Locally-Weighted-Regression Scatter-Plot Smoothing (LOWESS): A Graphical Exploratory Data Analysis Technique

Statisticians have long used moving average type smoothing and classical regression analysis techniques to reduce the variability in data sets and enhance the visual information presented by scatterplots. This thesis examines the effectiveness of Robust Locally Weighted Regression Scatterplot Smoothing LOWESS, a procedure that differs from other techniques because it smooths all of the points and works on unequally as well as equally spaced data. The LOWESS procedure is evaluated by comparing it to previously validated uniform and cosine weighted moving average and least squares regression programs. Interactive APL and FORTRAN programs and detailed user instructions are included for use by interested readers. Additional keywords Curve smoothing Curve fitting and APL programming language.