The HPBNET Procedure
Variable Selection
The network in Figure 1 does not include variables Sex or Duration, because PROC HPBNET automatically selects those variables by using independence tests. The following statements produce Figure 3, which shows the variable selection results. PROC HPBNET removes Duration from the network because the p-value of the chi-square and G-square statistics of Duration is greater than 0.05 (the default value for the ALPHA= option, which is not specified). Sex is conditionally independent of Pain given Treatment; therefore, PROC HPBNET also removes it from the network.
proc print data=varselect noobs label;
run;
Figure 3: Variable Selection
| Variable Name | Selected | Chi-Square Statistics | G-Square Statistics | P-Value of Chi-Square Statistics | P-Value of G-Square Statistics | Mutual Information | Degree of Freedom | Conditional Variables |
|---|---|---|---|---|---|---|---|---|
| Sex | NO | 7.2000 | 7.5945 | 0.06579 | 0.05518 | 0.34481 | 3 | Treatment |
| Treatment | YES | 13.7143 | 14.0230 | 0.00105 | 0.00090 | 0.45652 | 2 | |
| Age | YES | 14.6000 | 15.2712 | 0.00068 | 0.00048 | 0.47404 | 2 | |
| Duration | NO | 2.2579 | 3.3485 | 0.32337 | 0.18745 | 0.23298 | 2 |
Last updated: May 25, 2022