Page - (000260) - in Knowledge and Networks

256 To examine any field-specific relationships between novelty, conventionality, and hit papers, we calculated the subfield-specific probabilities of a “hit” by the four categories used in Fig. 12.2 and defined in the text. We then ranked these four cat- egories in each subfield, where 1 indicates the highest probability of a hit, 2 indi- cates the second highest probability of a hit and so on. Consistent with the main results, Table 12.2 shows that in 64.4 % of fields, a paper’s likelihood of being a hit paper was greatest when combining prior work characterized by high tail novelty and high median conventionality. This category (Row 3 in Table 12.2) is ranked first or second in 86.3 % of subfields. Notably, to the extent that this category is not dominant within a subfield, the category featuring a more general shift toward nov- elty (Row 1 in Table 12.2) appears prominently, suggesting that tail novelty is an especially generic feature of the highest-impact papers. Conversely, the category featuring low tail novelty and low median conventionality (Row 4 in Table 12.2) ranks lowest in 70.4 % of subfields. Thus, novelty and conventionality are not opposing factors in the production of science; rather, papers with an injection of novelty into an otherwise exceptionally familiar mass of prior work are unusually likely to have high impact. Next, we focus on the effect of teams on novelty. Fig. 12.7 Citation impact results generalize to broader definitions of left tail novelty. The figure presents the relationship between tail novelty and impact using alternative definitions of tail nov- elty. In each case, tail novelty is defined as an indicator for whether the eth percentile of a paper’s z-score distribution is less than zero. The x-axis indicates the value of e. It is seen that for e ≤ 20, high tail novelty combined with a high median conventionality (Green bars) outperforms other categories. The results in the main text, which use the 10th percentile, thus extend broadly to other definitions of tail novelty so long as the measure emphasizes the paper’s left tail of combinations. From Uzzi et al. (2013b, p. 14). Copyright 2013 by Science. Reprinted with permission from the authors and Science S. Mukherjee et al.