This article discusses the surname analysis of 2.5 million research papers in which all of the authors had US addresses. Our study showed that US-based authors with English surnames were more likely to have co-authors with English surnames than would occur by chance; those with Chinese names were more likely to have co-authors with Chinese names, and so on. The trend held for seven other groups, including Russian and Korean populations, between 1985 and 2008 in 11 scientific fields, including biomedicine, physics and geosciences. The results hint that scientific research is much like the rest of social life. Studies of social networks find that people eat with, work with and generally connect with others similar to themselves, a tendency that some sociologists call homophily. To the extent that surnames can be a proxy for ethnicity, homophily in scientific collaborations also seems to be related to a work's reception in the scientific community. After controlling for numbers of authors and for factors such as an ethnic groups' population density, we find that greater ethnic homogeneity among authors is associated with a paper's publication in lower-impact journals. It also predicts fewer citations. Papers with four or five authors of multiple ethnicities have, on average, one to two more citations than those written by authors all of the same ethnicity. This effect represents a 5–10% difference in the mean number of citations for a given publication.

Editor's Note: This article is part of a special collection of invited articles: "Diversity: A Nature & Scientific American Special Issue". See Related Materials in right menu for a link to the full collection.