Page Banner

United States Department of Agriculture

Agricultural Research Service

Title: Using Sas for Pls Calibrations of Spectroscopic Data

Authors
item Reeves Iii, James
item Delwiche, Stephen

Submitted to: NIR news (Near Infrared Reflectance News)
Publication Type: Popular Publication
Publication Acceptance Date: May 17, 2004
Publication Date: June 1, 2004
Citation: Reeves III, J.B., Delwiche, S.R. 2004. Using the Statistical Analysis System for partial least squares calibrations of spectroscopic data. NIR news (Near Infrared Reflectance News). 15(3):10-13.

Interpretive Summary: In Volume 11, Issue 6 (2003) of the J. NIRS, an article entited 'SAS®partial least squares regression for analysis of spectroscopic data' by Reeves and Delwiche appears, along with two SAS® programs (I and II) and accompanying documentation. The present article for NIR News discusses the philosophy behind those programs and also introduces two updates and some additions written since the acceptance of the article which significantly add to the original programs. While many programs dedicated to chemometrics are available, these programs are generally not as user friendly as might be desired when it comes to processing a data set, or many data sets, in many different ways, e.g., different derivatives, scatter corrections, etc. and combinations thereof. The SAS programs discussed were originally designed to allow the rapid analysis of data sets using many different data pre-treatments using the results from a cross validation analysis to compare different data pre-treatments. These programs were not, and still are not, designed to replace existing dedicated chemometrics programs, but rather to act as a prototyping tool. In the latest versions, additional features have been added which allow for the use of independent test data sets, the determination of many more error terms for both the calibration and test set, for random generation of the test set (Program I only) and for plotting of spectra and spectral correlations (Program I only).

Technical Abstract: In Volume 11, Issue 6 (2003) of the J. NIRS, an article entited 'SAS®partial least squares regression for analysis of spectroscopic data' by Reeves and Delwiche appears, along with two SAS® programs (I and II) and accompanying documentation. The present article for NIR News discusses the philosophy behind those programs and also introduces two updates and some additions written since the acceptance of the article which significantly add to the original programs. While many programs dedicated to chemometrics are available, these programs are generally not as user friendly as might be desired when it comes to processing a data set, or many data sets, in many different ways, e.g., different derivatives, scatter corrections, etc. and combinations thereof. The SAS programs discussed were originally designed to allow the rapid analysis of data sets using many different data pre-treatments using the results from a cross validation analysis to compare different data pre-treatments. These programs were not, and still are not, designed to replace existing dedicated chemometrics programs, but rather to act as a prototyping tool. In the latest versions, additional features have been added which allow for the use of independent test data sets, the determination of many more error terms (Rsquares, RMSD, SEP, and BIAS) for both the calibration and test set, for random generation of the test set (Program I only) and for plotting of spectra and spectral correlations (Program I only). In addition, Program I can be ran within a loop, with random generation of the test set at each iteration. Preliminary efforts have shown that some data pre-treatments may be more robust as indicated by less variation in results across randomly generated test sets.

Last Modified: 10/20/2014
Footer Content Back to Top of Page