Statistics 740 - Spring 2004

Spring 2004
Statistics 740 - Statistical Computing
Monday/Wednesday/Friday 10:10-11:00
210A LeConte

accov1 and accov2 (1/12/04) - R code to calculate the average conditional covariance for 0-1 data. You can either cut and paste it into R or use
source("http://www.stat.sc.edu/~habing/courses/740/dimenrcode.txt") to load the two functions and the data set in. accov2(a1291) will run the efficient version.
accov.for (1/14/04) - Fortran code to calculate average conditional covariance for 0-1 data. Save it in the c:\MinGW\Programs directory. Choose Run under the Start menu in windows, and type in command or cmd. This will bring up a dos window. Type c: and then cd \MinGW\Programs. You can then compile the program using
g77 accov.for -o accov.exe and run it simply by typing accov.
a1291nospace.txt - This is the a1291 data set you can use to check the above code on. To use with fortran, save it in the Programs subdirectory of the MinGW directory on the C drive. The code to read it into R is in between accov1 and accov2 in that file.
chartoint.for (Hmwk 1) - This file contains a program demonstrating the chartoint subroutine. This subroutine is designed to prompt the user for an integer in a certain (positive) range, and to keep re-asking if they don't enter an appropriate value.
count.for, count2.for, and count3.for (1/21/04) - These are the three computer arithmetic example programs from in class. The first program count demonstrates that integers wrap around when the maximum or minimum possible value is met. This example is for INTEGER*1 variables. count2 and count3 demonstrate that the order floating point numbers are added can make a difference when the magnitudes differ greatly. Note that in count2 we need to add the decimal place after the 1 in (1/i) so that it doesn't do integer division. Note that if I wanted to just have the computer give a certain sum or product I could remove the loop entirely and just use notation like sum=24*15+115 and write(*,*) sum.
randu (2/4/04) - An incautious implementation of the inadequate randu random number generator for R, along with the code to help demonstrate one of its inadequacies.
newrt (2/6/04) - Generates a single random number from a t-distribution using the accept-reject method and g(x)=1/4*min(1,1/x^2).
Fleishman's Power Method (2/9/04) - Contains the code for generating pseudo-random variables with specified first four moments as per: Fleishman, A.I. (1978). A method for simulating non-normal distributions. Psychometrika, 43, 521-532.
The function sandk will calculate the first four moments for the entered data set. The function rfleish will generate the random variables. The other two functions are called by rfleish.
Try: x<-rfleish(200000,1,1.5,2,18) and sandk(x) .
h2.for (Hmwk 2) - Some code to demonstrate calling a function and writing out to a file. You can use the read.table command to read the output file into R, for example:
h2<-read.table("c:/mingw/programs/h2example.txt")
rbootmedci (2/16/04) - Contains code for performing a naive bootstrap confidence interval for the median.
bbmedci and stbmedci (2/20/04) - Contains the code for calculating the basic bootstrap confidence interval for the median and for performing the studentized bootstrap confidence interval (with the naive variance estimate).
h3.txt (Hmwk 3 Solutions) - Contains R code for counting the number of rejections in the accept reject method for generating t random variables. Also contains R code for simulating multivariate normals given the correlation matrix instead of the covariance matrix.
bcaboots (2/23/04) - Contains the code (with some extra comments) for calculating the accelerated bias-corrected bootstrap confidence interval for the median and performing a brief simulation study.
regresboot and regptsboot (2/25/04) - Contains the code for estimating confidence intervals for the slope in simple linear regression using either resampling from the residuals or resampling from the data points. The percentile method is used.
jack.txt (2/27/04) - A small piece of code to demonstrate why the jackknife distribution needs to be multiplied by sqrt(n-1).
bisect and Newton (3/5/04) - Functions to perform the univariate bisection method and Newton's method for finding roots of equations. Accompanied with the extra code needed to find the mle of the shape parameter for the gamma distribution.
Nelder-Mead Algorithm (3/24/04) - An implementation of the Nelder-Mead optimization algorithm with no safe-guards. It will plot the successive triangles for a two-dimensional problem.
R's built in optimization (3/26/04) - An example of using R's built in functions for optimization and constrained optimization (optim and constrOptim).
EM widows (4/2/04) - Examining Thisted's data concerning the number of children widows have by finding both the MLE for the observed data distribution and by using the EM algorithm.
Metropolis-Hastings Example (4/23/04) - Using Metropolis-Hastings to examine the posterior of a Cauchy location parameter using a normal prior.
Gibbs Sampler Example (4/28/04) - Using a Gibbs sampler to simulate data from a bivariate normal distribution with mean vector (0,0), standard deviation vector (1,1), and correlation 0.9.
Integration Code for Final (Final) - Numerical integration to find P(X1 < a,X2 < b) for a bivariate normal distribution with mean vector (0,0), standard deviation vector (1,1), and correlation rho. intfun uses a two dimensional version of the midpoint method with a sqnp by sqnp grid of points. intfun2 uses the adapt library to estimate the same value.