MAGNET · COST Action CA25124
Background of MAGNET
A longer statistical history. A shared need for practical distributional modelling.
Why MAGNET?
Health decisions often depend on more than an average: whether a measurement crosses a centile, how frequently a threshold is exceeded, how variable an outcome is, or how uncertain a forecast may be. People, animals and environments can have similar average outcomes while facing very different risks.
MAGNET addresses the gap between established distributional methods and their routine applied use. The network connects statistical expertise with clinical, veterinary, environmental and public-health knowledge, helping researchers and practitioners use models that reflect the features of their data.
Distributional regression has a longer history
Modelling variation, quantiles and changing distributional features predates GAMLSS. Its development includes several related statistical traditions rather than a single starting point. GAMLSS is one flexible framework within this broader landscape.
The milestones below are a selected route towards MAGNET, not a complete history of distributional regression. The final milestone describes the Action’s ambitions rather than completed achievements.
1978 · Related foundations
Regression quantiles
Koenker and Bassett’s regression quantiles provided a foundational way to study conditional quantiles. This is one strand of the wider history of modelling beyond conditional means; quantile regression does not require specifying a complete parametric distribution.
1992 · Centile modelling
The LMS method
Cole and Green described smoothed reference centile curves using three changing parameters for median, variation and skewness. Their work is an important milestone in the development of modern growth and reference charts.
2005 · GAMLSS framework
Location, scale and shape
Rigby and Stasinopoulos presented the GAMLSS framework, allowing explanatory variables to influence multiple parameters of a response distribution through flexible predictors. This marks a major GAMLSS milestone within a much longer statistical history.
2007 and onwards · Software
An expanding open ecosystem
Stasinopoulos and Rigby documented GAMLSS in R in 2007. The ecosystem has continued to develop through open software and related distributional modelling approaches, including Bayesian and boosting-based implementations.
2026 · MAGNET begins
The first pan-European GAMLSS-focused project
MAGNET is the first pan-European project specifically focused on GAMLSS, building on earlier methodological developments, software and applications. COST Action CA25124 runs from 12 October 2026 to 11 October 2030 and connects methodological and applied communities across Public and One Health, infectious-disease modelling and predictive risk modelling.
2028–2030 · Planned milestones
From reusable workflows to routine use
The MoU sets targets of at least four reusable workflows by Month 24 and routine use of a workflow by at least two partners by Month 48. These are planned outcomes, supported by training, pilots and a long-term community strategy.
What changes when we look beyond the mean?
CENTILES
Where does an observation sit?
Reference curves can locate a measurement within a relevant population distribution, helping interpret age-related growth or other changing measurements.
THRESHOLDS
How likely is an exceedance?
A fitted conditional distribution can estimate the probability of crossing a clinically or environmentally meaningful threshold.
HETEROGENEITY
How does variation change?
Distributional models can examine differences in spread and shape, including skewness, heavy tails or excess zeros when a suitable model is chosen.
UNCERTAINTY
What range of outcomes is plausible?
Predictive distributions and intervals can express uncertainty around future observations. Their usefulness depends on validation, calibration and the quality of the underlying data.
From statistical capability to shared practice
MAGNET’s contribution is to connect expertise, coordinate accessible workflows and strengthen skills across disciplines. Shared examples, guidance, software and stakeholder dialogue are intended to make distributional modelling easier to understand, evaluate and apply.
Sources and further reading
CA25124 — official COST Action page
MAGNET Memorandum of Understanding and Technical Annex (PDF; university-hosted copy)
Koenker & Bassett (1978) — Regression Quantiles
Cole & Green (1992) — The LMS method and penalized likelihood
Rigby & Stasinopoulos (2005) — Generalized additive models for location, scale and shape
