MerQuR+Statistics Glossary

Statistics Glossary → Credible interval and highest density interval (HDI)

Credible interval and highest density interval (HDI)concept

In Bayesian analysis, an interval that contains a parameter with a given probability (for example 95%) according to its posterior distribution; the HDI is the narrowest such interval.

After seeing the data, a Bayesian analysis gives a posterior probability distribution over the values a parameter can take. A 95% credible interval is an interval containing 95% of this distribution, and it is read directly as 'given the data and the prior, the parameter lies in this interval with 95% probability'. The equal-tailed interval cuts off 2.5% from each end. The highest density interval (HDI) is the interval in which every point has a higher density than every point outside it; of all intervals with the same probability, it is the narrowest.

For symmetric posteriors the two intervals are almost identical; for skewed posteriors (such as variance or proportion parameters) the HDI shifts towards the dense part of the distribution and is more informative. A confidence interval has the frequentist interpretation 'in repeated samples, 95% of such intervals cover the true value', whereas a credible interval makes a probability statement from a single data set; the price is that the result depends on the choice of prior. With little data, an informative prior can affect the interval considerably. In some applications the HDI is compared with a pre-specified region of practical equivalence (ROPE) to reach a decision.

An everyday example

A café wonders how popular a new dessert will be. Of 40 customers, 28 say they 'loved it'. With a flat prior, the 95% HDI from the posterior distribution of the approval rate is roughly 0.56-0.83. The owner can read this directly as 'there is a 95% probability that the approval rate is between 56% and 83%'.

In MerQur

In the Bayesian t-Test, the 'Confidence Level' option sets the level of the credible interval and the highest density interval (HDI) of δ, and in Bayesian Correlation the level of the credible interval and HDI. Bayesian Linear Regression gives credible intervals for the coefficients and, as extra output, an HDI for σ²; Bayesian ANOVA reports 95% credible intervals in its descriptives.

Analyses where this concept appears

Bayesian t-Test (JZS Bayes Factor)Bayesian CorrelationBayesian ANOVABayesian Linear Regression

Related concepts

Posterior distribution and posterior predictive checkConfidence level and confidence interval