Normal-inverse-gamma distribution

In probability theory and statistics, the normal-inverse-gamma distribution (or Gaussian-inverse-gamma distribution) is a four-parameter family of multivariate continuous probability distributions. It is the conjugate prior of a normal distribution with unknown mean and variance.

normal-inverse-gamma
Probability density function
Probability density function of normal-inverse-gamma distribution for α = 1.0, 2.0 and 4.0, plotted in shifted and scaled coordinates.
Parameters location (real)
(real)
(real)
(real)
Support
PDF
Mean


, for
Mode


Variance

, for
, for

, for

Definition edit

Suppose

 

has a normal distribution with mean   and variance  , where

 

has an inverse-gamma distribution. Then   has a normal-inverse-gamma distribution, denoted as

 

(  is also used instead of  )

The normal-inverse-Wishart distribution is a generalization of the normal-inverse-gamma distribution that is defined over multivariate random variables.

Characterization edit

Probability density function edit

 

For the multivariate form where   is a   random vector,

 

where   is the determinant of the   matrix  . Note how this last equation reduces to the first form if   so that   are scalars.

Alternative parameterization edit

It is also possible to let   in which case the pdf becomes

 

In the multivariate form, the corresponding change would be to regard the covariance matrix   instead of its inverse   as a parameter.

Cumulative distribution function edit

 

Properties edit

Marginal distributions edit

Given   as above,   by itself follows an inverse gamma distribution:

 

while   follows a t distribution with   degrees of freedom.[1]

Proof for  

For   probability density function is

 

Marginal distribution over   is

 

Except for normalization factor, expression under the integral coincides with Inverse-gamma distribution

 

with  ,  ,  .

Since  , and

 

Substituting this expression and factoring dependence on  ,

 

Shape of generalized Student's t-distribution is

 .

Marginal distribution   follows t-distribution with   degrees of freedom

 .

In the multivariate case, the marginal distribution of   is a multivariate t distribution:

 

Summation edit

Scaling edit

Suppose

 

Then for  ,

 

Proof: To prove this let   and fix  . Defining  , observe that the PDF of the random variable   evaluated at   is given by   times the PDF of a   random variable evaluated at  . Hence the PDF of   evaluated at   is given by : 

The right hand expression is the PDF for a   random variable evaluated at  , which completes the proof.

Exponential family edit

Normal-inverse-gamma distributions form an exponential family with natural parameters  ,  ,  , and   and sufficient statistics  ,  ,  , and  .

Information entropy edit

Kullback–Leibler divergence edit

Measures difference between two distributions.

Maximum likelihood estimation edit

Posterior distribution of the parameters edit

See the articles on normal-gamma distribution and conjugate prior.

Interpretation of the parameters edit

See the articles on normal-gamma distribution and conjugate prior.

Generating normal-inverse-gamma random variates edit

Generation of random variates is straightforward:

  1. Sample   from an inverse gamma distribution with parameters   and  
  2. Sample   from a normal distribution with mean   and variance  

Related distributions edit

  • The normal-gamma distribution is the same distribution parameterized by precision rather than variance
  • A generalization of this distribution which allows for a multivariate mean and a completely unknown positive-definite covariance matrix   (whereas in the multivariate inverse-gamma distribution the covariance matrix is regarded as known up to the scale factor  ) is the normal-inverse-Wishart distribution

See also edit

References edit

  1. ^ Ramírez-Hassan, Andrés. 4.2 Conjugate prior to exponential family | Introduction to Bayesian Econometrics.
  • Denison, David G. T.; Holmes, Christopher C.; Mallick, Bani K.; Smith, Adrian F. M. (2002) Bayesian Methods for Nonlinear Classification and Regression, Wiley. ISBN 0471490369
  • Koch, Karl-Rudolf (2007) Introduction to Bayesian Statistics (2nd Edition), Springer. ISBN 354072723X