Sheffield Elicitation Framework (SHELF)
SHELF helps experts express their beliefs about uncertain quantities by turning their judgments into probability distributions. It supports individual and group input and provides visual feedback to refine those judgments.
At a glance
Use when
Empirical data is scarce or unavailable; quantifying uncertainty in model parameters; incorporating expert knowledge in decision-analytic models; supporting regulatory or health technology assessment submissions
Avoid when
Sufficient high-quality data exists to estimate parameters directly; experts lack relevant knowledge or are unwilling to participate; time or resources for proper elicitation training and facilitation are unavailable
Inputs
Expert-provided probability judgments (e.g., quantiles or cumulative probabilities), expert weights (for group elicitation), choice of distribution family
Outputs
Fitted probability distribution(s), graphical feedback (e.g., CDFs, percentiles), pooled distributions for multiple experts, survival extrapolations
How it works
SHELF implements structured expert judgment methods for eliciting probability distributions. Experts provide quantile-based probability assessments, which are used to fit parametric distributions with graphical feedback. Methods include fitting univariate distributions, weighted linear pooling for multiple experts, population distribution elicitation, multivariate distributions via Gaussian copula, Dirichlet distributions, variance components in meta-analysis, and survival extrapolation. R Shiny applications are available for interactive use.
- HTA domains
- Clinical Effectiveness, Patient and Social Aspects, Aspects Beyond HTA
- Assumptions
- Experts are well-calibrated and have relevant knowledge; their judgments can be represented by specified parametric forms; independence assumptions may apply in multivariate or hierarchical settings
- Strengths
- Provides structured, transparent, and reproducible expert elicitation; includes validation and feedback mechanisms; supports diverse elicitation contexts and multiple experts; integrates with modeling workflows via R
- Limitations
- Relies on cognitive ability and honesty of experts; potential for overconfidence or bias; fitting parametric distributions may misrepresent true beliefs if poor fit; setup requires training and facilitation
- Also known as
- SHELF, Sheffield Elicitation Framework
Questions this answers
- › What is the expert's belief about the likely value of an uncertain parameter?
- › How can multiple experts' judgments be combined into a single distribution?
- › What is the probability that a parameter exceeds a certain threshold, according to expert judgment?
- › How can uncertainty in model inputs be quantified when data is limited?
- › How can expert beliefs be formally incorporated into decision models?
- › What are plausible long-term survival trends when empirical data is sparse?
References & sources
Similar by meaning
Beta record. Generated from the primary source via AI extraction and independent audit, pending final human review.