A centralised Research Software Engineering (RSE) team is likely to provide stronger organisational resilience and lower burnout risk than models in which RSEs are embedded as isolated individuals across separate projects or departments. Centralisation does not eliminate project risk, but it changes the risk structure from person-dependent to team-dependent, which is generally more manageable at the institutional level.
The primary resilience advantage of a central team is that knowledge, responsibilities, and specialist skills are distributed rather than concentrated in one person per project. In embedded models, an individual may become the sole holder of technical context for software architecture, deployment, testing workflows, and operational handover, creating a single-point-of-failure risk. If that individual leaves or is reassigned, software continuity can deteriorate quickly. Centralised teams reduce this fragility by building redundancy in capability: work can be reallocated when needed, peer review is more available, and key practices are more likely to be standardised across projects.
Figure 1: Composite resilience score by survey year for respondents with and without a dedicated RSE group. The score is defined as the proportion reporting a transition plan minus the proportion reporting bus-factor risk (bus factor <= 1), so higher values indicate stronger resilience. Across all survey years, the dedicated-group line remains above the embedded-group line, indicating a consistently more favorable resilience profile for dedicated team structures.
Figure 1 demonstrates a consistent resilience advantage for dedicated RSE groups over embedded models across all survey years. Composite resilience, calculated as transition-plan percentage minus bus-factor risk percentage, is higher for the dedicated group in every wave, though both groups trend less favourably over time. Figure 2 clarifies the source of this advantage by decomposing the composite gap into its two components: transition-plan advantage and bus-factor risk reduction. Both favour dedicated RSE groups in every survey year, indicating the resilience benefit is broad-based rather than driven by a single factor.
Figure 2: Component-level resilience advantage of dedicated RSE groups by survey year, shown as percentage-point differences between the two organisational models. The two components represent transition-plan advantage (Yes minus No) and bus-risk reduction (No minus Yes), with positive values indicating outcomes that favour dedicated groups. Across all years, dedicated-group settings show advantages on both measures, indicating better transition planning and lower exposure to single-point-of-failure risk.
The case for centralisation is strengthened further when staff wellbeing is considered. The primary burnout advantage of a dedicated RSE group is that professional identity, peer support, and institutional visibility are structurally reinforced rather than left to chance. In embedded models, RSEs may lack peers who share their technical frame of reference, need to advocate for recognition without institutional backing, and remain peripheral to both the research teams they support and any wider RSE community.
The composite burnout risk score, constructed from low satisfaction, high turnover intent, low supervisor recognition, and willingness to accept an equivalent job elsewhere (each normalised to a 0–1 scale) captures this pattern across four survey waves. As shown in Figure 3, from 2017 onward embedded RSEs show consistently higher burnout risk than those in dedicated groups by approximately 0.06 points: 0.330 versus 0.391 in 2017 (p = 0.001), 0.307 versus 0.368 in 2018 (p < 0.001), and 0.279 versus 0.342 in 2022 (p < 0.001). Bootstrap confidence intervals for the difference are narrow and consistent across all three waves, indicating a stable signal rather than a one-off fluctuation. The 2016 UK-only wave shows a non-significant difference (p = 0.565), plausibly explained by the smaller embedded sample size.
Figure 3: Mean composite burnout risk score (0–1) by survey year for dedicated and embedded RSEs, with 95% bootstrap confidence intervals shown as shaded bands. Sample sizes are reported beneath each survey year, and significance markers (ns, **, ***) indicate the results of two-tailed Mann–Whitney U tests comparing the two groups within each survey wave. From 2017 onward, embedded RSEs show consistently higher burnout risk than those in dedicated groups, while both groups show an overall decline in burnout risk over time.
Figure 3 also shows that absolute burnout risk declines over time for both groups, suggesting the RSE community more broadly has benefited from increased institutional recognition and a more established professional identity. However, the relative gap remains. As shown in Figure 4, it is broad-based across all four burnout components, with turnover intent and supervisor recognition contributing most strongly, suggesting that relational and cultural conditions, particularly the experience of being seen, recognised, and supported within a professional community, are central mechanisms behind the difference.
Figure 4: Component-level burnout gap by survey year, shown as the difference in mean score between embedded and dedicated RSEs on the 0–1 scale. Positive values (red) indicate dimensions on which embedded RSEs are at greater disadvantage, while negative values (blue) indicate the reverse. From 2017 onward, all four components, low satisfaction, high turnover intent, low recognition, and job-switch willingness, show positive gaps, with turnover intent and recognition contributing most strongly to the overall burnout difference.
Taken together, the resilience and burnout findings support the same strategic conclusion. Centralised RSE teams appear to strengthen organisational continuity while also reducing staff welfare risks associated with isolation, weak recognition, and single-person dependency. For decision-makers, this supports consolidating dispersed embedded roles into a central RSE function where feasible, while preserving strong domain links to research teams through clear engagement models.