversity in knowledge enables the recombinant knowledge required for innovation. (p. 1)

This study focuses on the dilemma of diversity for fostering innovation in networked research. On the one hand, generating innovative ideas requires the ability for recombinant searches (Fleming & Sorenson, 2001) across diverse areas of the knowledge possessed within the team (West, 2002). On the other hand, it requires team members who are comfortable work- ing with each other (e.g., Guimera, Uzzi, Spiro, & Amaral, 2005; Taylor & Greve, 2006). We seek to address this dilemma by parsing the effects of different dimensions of diversity and network ties on innovation in networked research. More specifically, we examine the influence of cognitive, gender, and country diversity (or its opposite, homophily), as well as prior network ties on the emergence of collaborative networks of innovation over time. (p. 2)

Theoretical Background and Hypotheses (p. 2)

Research on teams has investigated several dimensions of diversity that affect creativ- ity and innovation (Cady & Valentine, 1999; Joshi & Roh, 2009). These dimensions have been broadly considered in two categories: whether they are relatively unobserv- able (such as cognitive or expertise diversity) or observable (such as gender and coun- try diversity). (p. 3)

Cognitive diversity. (p. 3)

Williams and O’Reilly (1998) noted that diversity of knowledge is useful for idea generation and performance in teams. (p. 3)

Hypothesis 1: Individuals that possess disparate knowledge are more likely to pro- duce innovative ideas through collaboration. (p. 3)

Gender homophily. Self-categorization theory suggests that individuals characterize themselves using categories such as age, gender, and race (e.g., Turner, Hogg, Oakes, Reicher, & Wetherell, 1987). Individuals then select similar others in order to reduce the potential conflict in their relations (Byrne, 1971), or to reduce the psychological discomfort that may arise from cognitive or emotional differences (Heider, 1958). (p. 3)

Gender homophily represents one dimension of homophily (p. 4)

Hypothesis 2: Individuals of the same gender are more likely to produce innova- tive ideas through collaboration. (p. 4)

Country homophily. (p. 4)

Hypothesis 3: Individuals located within the same country are more likely to pro- duce innovative ideas through collaboration. (p. 4)

Methodology (p. 4)

Data and Sample (p. 4)

We tested these hypotheses using archival and bibliographic data about teams that col- laborated to publish scientific articles in the recently emerging field of Oncofertility. The term oncofertility, coined in 2007, refers to research on fertility preservation for cancer patients. (p. 4)

We identified all scientific articles that were published in the field of Oncofertility using the keywords oncofertility, or cancer and ovarian tissue cryopreservation, or cancer and fertility preservation. We used the Web of Science (Wo S) database to con- struct researchers’ coauthorship relations (p. 5)

The data set for this study utilized all publications between 2007 and 2010, and comprised 469 publica- tions from 1,354 researchers. (p. 5)

Author name disambiguation is a recognized issue when constructing bibliometric measures (Torvik, Weeber, Swanson, & Smalheiser, 2005). To address this limitation, we took a conservative approach in identifying an author’s non-Oncofertility publications, con- sidering only those publications with identical author names, e-mail addresses, and Digital Author Identification system numbers, a unique internal ID used by Wo S data- base to disambiguate authors. (p. 5)

Demographic information and country affiliation were manually coded (p. 5)

Variables (p. 5)

Our dependent variable is the collaboration network (p. 5)

Coauthorship of a journal publication is an important measure of researchers’ collaborative relation- ship (Guimera et al., 2005). Thus, based on publication information, we defined a coauthorship relation between two researchers if they published a scientific article together in the field of Oncofertility. (p. 5)

four undirected 1,354 by 1,354 coauthorship relation matrices, one for each year of observation (2007, 2008, 2009, and 2010). (p. 5)

Our analysis was confined to a binary, rather than a weighted coauthorship network, because stochastic actor-oriented models, used for our analysis, are currently only developed for binary dependent network relations (p. 5)

Control variables. (p. 5)

we control for the potentially confounding effects associated with network structures and individual researchers’ attributes. In terms of the former, we control for the endogenous and exogenous effects of net- work structures on coauthorship relations. (p. 5)

Endogenous network effects control for (p. 5)

the tendency of creating new coauthorship ties based exclusively on prior ties of the same kind, such as previous coauthorship in the area of Oncofertility. Exogenous network effects control for the tendency to explain the creation of new ties based on the existence of prior ties of a different kind, such as prior coauthorship on non- Oncofertility publications. Additionally, we also controlled for country and gender as these two attributes may affect a researcher’s proclivity to coauthor Oncofertility publications. (p. 6)

We used three network structures to control for the endogenous network effects of prior Oncofertility coauthorships relation to future coauthorship of related publica- tions: density, transitive triads, and the number of actor pairs at Distance 2. (p. 6)

den- sity controls for the overall “baseline” propensity of any two researchers in the field to coauthor a publication. Like the intercept term in a regression or a grand mean in ANOVA, this measure reflects the probability of two researchers coauthoring a publi- cation in the area of Oncofertility if nothing is known (p. 6)

Additionally, transitive triads control for researchers’ proclivities to coauthor Oncofertility publica- tions with their prior coauthors’ prior coauthors and focuses on the formation of new coauthorship ties, while the number of actor pairs at Distance 2 controls for the inverse tendency of researchers to keep their coauthors’ prior coauthors at two degrees of separation by not coauthoring with them and focuses on the dissolution of ties in order to create a distance of 2. (p. 6)

