Home Statistical Dictionary Ratio Estimator

Ratio Estimator

A survey estimator that leverages a known population total of an auxiliary variable correlated with the outcome, estimating the population mean or total as the sample ratio of outcome to auxiliary variable scaled by the known auxiliary total.

Sampling Theory & Survey MethodsIntermediateFree

In Plain English

The ratio estimator is a way to squeeze extra precision out of a survey by using some other piece of information you already know about the population, say, you're estimating this year's total crop yield, and you happen to already know last year's total acreage planted from administrative records. If yield and acreage are correlated, you can estimate the ratio of this year's yield to acreage in your sample, and then multiply that ratio by the known population acreage total to get a more precise estimate of total yield than you'd get by just averaging your sample yields directly, since you're leaning on real, known population information rather than sample noise alone.

Definition

The ratio estimator estimates a population total $Y$ using an auxiliary variable $x$ whose population total $X$ is known, via $\hat{Y}_{R} = \frac{y ˉ}{x ˉ} X = \hat{R} X$ , where $\hat{R} = \overset{y}{ˉ} / \overset{x}{ˉ}$ is the sample ratio of the outcome to the auxiliary variable. The ratio estimator is more precise than the simple expansion estimator $N \overset{y}{ˉ}$ whenever $y$ and $x$ are strongly, positively, and roughly linearly correlated, and the regression line relating $y$ to $x$ passes approximately through the origin, since the estimator essentially assumes proportionality between the two variables; unlike the simple sample mean, the ratio estimator is only approximately, not exactly, design-unbiased (it carries a small bias that vanishes asymptotically as the sample size grows), a trade-off generally considered acceptable given the potentially large variance reduction, particularly in agricultural, economic, and business surveys where administrative auxiliary totals (prior-year figures, related economic indicators) are commonly available.

Formula

\hat{Y}_{R} = \frac{y ˉ}{x ˉ} X

Notation

\overset{y}{ˉ}, \overset{x}{ˉ}

—Sample means of the outcome and auxiliary variable

X

—Known population total of the auxiliary variable

\hat{R} = \overset{y}{ˉ} / \overset{x}{ˉ}

—Estimated ratio of outcome to auxiliary variable

Properties

The ratio estimator's precision gain over the simple expansion estimator grows with the strength of the correlation between the outcome y and the auxiliary variable x, when the correlation is weak, the ratio estimator offers little benefit and its small inherent bias becomes a comparatively larger share of its total error.
Unlike the simple sample mean, the ratio estimator is only approximately design-unbiased, not exactly unbiased, in finite samples, this bias is generally small relative to its variance reduction and shrinks toward zero as sample size grows, which is why it is considered a favorable trade-off in most applications where it is used.
The ratio estimator is a special case of the broader class of regression estimators, which generalize the idea by fitting a full linear regression of y on x (allowing a nonzero intercept) rather than assuming strict proportionality through the origin, the ratio estimator can be seen as a regression estimator with the intercept constrained to zero.

At a Glance

Requiresa known population total for an auxiliary variable correlated with the outcome

Works best wheny and x are strongly, linearly correlated through roughly the origin

Trade-offsmall asymptotic bias in exchange for potentially large variance reduction

Common inagricultural and business surveys with known prior-period auxiliary totals

Last updated September 8, 2026← Back to the Dictionary