Concept Architecture
Concept
Theoretically, Sobol Indices are variance-based sensitivity measures that quantify the contribution of individual model input parameters and their interactions to the overall uncertainty in model outputs. They are founded on functional analysis of variance (ANOVA) decomposition and provide a comprehensive framework for global sensitivity analysis. In health economics, Sobol indices are used to identify influential parameters, quantify interaction effects and prioritise future research within complex decision models.
Mathematically, Sobol indices partition the total variance of a model output into components attributable to individual parameters and combinations of parameters. First-order Sobol indices measure the direct contribution of each parameter, while higher-order and total-order indices quantify interaction effects. The indices are normalised by the total model variance and therefore range from 0 to 1.
In practice, Sobol indices are estimated using Monte Carlo or quasi-Monte Carlo simulation, often employing Sobol sequences to improve sampling efficiency. They are applied to decision trees, Markov models, microsimulation and other health economic models to rank uncertain parameters, assess structural complexity and guide efficient evidence generation.
Purpose
Used to quantify the contribution of individual parameters and parameter interactions to overall model uncertainty, supporting global sensitivity analysis and prioritisation of influential model inputs.
Mathematical Formulae
Primary Formula
First-order Sobol index:
S? = Var(E(Y|X?)) / Var(Y)
Supporting Formulae
Total-order Sobol index:
ST? = E(Var(Y|X??)) / Var(Y)
Second-order Sobol index:
S?? = V?? / Var(Y)
Variance decomposition:
Var(Y) = ?V? + ?V?? + ?V??? + ?
Related Mathematical Methods
- Global Sensitivity Analysis
- First-Order Index
- Total-Order Index
- Morris Method
- Variance Decomposition
- Monte Carlo Simulation
- Sobol Sequence
Example
A probabilistic sensitivity analysis evaluates uncertainty in treatment efficacy, costs, utility values and disease progression. Sobol analysis estimates a first-order index of 0.42 and a total-order index of 0.57 for treatment efficacy. The difference indicates that treatment efficacy contributes substantially to model uncertainty both directly and through interactions with other uncertain parameters.
Excel Implementation
| Function | Example Formula | Health Economics Application |
|---|---|---|
| VAR.S | =VAR.S(ResultRange) | Estimate the total variance of simulated model outputs. |
| AVERAGE | =AVERAGE(ResultRange) | Calculate conditional means required for Sobol index estimation. |
| SUMPRODUCT | =SUMPRODUCT(Array1,Array2) | Support variance decomposition calculations. |
| RANK.EQ | =RANK.EQ(IndexValue,IndexRange,0) | Rank parameters according to their Sobol sensitivity indices. |
VBA (Optional)
VBA can automate Monte Carlo sampling, estimate first-order and total-order Sobol indices and generate ranked sensitivity reports for health economic models.
Sources
- Sobol IM. Global sensitivity indices for nonlinear mathematical models and their Monte Carlo estimates. Mathematics and Computers in Simulation. 2001;55(1?3):271?280.
- Saltelli A, Ratto M, Andres T, et al. Global Sensitivity Analysis: The Primer. John Wiley & Sons.
- Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation. Oxford University Press.
- NICE. NICE Health Technology Evaluations: The Manual.
Related Concepts (3)
Library
Publications
1
Statistical Analysis of Cost-Effectiveness Data — Willan & Briggs, 1st Edition ed., 2006 (John Wiley & Sons)
A synthesis of statistical methods for analysing cost-effectiveness data, including net-benefit regression, confidence intervals for the ICER, cost-effectiveness acceptability curves, and covariate adjustment. Part of the Wiley Statistics in Practice series.
BookView source →
Frequently Asked Questions (6)
What are Sobol indices?
Variance-based global sensitivity measures decomposing a model's total output variance into portions attributable to each input parameter and their interactions.
Source: Sobol 2001
How do Sobol indices apportion a model's output variance?
Sobol indices divide the total variance in a model's output among its inputs, attributing to each the share of the variation it is responsible for. A first-order index gives the variance an input causes acting alone, while a total-effect index adds the variance it causes in combination with others, so together they show both direct and interactive influence. Because the shares are defined to sum in a principled way, they give a complete, quantitative picture of what drives the output. They are a full variance decomposition. Saltelli and colleagues (2008) describe them.
Source: Saltelli et al. 2008
How are Sobol indices calculated?
Sobol indices are calculated by decomposing the output variance into contributions from each input and their combinations, estimated by Monte Carlo procedures using specially structured samples that vary subsets of inputs while averaging over the rest. The first-order index is the variance of the conditional expectation given an input, divided by the total variance, and the total-effect index captures that input's contribution including all interactions. These estimates, each between zero and one, quantify the shares of output variance, though obtaining them accurately can require many model runs.
Source: Saltelli et al. 2008
What is the difference between first-order and total-effect Sobol indices?
The first-order Sobol index measures the output variance explained by an input varying alone, excluding interactions, while the total-effect index measures the input's full contribution, including all its interactions with other inputs. The total-effect index is therefore at least as large as the first-order index, and a gap between them indicates interaction effects. First-order indices suit factor prioritisation, identifying individual influence, while total-effect indices suit factor fixing, since an input can be fixed only if its total effect, not just its first-order effect, is negligible.
Source: Saltelli et al. 2008
Why are Sobol indices used?
Sobol indices are used because they provide a rigorous, model-independent quantification of how much each input and interaction contributes to output uncertainty, valid even for non-linear models with interactions, unlike simpler local measures. They support ranking inputs for prioritisation, identifying negligible inputs for fixing, and revealing interactions through the gap between first-order and total-effect indices. This makes Sobol indices a thorough tool in global sensitivity analysis where a precise decomposition of output variance among the inputs is wanted to understand and communicate the drivers of uncertainty.
Source: Sobol 2001
What are the limitations of Sobol indices?
Sobol indices can be computationally expensive to estimate accurately, since the Monte Carlo procedures require many model runs, which is demanding for complex or slow models, though screening methods can reduce inputs beforehand. They depend on the assumed input distributions, so the decomposition changes if these change, and they assume the inputs can be treated appropriately, with correlated inputs complicating interpretation. These limitations mean Sobol indices are applied with attention to computational cost and input assumptions, often after screening, where a precise variance decomposition justifies the effort.
Source: Saltelli et al. 2008
Trust Record
Verified by Dr Darrin Baines
British health economist
Professional identity: darrinbaines.org
Verification date: 29 Oct 2025
Content version: 1.0.0
Canonical Identity
- Persistent URI
- https://healtheconomics.wiki/concept/sobol-indices
- Term code
- HE-EM-UA-084
Stable URI · Machine-readable · Resolvable · CC BY 4.0