SS459/559b Winter 2000

Assignment 3

Due date:  March 2, 2000, in lab.

 

Note: You have to email your Splus codes (script codes) to hyu@stats.uwo.ca besides the usual hardcopy of your assignments.

 

1.      Implement qqplot and pplot functions for one of distributions: Gamma or Weibull, by following steps. The  principle of qqplot was discussed in class (a scatter plot of Qn(F(x)) against x=Q(p)). Do some error checking on the data x in your codes.

a)      Write a qqplot function with three arguments:

1)      x                a vector (data)

2)      shape        shape parameter

3)      rate (scale)       rate parameter for Gamma (scale parameter for Weibull)

Then the qqplot will do a scale plot of the empirical quantile function against the theoretical quantile function. Compare you qqplot with the trellis function qqmath by a example. 

b)      Write a ppplot function with three arguments:

1)      x                a vector (data)

2)      shape        shape parameter

3)      rate (scale)       rate parameter for Gamma (scale parameter for Weibull)

Then the ppplot will do a step function plot of Fn(Q(p)) against p, where Fn is the empirical distribution function and Q is the theoretical quantile function.  Since all points are in (0,0) to (1,1) box, a line from (0,0) to (1,1) should be in the plot which can be used to make compare with the empirical step function. Show your code works by a example.

c)      Write a function to estimate the two parameters based on methods of moments.

d)      Using the results of c) as initial values, write a function to do MLE estimation of the two parameters. The function should only have one argument x and returns two estimators.

e)      Rewrite a) and b) with one argument x only. The parameter values are taken from d).

f)       Run some simulation. Say, simulate two sets of data from Gamma and Weibull (n=100), then apply your qqplot and qqplot. Comments on similarity and difference of two plots.

 

2.       Write a function to run a simulation to compute the relative efficiency of median to mean under contaminated normal distribution.

a)      Use (create) a function to simulate contaminated normal distribution discussed in class.

b)      Then create a function to loop a) m times to simulate m sets of data. Use these datasets to compute mean and median and then estimate variances for median and mean. After that a value of RE of median and mean can be returned from a function. (you can combine a) and b) together in order to avoid looping if you want to try).

c)      Write a function to redo b) l times and return mean and standard deviation of RE values and a histogram of RE values.

d)      Run a set of simulation with n=100, m=100, l=10, mu=0, sig=1, k = 3 for eps=0%, 1%, 5%, 10%. Report your finding.