A SAS procedure for exact probability testing of difference between sample and population proportion.

Abstract

Statistical testing of the hypothesis that the proportion of subjects in a defined population having a certain attribute (proportion of "positives" in a population, P) is equal to some specified value (P0) is frequently encountered in biomedical research. For example, a study might be carried out to statistically test whether the postoperative wound infection rate in patients having undergone an operation with a "new" surgical procedure is 20%, the same value that has been observed for the "established" surgical procedure. The significance test for this hypothesis (e.g., test H0: P = 0.20 against HA: P not equal to 0.20) is usually based on the normal theory approximation method. However, when the sample size is "small", especially if P0 is close to 0 or 1, the normal theory method can yield grossly unreliable results. In contrast, the significance test based on the exact binomial probability procedure always yields reliable results. A computer program coded in SAS is described herein to perform this exact probability test procedure.