Changeset 20026 in main

Timestamp:

03/13/20 09:50:45 (3 weeks ago)

Author:

Paul Leo

Message:

Montana would like to use the following for computing confidence intervals...
use binomial exact confidence limits for all of them with counts over 20 (not suppressed). The formulas for these are below, where x is the numerator (such as births with prenatal care in the first trimester), and n is the denominator (such as all births).

Lower = (1 – betainv (.975, n – x – 1, x))(100%)
Upper = (1 – betainv (.025, n – x, x + 1))(100%)

File:

• adopters/mt/trunk/src/main/backend_qModules/birth23/MT_Average_data_frame.def (modified) (2 diffs)

adopters/mt/trunk/src/main/backend_qModules/birth23/MT_Average_data_frame.def

@@ -264 +264 @@
                         end;
 
-                        if 20<=count<=100 then do; 
-                        * for events between 20 and 100, use T distribution to calculate confidence intervals;
-                        * note LL and UL are calculated in step 4 above.
-                        
+                        * for events >= 20 use binomial exact confidence intervals;
+                        
+                        if count >= 20, then do;
+                                LL=put(t1, 8.2);
+                                LL=(1 – betainv (.975, popcount – count – 1, count))*100);
+                                UL=((1 – betainv (.025, popcount – count, count + 1))*100);
+                                LL=compress(LL);
+                                UL=compress(UL);
                                 redflag=put('-', $15.); 
-
-                        end;
-
-                        *  and normal approximation for more than 100 events. ;
-                        * which is also computed in step 4 above
-                        
-                        if count>100 then do;
-
-                                redflag=put('-', $15.); 
-                        end;
+                        end;
+                        
                 end;
                 
@@ -296 +292 @@
                         * report rate and n if popcount <300 and count >= 20 events.    ;
 
-                        if 20<=count<=100 then do; 
-                        * for events between 20 and 100, use T distribution to calculate confidence intervals;
-                        * note LL and UL are calculated in step 4 above.
-                        
-                                redflag=put('-', $15.); 
-
-                        end;
-
-                        *  and normal approximation for more than 100 events. ;
-                        * which is also computed in step 4 above
-                        
-                        if count>100 then do;
-
+                        * for events >= 20 use binomial exact confidence intervals;
+                        
+                        if count >= 20, then do;
+                                LL=put(t1, 8.2);
+                                LL=(1 – betainv (.975, popcount – count – 1, count))*100);
+                                UL=((1 – betainv (.025, popcount – count, count + 1))*100);
+                                LL=compress(LL);
+                                UL=compress(UL);
                                 redflag=put('-', $15.); 
                         end;