Appendix C: Known Issues in Using CAS Tables with SAS/STAT Procedures
Statistical Results Not Reliably Reproducible with Input CAS Tables
Table 13 summarizes known nonreproducibility issues of analysis methods in SAS/STAT procedures when you use a CAS table as input in the DATA= option. In these situations, the procedure issues a note about nonreproducibility or suppresses the nonreproducible statistical results.
Table 13: Statistical Methods That Do Not Reproduce Results Reliably with Input CAS Tables
| Procedures | Statistical Methods |
|---|---|
| ADAPTIVEREG, GLMSELECT, HPGENSELECT, HPLOGISTIC, HPPLS, HPQUANSELECT, HPREG, HPSPLIT, QUANTSELECT |
Partitioning of observations (for model training, validation, and testing) based on random assignments of roles via the FRACTION= option in the PARTITION statement |
| BGLIMM | Missing data that require imputations |
| BGLIMM, PHREG |
Bayesian methods with random effect specifications and/or imputations of missing responses |
| CAUSALMED, CAUSALTRT, QUANTREG, TTEST |
Bootstrap methods |
| CLUSTER | All cluster methods when there are tied distances |
| FMM, HPFMM |
Data augmentation methods |
| GLIMMIX | Estimation of the  matrix and its inverse, the Cholesky root of the matrix and its inverse, and the correlation matrix corresponding to the estimated matrix when the SUBJECT= option is not specified in the RANDOM statement |
| GLM | Autocorrelation statistics and Durbin-Watson D (results are suppressed) |
| GLMSELECT, HPREG |
Model averaging |
| MI | All multiple imputation methods |
| PHREG | Standard errors of Uno’s concordance statistics |
| PLS | Cross validation methods and tests |
| PSMATCH | Greedy matching when there are tied propensity scores |
| SURVEYSELECT | Probability sampling methods |
| SURVEYFREQ, SURVEYIMPUTE, SURVEYLOGISTIC, SURVEYMEANS, SURVEYPHREG, SURVEYREG |
Balanced repeated replication (BRR) and bootstrap methods |
| TREE | Tree diagrams |
The nonreproducibility issue is not limited to CAS table input in the DATA= option. An additional known issue that occurs is when you input a HADAMARD= data set as a CAS table for the SURVEYFREQ, SURVEYIMPUTE, SURVEYLOGISTIC, SURVEYMEANS, SURVEYPHREG, or SURVEYREG procedure, the numerical results of the balanced repeated replication (BRR) method are not reliably reproducible.