Welcome to Statistics and Operations Research

SAS has long developed software for data analysis, econometrics, operations research, and quality improvement. The purpose of these pages is to provide our users with technical information about using this software, including details about software capabilities, examples, papers, e-newsletter, and communities.

Featured News

Early-bird Registration for SAS Global Forum is Open

SAS Global Forum provides a host of analytical talks as well as dozens of short presentations on the exhibition floor and opportunities to talk shop with SAS developers. In addition, we offer a bunch of pre-conference tutorials that include presentations on mixed models, causal analysis, survey data analysis, and longitudinal data analysis. You can advance your ODS Graphics skills or learn all about modern machine learning techniques with SAS Visual Data Mining and Machine Learning software.

9.4M5 The DO Loop

Learn about the diffogram and other graphs for multiple comparisons of means as well as using the method of moments to choose initial parameter estimates for maximum likelihood estimation in these blog posts by Distinguished Research Statistician Developer Rick Wicklin. Also find out how to visualize multivariate regression models by slicing continuous variables and how to visualize patterns of missing values.

Blog Van Elteren Test for Nonparametric Two-Way Analysis

Lehmann (1975) and Dmitrienko et al. (2005) discuss and illustrate a nonparametric test proposed by van Elteren (1960) for stratified or blocked continuous response data. This test is an extension of Wilcoxon’s rank-sum test and is also a Mantel-Haenszel mean score test. As such, it can be obtained using either PROC FREQ or, beginning in SAS 9.4 TS1M3, PROC NPAR1WAY. It tests the null hypothesis of no treatment effect in the strata. Validity of the test depends only on large overall sample size and not on the strata sizes. Also, normality of the response distribution is not required, so this test can be used when a two-way analysis of variance might not be valid. The accompanying example from Dmitrienko et al. (2005) tests for drug effect in a data set from a clinical trial on urinary incontinence with patients from three strata. The response is the percentage change from baseline in the number of incontinence episodes per week. Because the distribution of this response is skewed and therefore not considered to be approximately normal, a nonparametric test is preferred.

May I Direct Your Attention to: The HPGENSELECT Procedure

The CUSTOM statement in PROC POWER in SAS/STAT performs power and sample size analyses for extensions of existing analyses that involve the chi-square, F, t, normal, or correlation coefficient distribution. Use cases include logistic regression with classification variables, Poisson regression, zero-inflated models, and other generalized linear models.

The HPGENSELECT procedure in SAS/STAT performs model building for generalized linear models. It provides the LASSO method. Learn more.