STA 721: Lecture 15
Duke University
Confidence Interverals from Test Statistics
Pivotal Quantities
Confidence intervals for parameters
Prediction Intervals
Bayesian Credible Regions and Intervals
Readings:
For the regression model
what is a plausible range for
what is a plausible set of values for
what is a a plausible range of values for
what is a plausible range of values for
Look at confidence intervals, confidence regions, prediction regions and Bayesian credible regions/intervals
For a random variable
In this case we say
there is some true value of
the randomness is due to
once we observe
Recall for a level
we reject
for each test we can construct:
these sets are complements of each other (for non-randomized tests)
Suppose we have a level
for each
then
This collection of hypothesis tests can be “inverted” to construct a confidence region for θ, as follows:
define
this is the set of
then
For the linear model
suppose you are testing
if
therefore if
define the acceptance region
we have that
Now construct a confidence interval for the true value by inverting the tests:
for
For a linear function of the parameters
Related to CI for
a
this use the idea of a pivotal quantity: a function of the data and the parameters that has a known distribution that does not depend on any unknown parameters.
for prediction,
Since
therefore
Rearranging gives a
In a Bayesian setting, we have a posterior distribution for
a set
lots of sets have this property, but we usually want the most probable values of
this motivates looking at the highest posterior density (HPD) region which is a
the HPD region is the smallest region that contains
For a normal prior and normal likelihood, the posterior for
the posterior density as a function of
so a highest posterior density region has the form
For unknown
for conjugate priors,
density of
For the reference prior
(generalized) posterior distribution:
marginal posterior distribution for
the posterior density
the quantity
then HPD regions for
marginals of
difference is in the interpretation of the regions i.e posterior probability that