Regression Graphics

‘Regression Graphics’
A Linear Model Based Statistical Analysis Program

‘REG/GRAPH’ or ‘REGRESSION GRAPHICS’ is a parametric statistical analysis program written for the New York State Education Department in the STSC APL 6.3 programming environment for IBM PC and 100% compatible systems by Thomas L. Hick.

REG/GRAPH is interactive and uses linear models as a generalized data analysis concept. (Ward & Jennings 1973).

It provides a data transformation process which converts desired variables into vector form and provides a simple way for the user to generate linear mathematical models using these variables.

REG/GRAPH provides for the testing of hypotheses by comparing changes in proportions of variance accounted for by full and restricted models of the data. F values in REG/GRAPH are derived from the National Bureau of Standards’ ‘Graphs and Tables of the Significance of F for the Fisher-Snedecor Variance Ratio’ (Graybill 1961). Associated P values are interpolated from that table.

It computes means, variances, biased and unbiased standard deviations, and maximal and minimal values for variables defined as components of linear models. REG/GRAPH also provides frequency distributions and Pearson correlation coefficients for pairs of variables of interest.

REG/GRAPH generates equation driven graphics to provide three dimensional representations of the surfaces of linear models used in analyses.

Before running the analyses for this study a trial run of REG/GRAPH was conducted using data from Green and Carrol (1976), and the results compared with the results of the multiple regression analysis reported in that text. Green and Carrol’s results were typically rounded to the nearest thousandth or hundredth. The results of the analysis generated by REG/GRAPH (e.g. regression coefficients, F tests and their significance values) agreed precisely with Green and Carrol’s results at those levels of rounding.