AI Glossary / Statistics and probability / Intermediate

Ratio estimation

Definition

Ratio estimation uses a relationship between a study variable and an auxiliary variable to estimate a population quantity. In a common survey application, an estimated ratio of totals is multiplied by a known auxiliary population total to estimate the study-variable total. The method can improve precision when the variables have a suitable relationship. Because a ratio is nonlinear, the resulting estimator is generally not exactly unbiased under repeated sampling, even when its component total estimators are unbiased.

What it means

Auxiliary information supplies a benchmark that the sample alone may miss. If a sample happens to contain smaller units than the population average, an unadjusted total estimate may be low for a variable associated with size. A ratio adjustment can compensate when the relationship is appropriate. This is different from simply averaging individual ratios. The target might be a population ratio itself or a total estimated using that ratio, so the intended quantity must be stated.

How it works

Select an auxiliary variable measured consistently in the sample and population. Estimate the study total and auxiliary total using weights appropriate to the sampling design. Form their ratio and, when estimating a total, multiply by the known auxiliary total. For a simple random sample with common weights, the ratio reduces to the sample sum of study values divided by the sample sum of auxiliary values. Estimate uncertainty using a method that accounts for the ratio and design, such as suitable linearization or replication.

Three practical examples

Example 1: A business survey estimates a current total using a known size measure related to the outcome. Example 2: A field survey uses a known land-area total to support estimation of an area-related quantity. Example 3: A statistical team calibrates weights so that an auxiliary total matches a reliable population benchmark. These applications require compatible definitions and coverage. An outdated or poorly related benchmark can introduce problems rather than improve the estimate.

Think of it this way

Think of estimating the weight of a collection by first learning an average weight-per-size relationship from a sample, then applying it to the collection's known total size. The approach uses more than the sample's count of objects. It works best when the relationship is stable enough for the intended population. A collection mixing very different object types may need a richer model or separate estimates.

Worked example

Suppose a simple random sample of three units from a population of 100 has auxiliary values x = [10, 20, 30] and study values y = [25, 35, 60]. The sample sums are 60 and 120, so the ratio estimate is 120/60 = 2. If the known population auxiliary total is 2,500, the ratio estimate of the study total is 2 × 2,500 = 5,000. The unadjusted expansion estimate is 100 × (120/3) = 4,000. Its estimated auxiliary total is 100 × (60/3) = 2,000, so ratio adjustment multiplies by 2,500/2,000 = 1.25. The mean of the individual ratios is (2.5 + 1.75 + 2)/3 = 25/12, approximately 2.0833, which is not the ratio of totals. These calculations do not establish which estimate is closer to the unknown true total.

Illustrative case study

Fictional case study: Poplar Surveys estimates business activity using a sampled outcome and a known employment total. Analysts inspect the relationship and find that very small and very large establishments behave differently. They evaluate separate adjustments and design-based uncertainty rather than assuming one overall ratio is optimal. The case is fictional and reports no measured precision gain. Its purpose is to show that auxiliary information must be assessed for relevance and consistency before it is used to adjust estimates.

Common misconceptions and mistakes

Common mistakes include averaging unit-level ratios when the target is a ratio of totals, ignoring sampling weights, and assuming the ratio estimator is always unbiased. A relationship with a substantial intercept may make a simple ratio adjustment inefficient or misleading. A near-zero denominator creates instability. Errors in the known auxiliary total can propagate into the result. Matching one benchmark also does not automatically correct nonresponse or coverage bias for every study variable.

Business use

Business use: Ratio estimation can improve official statistics, operational surveys and resource estimates when reliable auxiliary totals are available. Document the target, weights, benchmark source and adjustment method. Check whether the auxiliary variable predicts the study variable and whether the relationship varies across important groups. Compare uncertainty and sensitivity with an unadjusted estimate or alternative estimator. Treat the benchmark as data requiring quality checks, not as an unquestionable correction factor.

People also ask

Is a ratio of totals equal to the average of individual ratios?

Generally no; the worked example shows the difference. Q: Does using a known benchmark guarantee accuracy? A: No; the relationship and benchmark quality still matter. Q: Is ratio estimation limited to simple random samples? A: No, but complex designs require appropriate weighted estimators and variance calculations.

Check your understanding

Why does the example's adjusted estimate exceed the unadjusted one?

Answer: The sample-based auxiliary total is below the known total, so the common ratio adjustment scales it upward.

Does that prove the original study estimate was too low?

Answer: No. The adjustment is justified by the assumed relationship, while the true study total remains unknown.

Learn next

Calculate a ratio of totals and a mean of ratios from the same dataset and identify their different targets. Simulate repeated samples from a small known population to inspect bias and variance. Compare a proportional relationship with one having a nonzero intercept. Then study how survey weights and uncertain auxiliary information change the calculation.

More in Statistics and probability

• Rare events - Rare events are outcomes with low probability or low frequency relative to a specified population, exposure or time horizon.
• Rasch model - The Rasch model is a probabilistic measurement model relating a person's position on a latent trait to the difficulty of an item.
• Rashomon effect in machine learning - The Rashomon effect in machine learning is the presence of multiple different models that predict the same data similarly well.
• Ratio scale - A ratio scale is a measurement scale with meaningful equal intervals and a meaningful zero, so ratios of values can represent ratios of the measured quantity.
• Rational subgroups - Rational subgroups are deliberately formed groups of observations used in statistical process control so that within-subgroup variation represents a...
• Realized variance - Realized variance is an ex-post measure of variation calculated by summing squared returns over subintervals within a specified period.

Sources

Sources: Statistics Canada, Concepts, definitions and data quality, ratio estimation discussion: https://www.statcan.gc.ca/en/statistical-programs/document/2101_D3_V147 ; Statistics Canada, Weighting: https://www150.statcan.gc.ca/n1/edu/power-pouvoir/ch6/5214809-eng.htm . These primary statistical-agency sources explain auxiliary information and estimation adjustments. The three-unit calculation and fictional business-survey case are original illustrations.

Previous: Rating scale Next: Ratio scale