Basu's theorem

Lore & Background

Debabrata Basu published Basu's theorem in 1955. The theorem concerns a family of distributions on a measurable space, a statistic T that maps to another measurable space, and an ancillary statistic A. If T is boundedly complete and sufficient for a parameter θ, and A is ancillary to θ, then conditional on θ, T is independent of A.

Reader's Guide

Basu's theorem is a foundational result in statistical theory, providing a method to prove the independence of two statistics. Its proof relies on the properties of boundedly complete sufficient statistics and ancillary statistics. An example given in the source article is the independence of the sample mean and sample variance for a normal distribution, which is shown by first demonstrating the sample mean is complete sufficient and the sample variance is ancillary, then applying the theorem. The article notes that this property (independence of sample mean and sample variance) characterizes normal distributions. The theorem remains a standard tool in mathematical statistics.

Did You Know?

• Basu's theorem was published in 1955 by Debabrata Basu.
• An example of its use is proving the independence of sample mean and sample variance for a normal distribution.

The Core Statement and Its Conditions

Basu's theorem, established in 1955 by Debabrata Basu, delivers a clean independence result in mathematical statistics. The setting involves a family of probability distributions indexed by a parameter θ, defined on a measurable space, together with a statistic T that maps observations into another measurable space. The theorem requires two conditions on T: it must be sufficient for the parameter (meaning the conditional distribution of the data given T carries no further information about θ) and it must be boundedly complete (a technical condition ensuring that the only bounded function of T with zero expectation under every parameter value is the zero function). The second ingredient is an ancillary statistic A, whose distribution is entirely free of θ. Under these two requirements, the theorem concludes that T and A are conditionally independent given θ. This is a remarkably concise bridge between two seemingly unrelated properties—completeness and sufficiency on one side, parameter-freeness on the other—yielding a structural independence statement.

The Architecture of the Proof

The proof of Basu's theorem is elegant in its economy. One begins by writing the marginal probability of any measurable set B under the ancillary statistic A as an integral over the range of T, decomposing it through the law of total probability. Two critical observations then drive the argument forward. First, because A is ancillary, its marginal distribution is identical for every value of θ. Second, because T is sufficient, the conditional distribution of the data given T equals t is likewise free of θ. Subtracting the marginal probability from the conditional probability inside the integrand produces a function g(t) that depends only on t, not on θ. The integral of g(t) against the distribution of T equals zero for every θ. Bounded completeness of T now forces g(t) to vanish almost everywhere, which is precisely the statement that the conditional distribution of A given T equals the marginal distribution of A—in other words, independence.

A Workhorse for Proving Independence

In practice, Basu's theorem serves as a shortcut for establishing independence between two statistics without resorting to joint density calculations. The strategy is straightforward: identify one statistic as boundedly complete and sufficient for the parameter of interest, verify that the other statistic is ancillary, and then invoke the theorem to conclude independence. The canonical illustration involves the normal distribution, where the sample mean is complete sufficient for the location parameter while the sample variance is ancillary with respect to that same parameter. The theorem immediately yields their mutual independence. This result is not merely a curiosity; the independence of the sample mean and sample variance is in fact a characterizing property of the normal family. No other distribution in the broader location-scale family enjoys this separation, making Basu's theorem a key ingredient in one of the most celebrated characterizations in mathematical statistics.

Historical Placement and Lasting Influence

Debabrata Basu published this result in 1955, and it has since become a standard reference in theoretical statistics. The theorem occupies a distinctive niche: it does not introduce new distributional assumptions or require parametric forms beyond the general measurable-space framework. Instead, it extracts a structural consequence from two abstract properties—bounded completeness with sufficiency, and ancillarity—that are defined purely in terms of how distributions behave under reparameterization. Because the statement is so general, it applies across a wide range of models, from exponential families to more exotic constructions, wherever the two conditions can be verified. Its proof, relying on a single integral identity and the completeness condition, is short enough to appear in introductory graduate texts yet deep enough to connect sufficiency, completeness, and ancillarity in a way that reshapes how statisticians think about the geometry of statistical inference.