Statistics Glossary → Ratio estimator (Y/X)
Ratio estimator (Y/X)concept
An estimator for survey data that estimates the ratio of two totals, such as expenditure per person or cars per household, as the ratio of weighted totals.
The ratio is R = Σ wᵢyᵢ / Σ wᵢxᵢ, where w are the sampling weights. Because the denominator is also a random quantity estimated from the sample, the variance of the ratio cannot simply be derived from the variances of two means. It is therefore usually computed by Taylor linearization: the ratio is treated as a linear combination of the deviations in numerator and denominator (zᵢ = yᵢ − R·xᵢ), and the design-based variance of that combination is found.
The ratio estimator is slightly biased in small samples, but the bias vanishes quickly as the sample grows. If y and x are strongly and proportionally related, the ratio is estimated more precisely than the mean of y. This property is also used to estimate a population total from a known auxiliary total (ratio estimation of a total). In small subgroups where the denominator can be close to zero, the estimate becomes unstable.
An everyday example
A household budget survey aims to find monthly food expenditure per person. Because households differ in size, per-household averages are not used; instead the weighted total food expenditure is divided by the weighted total number of people. If total expenditure is 8.4 million and the total number of people is 3,000, the ratio is 2,800 per person, with its standard error computed by the Taylor method.
In MerQur
In the weighted mean, the 'Ratio estimate (Y / X)' option gives the ratio as total(wy) / total(wx) with a Taylor standard error.