WeiYa's Work Yard

A traveler with endless curiosity, who fell into the ocean of statistics, tries to write down his ideas and notes to save himself.

One Parameter Models

Posted on 0 Comments
Tags: Bayesian Inference

What is one-parameter model

A class of sampling distributions that is indexed by a single unknown parameter.

The binomial model

Happiness data

$n = 129$ individuals

If happiness, $Y_i = 1$, otherwise, $Y_i=0$.

A uniform prior distribution

\[p(\theta\vert y_1,\ldots, y_{129})\propto p(y_1,\ldots,y_{129}\vert\theta)\]

Data and posterior distribution

\[p(y_1,\ldots,y_{129}\vert\theta)=\theta^{118}(1-\theta)^{11}\]

An important formula:

\[\int_0^1\theta^{a-1}(1-\theta)^{b-1}d\theta=\frac{\Gamma(a)\Gamma(b)}{\Gamma(a+b)}\]

The beta distribution

\[p(\theta)=dbeta(\theta,a,b)=\frac{\Gamma(a+b)}{\Gamma(a)\Gamma(b)}\theta^{a-1}(1-\theta)^{b-1}\qquad \text{for }0\le \theta\le 1\]

Three properties:

  1. $mode(\theta)=\frac{a-1}{a-1+b-1}$
  2. $E(\theta)=\frac{a}{a+b}$
  3. $Var(\theta)=\frac{E(\theta)E(1-\theta)}{a+b+1}$

Confidence regions

Frequentist coverage lacks a post-experimental interpretation.

relationship

an interval that has 95% Bayesian coverage additionally has the property that

\[Pr(l(Y) < \theta < u(Y)\vert \theta) = .95+\epsilon_n\]

where $\vert \epsilon_n\vert<a/n$

quantile-based interval

$n=10$ conditionally independent draws of a binary random variable.

using a uniform prior distribution for $\theta$

a = 1; b = 1 ## prior
n = 10; y =2 ## data
qbeta(c(.025, .975), a+y, b+n-y)
## 0.06021773 0.51775585

Hightest posterior density region

There is a simple R code which can return the HPD region.

The Poisson Model

\[Pr(Y=y\mid \theta)=dpois(y,\theta)=\theta^ye^{-\theta}/y!\quad \text{for }y\in \{0,1,2,\ldots,\}\]

Two properties

  • $E(Y\mid\theta)=\theta$
  • $Var(Y\mid \theta)=\theta$

Gamma distribution

gamma(a,b)

\[p(\theta)=dgamma(\theta,a,b)=\frac{b^a}{\Gamma(a)}\theta^{a-1}e^{-b\theta}\]
  • $E(\theta) = a/b$
  • $Var(\theta) = a/b^2$
  • $mode(\theta)=(a-1)/b (if\;a > 1\;or\; 0)$

posterior distribution of $\theta$

Exponential families and conjugate priors

A one-parameter exponential family model is any model whose densities can be expressed as

\[p(y\mid\phi)=h(y)c(\phi)e^{\phi t(y)}\]

conjugate prior distribution

\[p(\phi\mid n_0,t_0)=\kappa (n_0,t_0)c(\phi)^{n_0}e^{n_0t_0\phi}\]

Published in categories Note