Throughout the last decades, tightening budgets and an increasing competition between research projects have stimulated the development of new approaches towards research evaluation. Yet, despite an important evolution, evaluations still have a critical limitation: the boundaries of the unit of analysis are established based on administrative, institutional or geographic contexts of its members rather than endogenous factors (such as co-authorship and collaboration patterns), overlooking the unique and self-organizing characteristics of the research endeavor. This approach stands in contrast with a growing recognition of the importance of the research group as the critical organizing force of the scientific endeavor. To overcome these limitations, this study develops and tests a research evaluation method that recognizes the self-organizing characteristics of research groups. Instead of ad-hoc definitions of the unit of analysis, the study utilizes patterns of collaboration and the specific body of knowledge that these collaborations entail to identify the frontiers of the focal units, as well as other units that qualify as relevant benchmarks. First, the boundaries of a research group (RG) are identified based on level of cohesiveness of the co-authorship patterns. Second, backward citations (found in the work published by each group) are used to establish its "knowledge footprint" and assess the degree of similarity between research groups. Once the groups are characterized and their peers identified, the performance and productivity of each RG are measured and benchmarked using publications and citations. The method is tested, first, by ranking groups within and across Mexico and Brazil; the new method is utilized with these countries because of an ongoing partnership (through the Principal Foreign Collaborator) with the Mexican Council for Science and Technology (CONACYT) and of the availability of the database of published papers in these countries. Once the method is tested and validated equivalent analyses for a few areas within the U.S. are performed. Two main results from this research are expected. First, this study develops a research evaluation method that identifies key research groups based on the strength and frequency of the collaboration patterns and contrast their performance against its peers at different levels of knowledge similarity. Second, this work identifies key research groups in the US, Mexico and Brazil; and benchmark their performance with other groups within and across countries. This study provides an interdisciplinary approach that combines in a creative way bibliometric and network analysis, to create a new research evaluation tool. In addition, the concept of evaluating and benchmarking research groups based on different levels of similarity in the knowledge footprint is an important departure from the current literature. Finally, the method is quite generic and can be extended to other type of focal units, as well as used with granted patents, to identify research and development or innovation teams across academia and industry. There are broader implications of this study for research administrators and evaluators. The study identifies and benchmarks key research groups within each country. A careful account of the cohesiveness of the patterns of co-authorship combined with the knowledge footprint of the research groups will yield more meaningful and precise evaluations. In addition, this method opens the institutional black box, helping these actors to note the differences between groups; and compare the performance of a group within similar entities.
