Concept Architecture
Concept
Theoretically, Unbiasedness is a statistical property of an estimator whereby its expected value equals the true value of the parameter being estimated. An unbiased estimator neither systematically overestimates nor underestimates the population parameter across repeated random samples. The concept is fundamental to statistical estimation theory and forms one of the principal criteria for evaluating the performance of estimators alongside consistency, efficiency and sufficiency.
Mathematically, unbiasedness is defined in terms of the expectation of an estimator. If the expected value of an estimator equals the true parameter, the estimator is unbiased and has zero bias. The concept is derived from probability theory and mathematical expectation and applies to both point estimators and, more generally, estimation procedures.
In practice, unbiasedness is evaluated theoretically through analytical derivation or empirically through simulation studies. Many commonly used estimators, such as the sample mean for the population mean, are unbiased under standard assumptions, whereas others may exhibit finite-sample bias. In health economics, unbiased estimation is essential when estimating treatment effects, healthcare costs, utility values and model parameters to ensure that statistical inference and economic evaluations are not systematically distorted.
Purpose
Used to assess whether an estimator provides, on average across repeated sampling, an accurate estimate of the true population parameter without systematic overestimation or underestimation.
Mathematical Formulae
Primary Formula
E(??) = ?
Supporting Formulae
Bias(??) = E(??) ? ?
An estimator is unbiased if:
Bias(??) = 0
Related Mathematical Methods
- Point Estimation
- Mathematical Expectation
- Maximum Likelihood Estimation
- Method of Moments
- Consistency
- Efficiency
- Mean Squared Error
Example
A health economist estimates the mean annual healthcare cost of a patient population using the sample mean from repeated random samples. Suppose the true population mean cost is �8,750. Across many repeated samples, the average of the estimated sample means is also �8,750. Therefore,
E(x?) = ? = �8,750
The sample mean is an unbiased estimator of the population mean because its expected value equals the true parameter.
Excel Implementation
| Function | Example Formula | Health Economics Application |
|---|---|---|
| AVERAGE | =AVERAGE(B2:B201) | Estimate an unbiased sample mean for healthcare costs or QALYs. |
| RANDARRAY | =RANDARRAY(100,1) | Generate repeated random samples for simulation studies assessing estimator bias. |
| AVERAGE | =AVERAGE(H2:H1001) | Calculate the average of repeated parameter estimates to assess unbiasedness. |
| ABS | =ABS(AVERAGE(H2:H1001)-J2) | Calculate the absolute empirical bias relative to the known parameter value. |
VBA (Optional)
Automate repeated Monte Carlo simulations to evaluate the empirical bias of alternative estimators under varying sample sizes and data-generating processes.
Sources
- Casella G, Berger RL. Statistical Inference.
- Lehmann EL, Casella G. Theory of Point Estimation.
- Cox DR, Hinkley DV. Theoretical Statistics.
- Drummond MF, et al. Methods for the Economic Evaluation of Health Care Programmes.
- Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation.
Related Concepts (2)
Library
Publications
1
Bayesian Methods in Health Economics — Gianluca Baio, 1st Edition ed., 2012 (Chapman & Hall / CRC Press)
An overview of Bayesian statistical methods for the analysis of health economic data, covering economic evaluation concepts, statistical cost-effectiveness analysis, Bayesian computation and MCMC, and applied health economic evaluation.
BookView source →
Frequently Asked Questions (6)
What is unbiasedness?
A property of an estimator whose expected value, across repeated samples, equals the true underlying value of the parameter being estimated.
Source: Casella G, Berger RL. Statistical Inference. 2nd ed. Duxbury; 2002.
What property does an unbiased estimator have on average?
An unbiased estimator is one whose average value, taken over many repeated samples, lands exactly on the true quantity being estimated. On average it neither overshoots nor undershoots: any single estimate may miss, but across repetitions the misses cancel and centre on the truth rather than beside it. This freedom from systematic error is valued, though it is not everything, since an unbiased estimator that varies wildly can be worse in practice than a slightly biased but far steadier one. Hitting the truth on average is what it guarantees. Kirkwood and Sterne (2003) describe this.
Source: Kirkwood & Sterne 2003
What does it mean for an estimator to be unbiased?
For an estimator to be unbiased means that its expected value, the average of its estimates over infinitely many repeated samples, equals the true parameter, so it does not systematically err in one direction. Individual estimates still vary around the truth, but their average is correct. So an unbiased estimator is one whose long-run average hits the true value, having no systematic bias, which is a property about the estimator's behaviour over repeated sampling rather than about any single estimate, and it is one criterion for a good estimator, alongside others such as low variance and efficiency.
Source: Casella & Berger 2002
Why is unbiasedness desirable?
Unbiasedness is desirable because an unbiased estimator does not systematically over- or underestimate the parameter, so on average it gives the correct value, which supports valid inference and avoids consistent error in one direction. So unbiasedness is a valued property because it ensures the estimator is correct on average, which is reassuring for estimation, though it is not the sole criterion, since an unbiased estimator can have high variance, making individual estimates imprecise, which is why unbiasedness is considered alongside variance and efficiency, and why a biased estimator with much lower variance is sometimes preferred.
Source: Casella & Berger 2002
How does unbiasedness relate to the bias-variance tradeoff?
Unbiasedness relates to the bias-variance tradeoff as one extreme, having zero bias, but an unbiased estimator may still have high variance, and the total error, measured by the mean squared error, combines both bias and variance. So unbiasedness minimises the bias component but not necessarily the total error, which is why the bias-variance tradeoff shows that a slightly biased estimator with much lower variance can have a smaller mean squared error and thus be preferable, meaning unbiasedness is desirable but not always optimal, since reducing variance at the cost of a little bias can improve overall accuracy.
Source: Casella & Berger 2002
Is an unbiased estimator always the best choice?
An unbiased estimator is not always the best choice, because an unbiased estimator with high variance can give imprecise individual estimates, and a biased estimator with much lower variance may have a smaller total error, measured by the mean squared error. So unbiasedness is a desirable but not sufficient property, since the best estimator depends on the balance of bias and variance, which is why in some settings, such as with regularisation, a small amount of bias is accepted to reduce variance and improve overall accuracy, meaning that unbiasedness is weighed against precision rather than pursued at all costs.
Source: Casella & Berger 2002
Trust Record
Verified by Dr Darrin Baines
British health economist
Professional identity: darrinbaines.org
Verification date: 26 Dec 2025
Content version: 1.0.0
Canonical Identity
- Persistent URI
- https://healtheconomics.wiki/concept/unbiasedness
- Term code
- HE-ES-SA-226
Stable URI · Machine-readable · Resolvable · CC BY 4.0