We used prior coauthorship of non-Oncofertility publications by researchers as an exogenous network control for future coauthorship of publications in the area of Oncofertility. (p. 6)

we extracted each scientific article published by the 1,345 researchers in the data set prior to the start of their Oncofertility collaborations. (p. 6)

Hypothesized variables. Our three hypothesized independent variables are cognitive similarity, gender similarity, and country similarity. Cognitive similarity was measured as citation similarity between two researchers using the Jaccard similarity coefficient. (p. 6)

Next, we coded the researcher’s gender as 1 for female and 0 for male and we identified a total of 34 unique countries, with the United States of America being the most represented (35.59%, see the appendix). (p. 6)

gender similarity and country similarity were calculated as 1 or 0, with 1 indicating a pair of actors having the same gender, or being from the same country. (p. 7)

Analysis (p. 7)

The diversity and homophily hypotheses were tested using SIENA, stochastic actor- oriented models for network dynamics (Snijders, Van de Bunt, & Steglich, 2010). This statistical model simulates the evolution of the network, using discrete observations over time, and estimates parameters for the underlying mechanisms of network dynam- ics between these discrete, incremental observations by combining random utility models, Markov processes, and simulation. (p. 7)

read SIENA, Snijders et al 2010 (p. 7)

The SIENA model is appropriate, as the dependent variable innovation is operationalized as a network tie, consisting of coau- thorship among two individuals publishing in the field of Oncofertility. In addition, an examination of the emergence of coauthorship ties requires a longitudinal approach which is offered by this model. (p. 7)

Results (p. 7)

Oncofertility Network Evolution and Descriptive Statistics (p. 7)

87% of the publi- cations reviewed were the result of multiauthor collaborations. Our results also show that 21% of researchers were involved in more than one team, a significant number given that we only observed teams whose work was published over a 4-year period. (p. 7)

Table 1 reports the changes within the Oncofertility col- laboration network over time, with the number of ties changing from 0 (no collaboration) to 1 (collaboration). Here we find that changes from 0 → 1 are greater than the number of ties changing from 1 → 0 between subsequent years. Thus, each year, more ties were created than dissolved. These results indicate that the network grew over the duration of our study, a result evident in Figure 1 (p. 7)

As demonstrated in Figure 1 and Table 2, the density of the Oncofertility coauthor- ship relation is very low in all 4 years. (p. 8)

Hypotheses Testing (p. 9)

Table 3 reports the correlation matrix and Table 4 the results of the coevolution model based on stochastic actor-based models. (p. 9)

Hypothesis 1 predicted that individuals possessing expertise in different knowledge or research areas are more likely to produce innovative ideas through collaboration. The results of the model indicate a negative and significant effect for cognitive simi- larity (−0.04, p < .05). A negative effect indicates dissimilarity in knowledge and hence such a result supports Hypothesis 1. (p. 9)

Hypothesis 2 predicted that individuals are more likely to achieve innovation with others of the same gender, drawing on arguments about attraction and ease of com- munication among similar people. However, we did not find support for this hypoth- esis (−0.02, p > .05), suggesting that diversity based on gender in not necessarily beneficial to collaboration. (p. 10)

Finally, Hypothesis 3 predicted that individuals are more likely to achieve innova- tion when collaborating with others within the same country. This hypothesis was supported (0.91, p < .05), revealing that despite the advance of ICTs and researchers’ familiarity with technology, distance continues to inhibit the creation of scientific innovation by teams (Cummings & Kiesler, 2005). (p. 10)

We found density to be significant and negative (−2.56, p < .05), suggest- ing that the costs of collaboration preempt researchers from creating coauthorship ties with random others. The transitive triad parameter on the other hand was found to be positive and significant (0.64, p < .05), suggesting that there is indeed a tendency for researchers to collaborate with “friends of their friends,” or more accurately prior coauthors’ of their prior coauthors. Consistent with this finding, estimates for the num- ber of actor pairs at Distance 2 is negative and significant, signaling a tendency toward network closure. (p. 11)

We also controlled for the exogenous network effects of prior coauthorship on non- Oncofertility publications. The results show that researchers are more likely to engage in new collaborations in this field with those whom they have previously collaborated with on non-Oncofertility collaborations (0.47, p < .05). (p. 11)

Discussion (p. 11)

Taken together with net findings about cognitive dissimilarity, this study suggests that innovation in net- worked research is most likely when teams draw on individuals with diverse knowl- edge resources, but who also have prior collaborations with each other or with common others. The key to sustaining innovation in such teams is to reduce the possibility that (p. 12)

close collaboration results in a reduction of the diversity in their knowledge expertise (Lungeanu et al., 2014). (p. 13)

networked research continues to reflect individuals’ natural proclivities to collaborate with others within their own dis- cipline and local unit (Dimitrova et al., 2013). (p. 13)

Chen, W., Rainie, L., & Wellman, B. (2012). Networked work. In H. Rainie & B. Wellman (Eds.), Networked: The new social operating system (pp. 171-196). Cambridge: MIT Press. (p. 14)

Dimitrova, D., Gruzd, A., Hayat, Z., Mo, G. Y., Mok, D., Robbins, T., … Zhuo, X. (2013). NAVEL gazing: Studying a networked scholarly organization. In E. Kranakis (Ed.), Advances in network analysis and its applications (pp. 287-313). New York, NY: Springer. (p. 14)