The second quarter of 2010 brings a new release of SAS/STAT software that includes significant new features and enhancements. Two new procedures, completely revised software for structural equation modeling, and substantial new coverage in postfitting inference are just some of the many enhancements. The following are highlights of this new release.

One of the strengths of SAS/STAT linear modeling procedures is the breadth of postfitting analyses available once you have fitted your model and estimated its parameters. Such statements as ESTIMATE, LSMEANS, and LSMESTIMATE provide the means of requesting this inference. In SAS/STAT 9.22, over 30 additional postfitting statements have been added to procedures such as GENMOD, LOGISTIC, MIXED, ORTHOREG, and PHREG. For example, the MIXED procedure picks up the LSMESTIMATE and SLICE statements, and the PHREG procedure picks up the ESTIMATE, LSMEANS, LSMESTIMATE, and SLICE statements.

In addition, the new PLM procedure performs postfitting inference with model fit information stored from these same procedures with the new STORE statement. PROC PLM inputs this information, saved as a SAS item store, and performs tasks such as testing hypotheses and scoring a new data set. These tasks are specified with the usual postfitting statements. Thus, you can perform additional analyses without refitting your model, and you can use PROC PLM to specify analyses that are not available in some procedures.

PROC PLM offers the most advanced postfitting inference techniques available in SAS/STAT software, including new
techniques such as step-down multiplicity adjustments for *p*-values, *F* tests with order restrictions,
analysis of means (ANOM), and sampling-based linear inference based on Bayes posterior estimates.

The experimental SURVEYPHREG procedure fits the Cox model for proportional hazards to sample survey data. The procedure provides design-based variance estimates, confidence intervals, and hypothesis tests concerning the model parameters and model effects. For statistical inference, PROC SURVEYPHREG incorporates complex survey sample designs, including designs with stratification, clustering, and unequal weighting. PROC SURVEYREG provides both Taylor series and replication variance estimation procedures, and it also provides domain analysis through the DOMAIN statement. In addition, PROC SURVEYREG offers postfitting inference as performed with the ESTIMATE, LSMEANS, LSMESTIMATE, and SLICE statements. You can save model information with the STORE statement for further use with the new PLM procedure.

The SURVEYFREQ procedure now provides plots that are created with ODS Graphics, including a weighted frequency plot, an odds ratio plot, a relative risk plot, and a risk difference plot. The CL option now offers additional confidence limit types, including the modified Clopper-Pearson (exact), modified Wilson (score), and logit. If you specify the DEFF option in the TABLES statement, PROC SURVEYFREQ computes design effects for the overall proportion estimates in the frequency and crosstabulation tables.

The SURVEYMEANS procedure now performs analysis for domain ratios. Variance estimation based on replication methods is available for domain means, totals, and ratios.

In the SURVEYSELECT procedure, the SAMPLINGUNIT statement names variables that identify the sampling units as groups of observations (clusters). The combinations of categories of SAMPLINGUNIT variables define the sampling units. If there is a STRATA statement, sampling units are nested within strata. The NMIN= option in the PROC SURVEYSELECT statement specifies the minimum stratum sample size for the SAMPRATE= option.

The CALIS procedure now includes updates that were previously made available in the experimental TCALIS procedure. These capabilities include the following:

- new modeling languages such as LISMOD, MSTRUCT, and PATH
- multiple group analysis
- improved mean structure analysis
- general parametric function testing
- improved effect analysis

In addition, PROC CALIS introduces several experimental features, including the full information likelihood method (FIML), mean structure analysis with the COSAN model, unnamed free parameter specification, and an extended path modeling language.

SAS/STAT software now provides seamless analysis and prediction of spatial processes. New features include the following:

- more correlation models
- automated semivariogram fitting
- item stores for passing semivariogram information to the KRIGE2D or SIM2D procedures
- customizable graphics for spatial data exploration and display of resulting predictions and simulations

Bayesian capabilities continue to grow in SAS/STAT software. The capabilities provided by the BAYES statement
in the GENMOD, LIFEREG, and PHREG procedures have been updated with new sampling methods. Conjugate sampling
for linear regression is now the default in the GENMOD procedure, reducing computation time. You can specify
either the Gamerman algorithm or the independent Metropolis algorithm in PROC GENMOD for other generalized
linear models. You can choose the random walk Metropolis algorithm as an alternative sampling method in the
PHREG procedure, and you can specify the Zellner *g*-prior for the regression coefficients.

The MCMC procedure introduces the PREDDIST statement, which enables you to create random samples from the posterior predictive distribution of the response variables. The posterior predictive distribution is the distribution of unobserved observations (predictions) conditional on the observed data.

