Documentation of XIRE - A Multivariable Tape Regression Analysis Program with Rule-of-Thumb Guides for Interpretation of Results by Nonstatisticians

摘要

XIRE is a multivariable regression analysis program which solves for correlation coefficients in an equation of the form Y = b(0) + b(1)x(2) + ...b(59)x(59) where Y, x(1),x(2), ...x(59) are terms consisting of variables which may be transformed in a variety of ways. Each x(i) may be a complex term built up by a combination of several variables and constants which may be transformed and/or raised to an exponential power. The equations may be made to fit theoretical or known relationships, be used to provide an empirical fit giving useful parametric equations expression the relative effect of several independent variables on some property (dependent variable), and be used to provide a method of predicting the most likely value of some property or structure given a set of known conditions. Using tape storage to accept an unlimited number of data points with up to 60 different input variables, the program will search for aberrant data values, discard them, and then complete the correlation. Double precision routines are used for matrix calculations in order to minimize the possibility of creating an apparently singular matrix. Output consists of the variables analyzed by name, their regression coefficients, and a value of each regression coefficient with a t distribution table to evaluate the confidence interval for a given regression coefficient. The program calculates the mean and standard deviation of the regression. It also back-calculates from the original data and compares the calculated value of the dependent variable to its measured value and gives confidence intervals on each point.