Heralded by a new release-numbering scheme, SAS/STAT 12.1, which was released in August 2012, comes loaded with new statistical capabilities. New development areas include model selection for quantile regression, quantile regression for censored data, and multivariate adaptive regression splines. Epidemiologists will like the STDRATE procedure for computing direct and indirect standardized rates and risks for study populations. The FMM procedure becomes production and includes new features such as support for additional distributions. Other notable enhancements include modeling missing covariates in the MCMC procedure and fitting Bayesian frailty models in the PHREG procedure.
Quantile regression is a modern analytic tool that models quantiles of a response variable conditional on explanatory covariates. It is especially useful for analyzing heterogeneous data in which the distribution of the response variable at the tails and central location vary conditional on the explanatory covariates.
The new QUANTSELECT procedure performs model selection for quantile regression. It supports model selection for a single quantile level and for a quantile process. The QUANTSELECT procedure offers capabilities similar to those offered by the GLMSELECT procedure. Selection methods include forward, backward, stepwise, and LASSO. PROC QUANTSELECT uses variable selection criteria such as AIC, SBC, and AICC. You can also partition your data into training, validation, and testing data sets, and then use the validation data set to measure the fitness of candidate models.
PROC QUANTSELECT is multithreaded so that it can take advantage of multiple processors. Numerous graphs illuminate the selection process.
Quantile regression also provides an alternative and flexible technique for the analysis of survival data. You can apply this technique to right-censored responses, thus providing covariate effects that are specific to quantiles and that directly predict lifetime. Two approaches are implemented: one is based on the idea of the Kaplan-Meier estimator, and the other is based on the Nelson-Allen estimator of the cumulative hazard function.
The new QUANTLIFE procedure provides interior point algorithms for estimation, Wald tests for the parameter estimates, survival plots, conditional quantile plots, and quantile process plots. It also supports the EFFECT statement so that it can fit regression quantile spline curves, and is multithreaded to take advantage of multiple processors when they are available.
The new STDRATE procedure computes direct and indirect standardized rates and risks for study populations. Direct standardization uses weights such as population-time from a standard or reference population to compute the weighted average of stratum-specific estimates in the study population. For two study populations with the same reference population, PROC STDRATE compares directly standardized rates or risks. For two study populations without a reference population, the procedure computes Mantel-Haenszel effect estimates, such as the rate difference.
The new ADAPTIVEREG procedure provides a nonparametric modeling approach for high-dimensional data. PROC ADAPTIVEREG fits multivariate adaptive regression splines as introduced by Friedman (1991). The approach is a nonparametric regression technique that combines both regression splines and model selection methods. It does not assume parametric model forms, and it does not require knot values for constructing regression spline terms. Instead, it constructs spline basis functions in an adaptive way by automatically selecting appropriate knot values for different variables, and it performs model reduction by applying model selection techniques. Thus, the ADAPTIVEREG procedure is both a nonparametric regression procedure and a predictive modeling procedure.
The ADAPTIVEREG procedure supports models with classification variables, and it provides options for improving modeling speed. PROC ADAPTIVEREG extends the method to data with response distributions from the exponential family, such as binomial and Poisson. PROC ADAPTIVEREG is multithreaded so that it can take advantage of multiple processors.
PROC ADAPTIVEREG enables you to force effects into the final model or restrict variables in linear forms; supports options for fast forward selection; supports partitioning of data into training, validation, and testing roles; and provides graphical representations of the selection process.
The FMM procedure for fitting finite mixture models, which was experimental in SAS/STAT 9.3, becomes production in SAS/STAT 12.1. PROC FMM fits statistical models to data for which the distribution of the response is a finite mixture of univariate distributions. These models are useful for applications such as estimating multimodal or heavy-tailed densities, fitting zero-inflated or hurdle models to count data with excess zeros, modeling overdispersed data, and fitting regression models with complex error distributions.
The MCMC procedure provides many new capabilities in SAS/STAT 12.1, including the following:
Other enhancements of SAS/STAT 12.1 include
SAS/STAT 12.1 is scheduled for release during the third quarter of 2012. See the details at support.sas.com/statistics/
See Look Out: After SAS/STAT 9.3 Comes SAS/STAT 12.1 for more details and examples of the new features contained in this release of SAS/STAT software. Another great resource is the What's New in SAS/STAT 12.1 chapter in the SAS/STAT documentation.
Download pdf version.