If you specify multiple quantiles in a MODEL statement of the QUANTREG procedure, additional analyses (such as those specified in the TEST statement) are now produced for each quantile specified. The RANKSCORE option in the TEST statement enables you to perform rank tests. Available score functions provide normal scores, Wilcoxon scores, and sign scores, which are asymptotically optimal for the Gaussian, logistic, and Laplace location shift models, respectively.

The GLMSELECT procedure now provides model averaging with the experimental MODELAVERAGE statement, which requests model selection on resampled subsets of the input data. An average model is produced by averaging the parameter estimates of the selected models that are obtained for each resampled subset of the input data.

The ADAPTIVE option of the SELECTION=LASSO method specifies adaptive lasso selection, which is a modification of lasso selection where weights are applied to each of the parameters in forming the lasso constraint.

The experimental EFFECT statement, which defines a richer class of linear models, is now available in the HPMIXED, GLIMMIX, GLMSELECT, LOGISTIC, ORTHOREG, PHREG, PLS, QUANTREG, ROBUSTREG, SURVEYLOGISTIC, and SURVEYREG procedures. With this statement, you can define effect types such as splines, multiclass effects, lag effects, and polynomial effects. The EFFECTPLOT statement uses ODS Graphics to create plots of model effects in the GENMOD, LOGISTIC, and ORTHOREG procedures.

The LIFEREG procedure now reports fit criteria based on the distribution of the response on the original scale (rather than on the log of the response) if you specify the Weibull, exponential, lognormal, log-logistic, or gamma distribution.

The LIFETEST procedure now enables you to request the Breslow and Fleming-Harrington estimates of the survivor function with the METHOD= option in the PROC LIFETEST statement. The number of subjects at risk can be displayed with the product-limit estimates, the Breslow estimates, and the Fleming-Harrington estimates.

The PHREG procedure now offers the ATRISK option in the PROC PHREG statement, which displays a table that contains the number of units at risk at each event time and the corresponding number of events in the risk sets. Likelihood ratio tests of model parameters are also available with the TYPE1 and TYPE3 options in the MODEL statement except when the robust sandwich estimate for the covariance matrix is specified.

The POWER procedure now enables you to parameterize computations for survival analysis in terms of the expected number of events, in addition to sample size (see the EVENTSPERGROUP=, EVENTSTOTAL=, and GROUPEVENTS= options in the TWOSAMPLESURVIVAL statement). Parameterization in terms of sample size accrued per unit time is also available in this statement with the ACCRUALRATEPERGROUP=, ACCRUALRATETOTAL=, and GROUPACCRUALRATES= options.

ODS Statistical Graphics technology is used by even more procedures with this latest release. The majority of SAS/STAT procedures now produce graphics as systematically as they produce tables, with over 400 easily specified graphs. Graphics have arrived in the SURVEYFREQ and TPSPLINE procedures, and new graphs are now available in the FREQ, SIM2D, KRIGE2D, and VARIOGRAM procedures.

The GENMOD procedure now fits exact Poisson regression, exact logistic regression, and the zero-inflated negative binomial models.

The FACTOR procedure output has been enhanced to include a table with the number of observations used in the analysis.

The TPSPLINE procedure now provides plots created with ODS Graphics, including residual plots, diagnostic plots, and fit plots. You can also request confidence bands for the expected value of the dependent variables by using the UCLM and LCLM keywords in the SCORE statement.

The VARCOMP procedure now enables you to request generalized confidence limits for the parameters with the CL=GCL option in the MODEL statement.

The HPMIXED procedure and the LOESS smoother in the MODEL statement of the GAM procedure are now production.

The FREQ procedure now offers exact

*p*-values for tests of the following measures: Kendall’stau-b , Stuart’stau-c , Somers’ D(C|R), and Somers’ D(R|C). The GAILSIMON option in the TABLE statement specifies the Gail-Simon test for qualitative interactions, and the MANTELFLEISS suboption of the CMH option requests the Mantel-Fleiss criterion for the Mantel-Haenszel statistic for stratified 2 x 2 tables. Relative risk plots and risk difference plots are also now available.The GLIMMIX procedure now enables you to perform a joint test under one-sided restrictions with the LSMESTIMATE statement (Silvapulle and Sen 2004); for example, you can test ordered alternatives. PROC GLIMMIX computes a simulation-based chi-bar-square statistic and produces a

*p*-value for the constrained joint test.The VARIOGRAM procedure now enables you to save the context and results of the semivariogram model fitting analysis in an item store by using a STORE statement. The contents of item stores can be processed with the KRIGE2D or the SIM2D procedures. After you save results in an item store, you can use them at a later time without having to refit the model.

See
The Next Generation: SAS/STAT^{®} 9.22
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 9.22 chapter in the SAS/STAT documentation.

SAS/STAT 9.22 is currently available. To obtain more information, ask your organization's SAS representative to contact the SAS Customer Interaction Center at 1.800.727.0025.

Download pdf version.