Teacher burnout is often addressed with broadly similar wellbeing programmes offered to everyone, even though the combination of exhaustion, detachment, rumination and low efficacy can look very different from one teacher to another. A cluster-randomised trial across 84 public schools tested a more personalised alternative: use repeated measurements of each teacher’s emotional state to identify which parts of that individual’s psychological network appear most influential, then prioritise brief interventions around those targets.
The result was a measurable advantage over giving teachers access to the same intervention library without the personalised network information. By week seven, teachers receiving the network-targeted allocation reported happiness scores 4.4 points higher on a 0 to 100 scale than the active-control group. The estimated difference had a 95% credible interval from 2.7 to 6.0 points.
The study, published in Acta Psychologica, does not show that an app or algorithm can solve the organisational causes of teacher burnout. It does, however, provide a proof of concept for a different way of allocating support: instead of assuming the same intervention is equally relevant to everyone, the system tries to identify where a small change may have the greatest influence within each person’s pattern of experiences.
A trial built around everyday emotional patterns
Researcher Shideng Chen conducted a two-arm cluster-randomised platform trial in 84 public schools across four urban districts in H Province. The school-level randomisation is important because it reduces the risk that teachers within the same school receive different allocation strategies and inadvertently contaminate the comparison through shared experiences or discussion.
Before the intervention phase, teachers completed 14 days of ecological momentary assessment, or EMA. Rather than relying on a single retrospective questionnaire asking how someone had felt over a long period, EMA repeatedly captures experiences close to the time they occur. The baseline assessments covered happiness, exhaustion, detachment, efficacy and rumination.
Those repeated observations were used to estimate person-specific partial-correlation networks. In simplified terms, each psychological state was treated as a node, while the statistical relationships between states formed connections. The researchers calculated expected influence, a network measure intended to identify nodes whose pattern of positive and negative connections makes them potentially important within the system.
An optimisation engine then prioritised three types of brief micro-intervention for each teacher according to that baseline network information. The available intervention library included techniques such as cognitive defusion, values reminders and gratitude-related exercises. Teachers in the active-control arm received content from the same library, but their allocation did not use the person-specific network information.
This makes the comparison more informative than simply testing an intervention against no support. Both groups were exposed to intervention content. The experimental difference was whether the system used the teacher’s estimated emotional network to decide which content to prioritise.
Happiness improved more under network targeting
Ecological momentary assessment continued for eight weeks during the intervention period. The analysis used Bayesian multilevel models, permutation-based mediation and benchmarking against simpler allocation rules.
The network-targeted group showed a larger reduction in the study’s composite expected-influence change index than the active control. The mean between-group difference was 0.11, with a 95% credible interval from 0.08 to 0.14.
The more immediately understandable result appeared in everyday happiness. At week seven, the network-targeted group was estimated to be 4.4 points higher on the 0 to 100 EMA happiness scale, with a 95% credible interval of 2.7 to 6.0. The trajectory also differed over time: the arm-by-week slope difference was 0.62 points per week, with a 95% credible interval from 0.39 to 0.85.
These figures do not imply a dramatic transformation in teacher wellbeing. A difference of several points on a 100-point scale is modest. The relevance is that the experimental group received the same broad intervention library as the control group, so the observed advantage was associated with how support was selected rather than simply whether support existed.
About half of the happiness difference tracked network change
The researchers also examined whether changes in the network measure statistically accounted for some of the difference in happiness. Model-based mediation estimates were consistent with roughly half of the happiness difference being associated with change in the composite expected-influence index.
The average conditional mediation estimate was 3.5 points, with a 95% credible interval from 2.0 to 5.2. This is best interpreted cautiously. Statistical mediation can show that changes move together in a pattern consistent with a proposed mechanism, but it does not prove that altering the network itself caused the improvement in happiness.
That distinction is particularly important in psychological network research. The connections are estimated from repeated observations and can help describe the structure of a person’s reported experiences, but they are not direct measurements of causal pathways in the brain or workplace.
