Risk and Possible Solution Associated with Sampling in Co-authorship Network

Abstract

There is already a reasonable set of papers that applied sociometry and network theory to characterize the arrangement of researchers and/or research institutions underlying a specific object of interest, whether it be a Postgraduate Program, periodical, conference, or theoretical or technical field. If the object of interest is circumscribed, the census is reasonable and can be employed. Otherwise, you work with a fraction of the sample population. The use of such a sample has risks that, in theory, may compromise the findings. To meet this study’s objectives – to evaluate the risks arising from the use of samples in co-authorship network analysis and to propose a more adequate approach than to simply disregard them – 300 simulations of a co-authorship network were made, reducing it by 5, 10, 15, 20, 25 and 30% to subsequently test the extrapolation from the linear model. Results indicate that even relatively large samples can be misleading. However, it was possible to infer some of the structural characteristics of the population from the method under consideration, in such a way that this development can turn out to be an interesting resource to confer greater reliability to the research in the area.

Downloads