Session: Undergraduate Research Posters: Analytical Chemistry:

Least squares estimation is widely used in many chemistry applications when analyzing data and fitting linear models. Many times a straight line fit through a set of data points is fit by minimizing the sum of squares error i.e., least squares line, with examples ranging from atmospheric chemistry, organic chemistry, chemical kinetics as well as analytical chemistry where the authors modeled a dependence of signal strength against the concentration of a substance in the form of a calibration curve. Others mapped globally averaged methane dry-air mole fractions over time; such fractions as those direct atmospheric measurements from 1978 to 2011. In this poster simulations are used to compare linear models fit by ordinary least squares (OLS), reverse OLS, maximum likelihood estimation (MLE), orthogonal regression, geometric mean regression, robust regression and also from using the Bootstrap to find an improved estimate by incorporating bias in estimation. Comparisons are made using simulated data, with the introduction of data that has increased variability on the error terms and with the introduction of known influential/outlier observations. Two data sets were used. One in which estimates were such that models could not be distinguished from each other; and one where models could be separated, distinguished and compared.

Kate Johnson

Student, University of California Davis

Appearances