The study nevertheless provides an unusually concrete test of the centrality idea. Instead of using network analysis only to describe burnout after the fact, the researchers used the baseline network to make an intervention decision and then compared that strategy with an active control.
Simpler targeting rules produced smaller gains
The team benchmarked the network-informed approach against simpler rules based on severity, thresholds or group-level centrality. Those alternatives produced smaller gains in the study’s analyses.
This addresses a practical question. Personalisation is only valuable if the additional measurement and modelling produce information that simpler approaches do not. If targeting the most severe symptom were equally effective, building person-specific networks would add complexity without much benefit.
The trial also found evidence that personality mattered. Effects were stronger among teachers higher in conscientiousness, while the reported results did not show the same moderating pattern for agreeableness. That suggests personalisation may eventually need to extend beyond selecting psychological targets to understanding who is most likely to engage with particular forms of support.
Precision support is not the same as fixing working conditions
There is an important policy boundary around these findings. Teacher burnout can arise from workload, role conflict, staffing pressures, administrative demands and broader organisational conditions. A personalised psychological intervention may help an individual respond to strain, but it cannot by itself reduce class sizes, change workloads or repair dysfunctional management.
That matters when interpreting the promise of precision mental health in education. A system that becomes better at helping teachers cope should not become a reason to ignore preventable workplace demands. Individual and organisational interventions address different parts of the problem.
Recent evidence mapping also shows how heavily teacher-burnout intervention research has concentrated on individuals and classroom practice rather than organisational change. The present trial therefore sits within a field where personalised support may become more sophisticated even while structural intervention remains comparatively under-studied.
The design is promising, but the evidence remains early
Several limitations keep the findings in proof-of-concept territory. The study was conducted in public schools in one provincial setting, so the results may not generalise to different education systems, school cultures or labour conditions.
The network itself was built from a limited set of measured states. A sensitivity analysis in a subgroup added anxiety, stress, workload and organisational support, illustrating how network estimates can depend on which variables researchers choose to include. Psychological experiences that are not measured cannot become nodes, regardless of how important they may be in a teacher’s real life.
Contemporaneous partial correlations also do not establish temporal causality. Changes in network summaries can reflect genuine psychological change, but they may also be influenced by measurement variation, regression toward the mean and the statistical properties of the estimated network.
The eight-week intervention period is another constraint. The trial can show that the targeted allocation was associated with better outcomes over the observed period, but longer follow-up is needed to establish whether the advantage persists after repeated measurement and intervention prompts end.
There is also a practical implementation cost. Person-specific targeting depends on frequent data collection, sufficient participation to estimate stable patterns, analytical infrastructure and a delivery system capable of translating those estimates into intervention choices. Districts would need to know whether the incremental benefit justifies that complexity.
From generic wellbeing programmes to adaptive support
The broader contribution of the trial is not a claim that one network metric has solved teacher burnout. It is the demonstration that the allocation of wellbeing support itself can be experimentally tested.
Most workplace programmes ask whether an intervention works on average. Precision approaches ask an additional question: which intervention should be offered to this person, at this point, given the pattern visible in their own data?
Across 84 schools, using individual emotional-network information to answer that question produced better measured outcomes than distributing the same intervention content without that information. The next test is whether the effect can be reproduced in other regions, maintained over longer periods and implemented without shifting attention away from the working conditions that contribute to burnout in the first place.
Source Information
Study Title: Precision targeting of teacher burnout using network-informed ecological momentary interventions
Author: Shideng Chen
Journal: Acta Psychologica
Volume: 269
Article: 107537
Publication: September 2026 issue; electronically published 14 August 2026
DOI: 10.1016/j.actpsy.2026.107537
Design: Two-arm cluster-randomised platform trial across 84 public schools in four urban districts, with a 14-day EMA baseline followed by eight weeks of EMA and micro-interventions
Main finding: Network-informed allocation produced a 4.4-point higher week-seven happiness score on a 0 to 100 scale than content-matched active control, alongside a larger reduction in the composite expected-influence change index.
