Academic collaborations and movements towards successful careers in physics

Abstract

Collaboration networks evolve throughout academic careers, yet few studies systematically examine how these network dynamics relate to long-term career success and mobility. Analysing 35,708 physicists' careers spanning at least 15 years, we use time series clustering to identify ten distinct evolution patterns of network size and clustering coefficient across career years 5 to 15. We report three key results. First, authors who begin with loosely connected networks and progressively tighten their networks while expanding network size during mid-career achieve the highest PI attainment rates, publication output, and citation impact. Second, despite different starting points, network evolution patterns associated with better outcomes converge toward moderate clustering by career year 15, suggesting an optimal balance between core team cohesion and diverse external connections. Third, mobility is positively associated with these successful network evolution patterns and remains positively associated with scientific outcomes even after controlling for network evolution patterns.

Figures (8)

• Figure 1: Network evolution and outcome based clustering. (A) We track authors' collaboration network evolution between the fifth and fifteenth year since their first publication. (B) Four outcome dimensions assess whether (i) scientists reached PI status, (ii) the time it took to reach it, (iii) their productivity and (iv) impact. (C) We apply Time Series K-Means clustering on the network evolution of all 35,708 authors to recursively assign them to one of two groups. Permutation tests detect significant differences between two groups in each of the four outcome dimensions. The splitting is repeated for each group until no significant difference can be detected for each of the four outcome variables. The last valid split is then considered as one of the final network evolution clusters (black border and marker).
• Figure 2: Evolution of adjusted network size and clustering coefficient across 10 clusters. Each subplot shows the trajectories of one cluster for adjusted network size (NS, blue solid line) and adjusted clustering coefficient (CC, orange solid line) over career years 5--15. Solid/hollow markers indicate adjusted network size trends (increasing/decreasing); marker shapes represent clustering coefficient patterns. The vertical dashed line separates clusters by clustering coefficient trends (left: $\searrow$ decreasing, right: $\nearrow$ increasing), while the horizontal dashed line separates clusters by network size trends (bottom:$\nearrow$ increasing, top: $\searrow$ decreasing). The figure is divided into four quadrants based on trajectory trends: Panel A (shrinking and loosening, top-left): NS$\searrow$, CC$\searrow$; Panel B (shrinking and tightening, top-right): NS$\searrow$, CC$\nearrow$; Panel C (growing and loosening, bottom-left): NS$\nearrow$, CC$\searrow$; Panel D (growing and tightening, bottom-right): NS$\nearrow$, CC$\nearrow$.
• Figure 3: Network evolution patterns and associated scientific outcomes. Panel A shows mean changes in adjusted network size and adjusted clustering coefficient from career year 5 to year 15 for each cluster. Each point represents a distinct network evolution pattern, as shown in Figure \ref{['fig:network_evolution_patterns']}, with reference lines showing overall means. Colours indicate mean mobility score for each cluster (indigo: low mobility, grey: medium mobility, gold: high mobility). Panels B-E compare scientific outcome measures across clusters: (B) likelihood of becoming a PI, (C) time to become a PI, (D) total publication output, and (E) citations per paper. Error bars represent standard errors of the mean. Clusters are ordered by performance within each outcome measure and the reference lines indicate the global mean.
• Figure 4: Early and mid-career mobility patterns across network evolution clusters. The scatter plot shows mean mobility scores with error bars indicating standard errors. Each point represents a cluster's average early-career mobility (x-axis) and mid-career mobility (y-axis). Reference lines indicate overall population means. Cluster markers correspond to network evolution patterns shown in Figure \ref{['fig:network_evolution_patterns']}.
• Figure 5: Mobility effects on scientific outcomes across network evolution clusters. Clusters are grouped by network evolution based on changes in adjusted network size (NS: $\nearrow$ increasing, $\searrow$ decreasing) and adjusted clustering coefficient (CC: $\nearrow$ increasing, $\searrow$ decreasing). Forest plots show regression coefficients with 95% confidence intervals for early-career mobility (purple lines) and mid-career mobility (teal lines) effects within each cluster. Panel A: Logistic regression coefficients for likelihood of becoming a PI. Panel B: Linear regression coefficients for years to PI attainment. Panel C: Linear regression coefficients for total publications. Panel D: Linear regression coefficients for citations per paper.