反正講開,不如講埋佢。
之前 #2 講到有幾企硬,好似好兒戲。但其實用甚麼的 Prior ,只要有再多的證據,用甚麼 Prior 影響都不會大。
Bayesian analysis 一個好處,是有新的數據,可以即時修正現有的結果。假設阿 Paul 還有命估 Euro 2012 ,今雲佢估六中三。我們可以用他世界盃後的數據作 Prior ,再加入這個估六中三的結果。理論上是會求出更貼近真實的 p 。

# chew that < -
p <- seq(0.00,1,by=0.01)
prior <- dbeta(p,alpha(4/6,0.03),beta(4/6,0.03))
post.d <- dbeta(p,alpha(4/6,0.03)+8,beta(4/6,0.03)+0)
post.euro12.d <- dbeta(p,alpha(4/6,0.03)+8+3,beta(4/6,0.03)+0+3)
plot(p,prior, type="l", ylab="Density", lty=1, ylim=c(0,5))
lines(p,post.d, lty=2)
lines(p,post.euro12.d, lty=3)
legend(0,5,c("Beta Prior","Post World Cup","Post Euro 2012"),lty=c(1,2,3))