Concept Architecture
Concept
Theoretically, Variance-Covariance Matrix is a symmetric matrix that summarises the variances of multiple random variables together with the covariances between every pair of variables. It provides a complete second-order description of joint variability and dependence among model parameters. In health economics, variance-covariance matrices are used to preserve parameter correlations during probabilistic sensitivity analysis, multivariate statistical estimation and Bayesian modelling.
Mathematically, the diagonal elements of the matrix contain the variances of individual parameters, while the off-diagonal elements contain pairwise covariances. Because covariance measures joint variation, the variance-covariance matrix captures both the magnitude of uncertainty and the dependence structure among parameters. The matrix is symmetric and positive semi-definite, properties that permit multivariate probability distributions to be constructed for correlated model parameters.
In practice, variance-covariance matrices are estimated from regression models, maximum likelihood estimation, Bayesian posterior samples or multivariate meta-analysis. Health economic models use these matrices to generate correlated parameter samples through multivariate normal simulation or Cholesky decomposition, ensuring that uncertainty analyses preserve observed relationships between costs, treatment effects, utilities and other model inputs.
Purpose
Used to quantify the joint uncertainty and correlation among multiple parameters, enabling correlated parameter sampling and multivariate statistical estimation in health economic models.
Mathematical Formulae
Primary Formula
For random vector
X = (X?, X?, ?, X?)?
the variance-covariance matrix is
� = E[(X ? ?)(X ? ?)?]
where:
- � = variance-covariance matrix
- ? = vector of parameter means
Supporting Formulae
Matrix form:
� =
? Var(X?) Cov(X?,X?) ? Cov(X?,X?) ?
? Cov(X?,X?) Var(X?) ? Cov(X?,X?) ?
? ? ? ? ? ?
? Cov(X?,X?) Cov(X?,X?) ? Var(X?) ?
Covariance:
Cov(X,Y) = E[(X ? ??)(Y ? ??)]
Relationship with correlation:
Cov(X,Y) = ???�?�?
Related Mathematical Methods
- Multivariate normal distribution
- Cholesky decomposition
- Regression analysis
- Maximum likelihood estimation
- Bayesian estimation
- Markov Chain Monte Carlo
- Probabilistic sensitivity analysis
- Multivariate meta-analysis
Example
A cost-effectiveness model simultaneously estimates treatment cost and health utility.
Estimated parameter statistics are:
- Variance of cost = 250,000
- Variance of utility = 0.010
- Covariance = 12.5
The variance-covariance matrix is
� =
?250000 12.5 ?
? 12.5 0.010?
This matrix is used with Cholesky decomposition to generate correlated random samples during probabilistic sensitivity analysis, preserving the observed association between treatment cost and health utility.
Excel Implementation
| Function | Example Formula | Health Economics Application |
|---|---|---|
| COVARIANCE.S | =COVARIANCE.S(B2:B201,C2:C201) | Estimate covariance between two model parameters |
| VAR.S | =VAR.S(B2:B201) | Estimate parameter variance |
| CORREL | =CORREL(B2:B201,C2:C201) | Estimate parameter correlation |
| MMULT | =MMULT(A2:B3,C2:D3) | Perform matrix multiplication during multivariate calculations |
| TRANSPOSE | =TRANSPOSE(A2:B3) | Construct or manipulate symmetric covariance matrices |
VBA (Optional)
Automate construction of variance-covariance matrices and generation of correlated parameter samples for probabilistic sensitivity analysis.
Sources
- Briggs AH, Claxton K, Sculpher MJ. Decision Modelling for Health Economic Evaluation. Oxford University Press; 2006.
- Gelman A, Carlin JB, Stern HS, Dunson DB, Vehtari A, Rubin DB. Bayesian Data Analysis. 3rd ed. CRC Press; 2013.
- Johnson RA, Wichern DW. Applied Multivariate Statistical Analysis. 6th ed. Pearson; 2007.
- Anderson TW. An Introduction to Multivariate Statistical Analysis. 3rd ed. Wiley; 2003.
- ISPOR-SMDM Modeling Good Research Practices Task Force Reports.
