Concept Architecture
Concept
Theoretically, Variance Reduction is a collection of mathematical techniques designed to improve the efficiency of stochastic simulation by reducing the variance of simulation estimators without increasing the number of simulation iterations. It is founded on statistical estimation theory and exists to obtain more precise estimates of expected outcomes while maintaining unbiasedness. In health economics, variance reduction techniques are used to improve the computational efficiency of Monte Carlo simulation and probabilistic sensitivity analysis.
Mathematically, variance reduction methods modify the simulation procedure to decrease the sampling variance of an estimator while preserving its expected value. Common approaches include common random numbers, antithetic variates, control variates, importance sampling and stratified sampling. These methods reduce Monte Carlo error and accelerate convergence towards the true expected value.
In practice, variance reduction techniques are incorporated into stochastic simulation models to improve the precision of estimates of costs, quality-adjusted life-years and cost-effectiveness. They are routinely applied in probabilistic sensitivity analysis, microsimulation and discrete event simulation when computational efficiency or model complexity makes conventional Monte Carlo simulation less practical.
Purpose
Used to improve the precision and computational efficiency of stochastic simulation by reducing estimator variance without increasing the number of simulation iterations.
Mathematical Formulae
Primary Formula
There is no universally recognised canonical mathematical formula.
Supporting Formulae
Variance of the sample mean:
Var(X?) = Var(X) / n
Monte Carlo standard error:
SE = s / �n
Monte Carlo convergence:
Error ? 1 / �n
Related Mathematical Methods
- Common Random Numbers
- Antithetic Variates
- Control Variates
- Importance Sampling
- Stratified Sampling
- Latin Hypercube Sampling
- Monte Carlo Simulation
- Probabilistic Sensitivity Analysis
Example
A probabilistic sensitivity analysis initially requires 100,000 Monte Carlo simulations to estimate expected net monetary benefit with acceptable precision. By implementing Latin Hypercube Sampling and common random numbers, the same level of precision is achieved using 25,000 simulations, substantially reducing computational time while maintaining unbiased estimates.
Excel Implementation
| Function | Example Formula | Health Economics Application |
|---|---|---|
| RAND | =RAND() | Generate random values for stochastic simulation. |
| AVERAGE | =AVERAGE(ResultRange) | Estimate expected model outcomes after variance reduction. |
| STDEV.S | =STDEV.S(ResultRange) | Quantify simulation variability before and after applying variance reduction techniques. |
| COUNT | =COUNT(ResultRange) | Compare the number of simulation iterations required to achieve a target precision. |
VBA (Optional)
VBA can automate implementation of variance reduction techniques such as common random numbers or stratified sampling and compare estimator precision across alternative simulation methods.
Sources
- Glasserman P. Monte Carlo Methods in Financial Engineering. Springer.
- Robert CP, Casella G. Monte Carlo Statistical Methods. Springer.
- Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation. Oxford University Press.
- Law AM. Simulation Modeling and Analysis. McGraw-Hill.
- NICE. NICE Health Technology Evaluations: The Manual.
Related Concepts (2)
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 (6)
What is variance reduction?
Statistical techniques, including antithetic variates and control variates, used to reduce a Monte Carlo simulation's statistical variability for a given number of iterations.
Source: Law & Kelton 2000
How does variance reduction get more precision from the same number of runs?
The precision of a simulation estimate usually improves only slowly with more runs, so reaching high precision by brute force can be costly. Variance reduction techniques instead restructure how the sampling is done, using devices such as pairing opposite draws or exploiting a related known quantity, so that the same number of runs yields a tighter estimate. They achieve more precision without more computation by removing avoidable noise rather than by averaging it away. The gain is efficiency for a fixed effort. Law (2015) describes these techniques.
Source: Law 2015
What techniques are used for variance reduction?
Techniques used for variance reduction include antithetic variates, pairing each draw with a negatively correlated complement so their outputs partly cancel; control variates, using a related quantity with a known mean to correct the estimate; common random numbers, sharing random streams across compared alternatives to reduce the variance of their difference; importance sampling, oversampling important regions and reweighting; and stratified or quasi-random sampling for even coverage. Each reduces variance in a different way, and the choice depends on the structure of the simulation and the quantity of interest.
Source: Press et al. 2007
Why is variance reduction used?
Variance reduction is used because Monte Carlo estimates carry sampling error that falls only slowly with the number of runs, so reducing the variance for a given number of iterations improves precision without the cost of many more runs. This matters when each run is expensive or when precise estimates are needed, as in probabilistic analysis or value-of-information calculations. By making simulation more efficient, variance reduction techniques allow better estimates within a limited computational budget, which is their central purpose in stochastic simulation.
Source: Law & Kelton 2000
How does variance reduction improve efficiency?
Variance reduction improves efficiency by lowering the variance of the estimate for a given number of runs, so that the same precision is reached with fewer runs, or better precision with the same number. Because Monte Carlo error depends on the variance of the output and the number of samples, reducing the variance directly sharpens the estimate. The techniques achieve this by exploiting structure, such as negative correlation, known quantities, or even coverage, to remove some of the random noise, giving more information per run.
Source: Law & Kelton 2000
What are the limitations of variance reduction?
Variance reduction techniques are problem-specific: each works only under certain conditions, such as monotonic responses for antithetic variates or a well-correlated control with a known mean for control variates, and a poorly chosen technique may give little benefit or even increase variance. They add complexity and require correct implementation. Their gains vary with the model and quantity of interest. These limitations mean variance reduction is applied selectively, matching the technique to the problem and checking that it genuinely improves the efficiency of the estimate.
Source: Press et al. 2007
Trust Record
Verified by Dr Darrin Baines
British health economist
Professional identity: darrinbaines.org
Verification date: 30 Oct 2025
Content version: 1.0.0
Canonical Identity
- Persistent URI
- https://healtheconomics.wiki/concept/variance-reduction
- Term code
- HE-EM-UA-083
Stable URI · Machine-readable · Resolvable · CC BY 4.0