You are on page 1of 2

Homework 12, Statistics 512, Spring 2005

This homework is due Thursday, April 21st at the beginning of class. 1. Hogg, McKean and Craig, 7.1.5. 2. Hogg, McKean and Craig, 7.1.7. 3. A possibly biased coin has an unknown probability of landing heads. It is thrown once, and you observe the outcome. You then have a choice whether or not to play a certain game. In the game, you predict the next toss. If you predict correctly, you are paid $2; if you predict incorrectly, you lose $1. If you dont play the game, you lose nothing and gain nothing. You thus have three possible actions -- a1 : play and predict heads; a2 : play and predict tails; a3 : dont play. There are nine decision rules shown in the following table that prescribe what action to take based on whether the first toss is heads or tails: First Toss Heads Tails d1 a1 a1 d2 d3 d4 d5 d6 d7 d8 d9 a1 a1 a2 a2 a2 a3 a3 a3 a2 a3 a1 a2 a3 a1 a2 a3

(a) What are the risk functions for the nine decision rules? (b) What is the minimax rule? (c) For a uniform [0,1] prior distribution on , what is the Bayes rule? (d) Which rules are admissible and which rules are inadmissible? 4. Suppose that X has a binomial distribution with n=2 and probability of success p . Use R to graph the risk functions for the following estimates of p using squared error loss: X p1 = 2 X +1 p2 = 3

X +1 4 Are any of these estimators dominated by the others? p3 = Hint: Here is R code for the plot in Notes 26 that you can use as a guide. The command lines(vector of x coordinates, vector of y coordinates, lty=) draws a line on an existing plot. lty controls what type of line: 1=solid, 2=dashed, 3=dotted, 4=dotdash, 5=longdash, 6=twodash. thetaseq=seq(0,1,.01); bayesrisk=1/9-1/3*(thetaseq)+1/3*(thetaseq^2); mlerisk=thetaseq*(1-thetaseq); plot(thetaseq,bayesrisk,type="l",xlab="Theta",ylab="Risk",ylim=c(0,.30)); lines(thetaseq,mlerisk,lty=2);

You might also like