Expertise diversity of teams predicts originality and long-term impact in science and technology

TL;DR

This paper addresses how the prior diversity of team expertise relates to originality and long-term impact in science and technology. It introduces a novel Expertise Distance metric that uses authors' field distributions of prior work and a field interaction matrix to compute pairwise distances between coauthors, with team diversity defined as the average of these distances. Applied to 22.8 million papers and 4.4 million patents from the MAG dataset (1950–2019), the approach finds that expertise-diverse teams produce more original work (higher disruption) and enjoy a long-term impact premium at 10 years, driven in part by cross-disciplinary influence; short- and mid-term impact show no consistent advantage. The findings carry implications for funding and team assembly, showing that fostering expertise diversity—especially within constrained institutional or national contexts—can promote enduring, cross-disciplinary impact, while acknowledging correlational limitations and data caveats.

Abstract

Despite the growing importance of teams in producing innovative and high-impact science and technology, it remains unclear how expertise diversity among team members relates to the originality and impact of the work they produce. Here, we develop a new method to quantify the expertise distance of researchers based on their prior career histories and apply it to 23 million scientific publications and 4 million patents. We find that across science and technology, expertise-diverse teams tend to produce work with greater originality. Teams with more diverse expertise have no significant impact advantage in the short- (2 years) or mid-term (5 years). Instead, they exhibit substantially higher long-term impact (10 years), increasingly attracting larger cross-disciplinary influence. This impact premium of expertise diversity among team members becomes especially pronounced when other dimensions of team diversity are missing, as teams within the same institution or country appear to disproportionately reap the benefits of expertise diversity. While gender-diverse teams have relatively higher impact on average, teams with varied levels of gender diversity all seem to benefit from increased expertise diversity. Given the growing knowledge demands on individual researchers, implementation of incentives for original research, and the tradeoffs between short-term and long-term impacts, these results may have implications for funding, assembling, and retaining teams with originality and long-lasting impacts.

Figures (5)

• Figure 1: Illustrating the estimation of expertise distance among team members and its correlation with multidisciplinarity. (a) We first indicate the prior career histories of two authors $1$ and $2$, distributed across three research fields $1$, $2$, and $3$. Then we display the expertise vectors $\boldsymbol{q_1}$ and $\boldsymbol{q_2}$ and their prior expertise distance $d_{12}$ in the geometric space of research fields, in which the field vectors $\boldsymbol{e_j}$ are usually non-orthogonal to each other due to the heterogeneity of interactions among fields, where $j \in (1,2,3)$. To estimate the expertise distance, we obtain the expertise vectors $\overline{\boldsymbol{q_i}}$ of unit length $1$ by normalizing the publication vectors $\overline{\boldsymbol{p_i}}$ of authors, where $i \in (1,2)$. The expertise vector $\boldsymbol{q_i}$ shown in the space of fields can be related to the expertise vector $\overline{\boldsymbol{q_i}}$ as $\boldsymbol{q_i} = \overline{\boldsymbol{q_i}} (\boldsymbol{e_1}, \boldsymbol{e_2}, \boldsymbol{e_3})^T$. We then compute the coauthor expertise distance $d_{12}$ by taking into account the interaction matrix $\boldsymbol{M}$ that explicitly embeds in the relatedness of fields. (b) The distribution of the average expertise distance among team members for over 11 million research papers published in 1970-2019. Then we mark the skin cancer classification paper using neural networks ($d = 1.00$, top $2\%$) that is more interdisciplinary in terms of both the composition of team expertise and the produced work, and the image recognition paper ($d=0.08$, bottom $5\%$) that is conducted by a team with members focusing primarily on a specific research area. Both papers are highly cited. (c) Contemporary trends of teams' expertise diversity for papers and patents, from 1970 to 2019. Shaded areas represent 95% confidence intervals. (d) To validate the efficacy of the distance metric, we show the interplay between teams' expertise distance percentile and the multidisciplinary inspiration (solid lines) and impact (dashed lines) of papers and patents. (e) We operationalize several alternative methods to assess expertise and show the correlation matrix between them and the expertise distance metric. (f) Probability distribution of all expertise metrics. (g) ROC curves of different expertise metrics predicting multidisciplinarity of teams' work.
• Figure 2: Originality and its correlation with expertise diversity of teams and multidisciplinarity of the teams' work. We present the interplay between expertise diversity of teams and disruption of the teams' work and find that high expertise diversity is correlated with a high disruption score for both papers (a) and patents (b). In contrast, we show the interplay between multidisciplinary inspiration and disruption of the teams' work. We find that multidisciplinary inspiration has no consistent correlation with disruption for papers (c), and appears to be negatively correlated with disruption for patents (d). (e) Contemporary trends of expertise diversity and geographic distance of international teams from 1970 to 2019. (f) Correlation between the average geographic distance and disruption of research by international teams. Shaded areas represent 95% confidence intervals.
• Figure 3: Distance metric and impact in science and technology. We show the interplay between research impact and the teams' expertise diversity divided into five uniform distance percentile bins. We consider three different lengths of citation time windows of two years, five years, and ten years after publication. We identify three existing citation patterns among science and technology: (a-d), paper-to-paper citations; (e-h), patent-to-paper citations; and (i-l), patent-to-patent citations. For each pattern, results are presented in separate panels categorized by the teams' size. Shaded areas represent $95\%$ confidence intervals.
• Figure 4: Teams' expertise diversity and origins of citations in science and technology. We decompose the citation origins of fields for scientific and technological research according to the teams' expertise diversity. We compare the number of citations coming from the same field or from different fields, accrued within two different time windows of 2 years and 10 years after publication, respectively. Consistent with previous results, we use three citation patterns among science and technology: (a-d), paper-to-paper citations; (e-h), patent-to-paper citations; and (i-l), patent-to-patent citations. For each pattern, results are presented in separate panels categorized by the team's size. Shaded areas represent $95\%$ confidence intervals.
• Figure 5: The synthesized effects of teams' expertise diversity and other dimensions of team diversity on long-term impact. We present the temporal trends of team expertise diversity and its correlation with disruption and two types of impact in the long run, conditional upon other dimensions of diversity among team members including affiliations, nationality, and gender. (a-c) Temporal trends of team expertise diversity based on the diversity of team members' affiliations, nationality, and gender. The effect of team diversity and expertise diversity on (d-f) disruption, (g-i) paper-to-paper citations, and (j-l) patent-to-paper citations. In particular, we consider three types of diversity among team members. Left column, whether it is a single-institutional or between-school team. Middle column, whether the team involves international collaborators. Right column, whether the team is gender-diverse or gender-homogeneous. Teams with diverse background collaborators, i.e., having between-school collaborations, having international collaborations, or having female researchers on board, are indicated by solid lines. Teams with simple background collaborators are shown in dotted lines. Shaded areas represent $95\%$ confidence intervals.