Related Concepts (4)
Library
Publications
1
Parameter Estimation and Uncertainty: A Report of the ISPOR-SMDM Modeling Good Research Practices Task Force-6 — Briggs, Weinstein, Fenwick, Karnon, Sculpher & Paltiel, Task Force Report 6 ed., 2012 (Value in Health / Medical Decision Making)
Best-practice guidance on parameter estimation and the characterisation of uncertainty in decision models, covering probabilistic sensitivity analysis, distributional choices, and correlation between parameters.
Journal ArticleView source →
Frequently Asked Questions (7)
What is a variance covariance matrix?
A table showing each parameter's variance and the covariance between each pair, summarising both individual uncertainty and interdependence among model inputs.
Source: Pearson 1896
What statistical array is a variance-covariance matrix?
A variance-covariance matrix is a square array that records the variances of a set of parameters on its diagonal and the covariances between each pair off the diagonal. It summarises both how uncertain each parameter is and how the parameters move together. In economic modelling it is often taken from a regression and used to generate correlated parameter draws in probabilistic sensitivity analysis. Preserving the off-diagonal covariances matters, because ignoring them would treat related parameters as independent and misstate overall uncertainty. It is a compact description of joint parameter uncertainty.
Source: Briggs, Claxton and Sculpher 2006
What is a variance-covariance matrix?
A variance-covariance matrix is a table showing each parameter's variance on the diagonal and the covariance between each pair of parameters off the diagonal, summarising both the individual uncertainty in each parameter and the interdependence among them. Symmetric by construction, it captures how much each parameter varies and how parameters vary together. In modelling, it represents the joint uncertainty of a set of parameters, which is needed to sample them correctly when they are correlated.
Source: Pearson 1896
What does a variance-covariance matrix show?
A variance-covariance matrix shows, for a set of parameters, the variance of each, indicating its individual uncertainty, and the covariance between each pair, indicating how they move together. A positive covariance means the parameters tend to rise and fall together, a negative one that they move oppositely, and zero that they are uncorrelated. By combining variances and covariances, the matrix fully describes the joint spread of the parameters under a multivariate normal assumption, summarising their uncertainty and interdependence in one structure.
Source: Pearson 1896
How does a variance-covariance matrix differ from a correlation matrix?
A variance-covariance matrix contains the variances of the parameters and the covariances between them, so it carries information about both the scale of each parameter's uncertainty and their joint variation, whereas a correlation matrix standardises these to correlation coefficients between minus one and one, showing only the strength and direction of association, not the magnitudes. The correlation matrix can be derived from the variance-covariance matrix by scaling. So the variance-covariance matrix retains the scales, while the correlation matrix shows relationships independent of scale.
Source: Pearson 1896
How is a variance-covariance matrix used in probabilistic analysis?
In probabilistic sensitivity analysis, a variance-covariance matrix is used to sample correlated parameters jointly, so that their variances and covariances are preserved across iterations rather than sampling each independently. Where parameters are estimated together, for instance from a regression, the estimated variance-covariance matrix captures their joint uncertainty, and sampling from a multivariate distribution using it generates draws with the correct correlations. This ensures the propagated uncertainty reflects both the individual uncertainties and the dependencies among the parameters.
Source: Briggs, Claxton & Sculpher 2006
Why is capturing covariance between parameters important?
Capturing covariance between parameters is important because parameters are often not independent, and sampling them independently in probabilistic analysis would misrepresent their joint uncertainty, potentially giving impossible or misleading combinations and biasing the estimated uncertainty in the results. The covariances ensure that parameters that move together are sampled together. Where parameters are estimated jointly, ignoring their covariance loses information about their relationships, so using the variance-covariance matrix preserves the correct joint uncertainty, which matters for a valid representation of uncertainty in the outputs.
Source: Briggs, Claxton & Sculpher 2006
Trust Record
Verified by Dr Darrin Baines
British health economist
Professional identity: darrinbaines.org
Verification date: 13 Oct 2025
Content version: 1.0.0
Canonical Identity
- Term code
- HE-EM-MP-046
Stable URI · Machine-readable · Resolvable · CC BY 4.0