...one of the most highly
regarded and expertly designed C++ library projects in the
world.
— Herb Sutter and Andrei
Alexandrescu, C++
Coding Standards
boost::histogram::utility::wald_interval — Wald interval or normal approximation interval.
// In header: <boost/histogram/fwd.hpp> template<typename ValueType> class wald_interval : public boost::histogram::utility::binomial_proportion_interval< ValueType > { public: // public member functions explicit wald_interval(deviation = deviation{1.0}) noexcept; interval_type operator()(value_type, value_type) const noexcept; };
The Wald interval is a symmetric interval. It is simple to compute, but has poor statistical properties and is universally rejected by statisticians. It should always be replaced by another iternal, for example, the Wilson interval.
The Wald interval can be derived easily using the plug-in estimate of the variance for the binomial distribution, which is likely a reason for its omnipresence. Without further insight into statistical theory, it is not obvious that this derivation is flawed and that better alternatives exist.
The Wald interval undercovers on average. It is unsuitable when the sample size is small or when the fraction is close to 0 or 1. e. Its limits are not naturally bounded by 0 or 1. It produces empty intervals if the number of successes or failures is zero.
For a critique of the Wald interval, see (a selection):
L.D. Brown, T.T. Cai, A. DasGupta, Statistical Science 16 (2001) 101-133. R. D. Cousins, K. E. Hymes, J. Tucker, Nucl. Instrum. Meth. A 612 (2010) 388-398.
wald_interval
public member functionsexplicit wald_interval(deviation d = deviation{1.0}) noexcept;Construct Wald interval computer.
Parameters: |
|
interval_type operator()(value_type successes, value_type failures) const noexcept;Compute interval for given number of successes and failures.
Parameters: |
|