The exploration in the size of scientific collaboration team using kernel density estimation

Purpose: Scientific collaboration is becoming a common pattern in the social organization of knowledge production. The paper tries to figure out the relationship between scientific collaboration team size and scientific output. Design/methodology/approach: Based on ESI database from year 2009–2019, the paper describes changes of collaboration team size from one author to more than 10 authors in 22 disciplines. Kernel density estimation and multidimensional kernel density estimation method are used to calculate optimal collaboration team size and appropriate collaboration team size in 22 disciplines. As bandwidth is one of the major issues in construction of kernel density estimation, the paper uses five different algorithms to calculate bandwidth. The method with the lowest mean absolute percentage error is chosen. Robustness test is conducted based on different sets of data. Findings: The results show that scientific collaboration becomes more widely and deeply. As time goes by, collaboration team size is becoming larger and larger. Natural science disciplines have larger collaboration team size and faster growth rate than social science disciplines. Considering both qualitative and quantitative measures, the paper proves the universality of optimal and appropriate scientific collaboration team size among 22 disciplines and calculates the specific number. Originality/value: The paper tries to investigate the law of scientific collaboration team size variation and provide a full picture of evolution of collaboration team size among 22 disciplines in 10 years. The paper first applies distribution method to figure out the relationship between scientific collaboration team size and scientific output and provides optimal collaboration team size and appropriate collaboration team size.