Comparing several proportions when the exposure variable is either ordinal or ratio which has been grouped into...

The 2-by-C chi 2 test is used for comparing the proportion of "positives" of a binomial response variable among the C levels of an exposure variable. There are actually several versions of the chi 2 test that are fundamentally different, and each is meaningful only for a particular measurement scale of the exposure variable. The ordinary chi-square, which is often used indiscriminately for comparing proportions, is in fact only valid for nominal exposure variable, the chi 2 test for trend is meaningful only for ordinal exposure variable, and the regression-like chi 2 is suitable only for ratio exposure variable which has been grouped into a few discrete classes. Two SAS programs (PC SAS Release 6.04) to carry out the 2-by-C chi 2 test, one for ordinal exposure and the other for discretized-ratio exposure, were described. In addition to the chi 2 test, the programs also compute the exact probability confidence interval of the proportion for each exposure level. Numeric examples were used to illustrate the application of the programs.